What if the most significant revolution of our time isn't just about faster computers or smarter algorithms, but about a fundamental re-wiring of how we experience reality itself? In an age often polarized between unbridled techno-optimism and dire predictions of collapse, I believe a more nuanced, pragmatic approach is essential. We must look beyond the hype and the fear to understand the true nature of the technological shifts unfolding around us.

In this essay, I will argue that we are not merely witnessing the rise of powerful new tools; we are entering the Era of Artificial Intelligence, a profound societal transformation where AI increasingly mediates our relationships with objective reality and, crucially, with each other.

To understand the magnitude of this shift, consider how past technological eras have reshaped human existence. For the purpose of this article, I'll define an "era" as a period marked by a fundamental shift where the majority of relationships, both with objective reality and among people themselves, become mediated by a specific technology stack.

Let's look at a few examples.

First, consider what we may call the Era of Electricity. Slowly first, but increasingly accelerated, our interactions with the physical world were completely transformed. Light, heat, power, all mediated by electric currents. Work in factories and farms became mediated by electricity, then transportation, then entertainment and education. Electricity even changed how we sleep. As people started staying outside after dark, we changed from a sleep cycle, often biphasic, that mirrored the behavior of the Sun, to our monophasic cycle that is completely artificial—we go to sleep way before its dark, and we wake up often before the Sun comes out.

Another example is the Era of Computers. Our engagement with reality became even more profoundly mediated, not just by physical devices, but by the logic and processing power of silicon. From personal computing to industrial automation, the computer became the invisible hand guiding our interaction with information and machinery. Think how many jobs started to function exclusively with a computer mediating between our intentions and their results. We started to use computers to pay for things, where before a simple exchange of paper was enough. Even our language changed with the advent of computers.

The Internet Era took this mediation a step further, fundamentally altering human relationships. We started flirting over the internet, ordering food, arguing about politics, and making friends over the Internet. Our social fabric itself became woven through digital threads. All our life is intricately connected to the Internet, there is almost no interaction between ourselves and with the real world that doesn't happen through some sort of online application or platform. Even this interaction between me, the author, and you, the reader, is only possible because the Internet mediates. We wouldn't know each other otherwise.

Now, Artificial Intelligence stands poised to inherit this mantle. In the remaining of this article I want to argue why AI is already becoming the primary mediator of most meaningful human interactions, and will continue to do so in the near future. And while no one can deny the progress of technology brings incredible benefits---and I'm the first to acknowledge it and embrace the immense power that AI grants---it is also undeniable that this mediation brings some pretty scary consequences, some which are already unfolding.

Subscribe now

AI as the Ultimate Mediator

AI's mediating power extends across every facet of our lives, subtly, yet profoundly, shaping how we perceive, interact with, and relate to the world and each other.

Consider how AI is already changing our perception and understanding of the world. In medicine, AI analyzes X-rays and MRIs, often detecting anomalies like tumors with greater speed and accuracy than the human eye. It mediates how doctors and patients perceive health conditions.

In industry, AI monitors machinery to predict failures, mediating our understanding of the physical world's future state. AI even helps us perceive the planet's health by processing satellite imagery to track deforestation or pollution.

Think about your phone's camera: when you take a photo, AI often automatically adjusts the lighting, focus, and even smooths skin, mediating how you perceive the scene or person you're photographing. It's not just capturing reality; it's enhancing or altering it before you even see the final image.

Beyond perception, AI is increasingly mediating our interaction and control over reality. Autonomous vehicles, from cars to drones, are changing how we physically move through and interact with our environment.

Smart home systems, managed by AI, mediate our control over our immediate living spaces. In manufacturing, AI-driven robots optimize production lines, mediating how we create and manipulate physical goods. When you ask a voice assistant like Alexa or Google Home to play music or turn off lights, AI is mediating your control over your environment through voice commands, rather than you physically interacting with switches or devices.

Many modern cars also have AI-powered features like adaptive cruise control or lane-keeping assist, mediating your interaction with the road by taking over some aspects of driving.

And then there's the realm of creation and augmentation. Generative AI, like Midjourney or ChatGPT, creates art, music, and text from simple prompts, mediating our creative output and expanding what's possible.

AI accelerates scientific discovery by analyzing vast datasets, mediating our ability to discover and augment knowledge. When you use an AI assistant or just an innocent grammar checker like Grammarly, AI is mediating your writing process and creative output, suggesting changes to make your sentences clearer or more impactful. Many video editing apps now have AI features that can automatically cut scenes or add background music, mediating your creative process in video production.

Our daily consumption of information is also heavily mediated by AI. Algorithms curate our news feeds, suggesting articles and entertainment based on past behavior, mediating what information and experiences we consume from the vast digital reality. Think about your Netflix recommendations; AI is constantly analyzing your viewing habits to mediate what shows and movies you see, often leading you down rabbit holes of similar content.

But the impact of AI on human relationships is even more profound. Communication itself is being redefined. Real-time language translation, powered by AI, enables seamless conversations across linguistic divides, mediating interpersonal understanding. AI-powered chatbots handle customer service, mediating interactions between businesses and consumers. And in a more intimate, and perhaps concerning, development, AI companions are emerging, offering conversational interaction and emotional support, mediating personal and emotional relationships in ways we're only beginning to comprehend.

Have you ever used a dating app? Or just Facebook or LinkedIn? AI algorithms are mediating who you see as potential partners by suggesting matches based on your profile and preferences, fundamentally changing how many people initiate relationships, romantic or otherwise.

When you send an email and your email client suggests auto-completing your sentences or offers quick replies, AI is mediating your written communication, subtly influencing your phrasing and the very nature of your interaction with the person at the other side of that email.

Consider how ride-sharing apps like Uber or Lyft connect drivers and passengers; AI mediates who you interact with for transportation, replacing traditional hailing or dispatch systems. Many online gaming platforms also use AI to match players for competitive games, mediating who you collaborate or compete with.

Finally, AI is beginning to mediate our very identity and self-perception. Deepfakes and synthetic media, generated by AI, create realistic images and videos of people, blurring the lines of authenticity and mediating how we perceive others and ourselves online. Personalized digital avatars, often AI-assisted, mediate our online identity.

Think about the filters you use on social media that can alter your appearance in photos or videos; AI is mediating how you present yourself to others online and, in turn, how you might perceive your own image. AI is starting to mediate even your relationship with your own mind.

The Perils of AI Mediation

The immediate benefits of Artificial Intelligence are undeniable. AI offers unprecedented efficiency, personalized experiences, and the ability to solve complex problems that have long eluded us. This potential for human flourishing is immense, provided we make conscious choices about its use.

However, this era, like all technological advancements, comes with no free lunch. While the benefits are compelling, they are inextricably linked to profound, often insidious, risks. To focus solely on these gains would be to miss the critical trade-offs inherent in this new era. As AI imposes more layers of mediation between us and our world, we risk a dangerous detachment from immediate reality, from each other, and ultimately, from our own agency.

This is where the prescient critique of Guy Debord, articulated in his seminal 1967 work The Society of the Spectacle, becomes chillingly relevant. Debord argued that modern capitalist society had transformed "all that was once directly lived" into "mere representation." The spectacle, for Debord, was not just a collection of images, but a social relationship mediated by images, where appearances replaced authenticity, and passive consumption supplanted active participation.

He warned of a world where reality itself was subsumed by its image, leading to profound alienation and a loss of genuine human experience. If Debord's mid-20th century world was a spectacle, the Era of AI threatens to usher in a Hyper-Spectacle—a phenomenon of mediation so pervasive and sophisticated that it would dwarf anything he could have imagined. I think there are three fundamental, distinct concerns about this hyper-spectacular reality that we should analyze in depth.

Disconnection from Immediate Reality

In the Hyper-Spectacle of AI, our direct engagement with reality is increasingly replaced by an algorithmic illusion. When AI curates all our information, news, and even sensory input, we risk living in an algorithmic filter bubble on steroids. Imagine an AI-powered AR overlay that filters out undesirable elements of reality, presenting a highly curated, potentially distorted, and ultimately unreal version of the world.

This is the ultimate triumph of Debord's "mere representation", where the mediated reality becomes more compelling, more perfectly tailored, than the actual reality. We become passive consumers of AI-generated solutions, outsourcing our direct engagement with the physical world and its challenges. This leads to a loss of practical skills and a diminished sense of agency, as our capacity for direct experience atrophies.

Think about how many people rely solely on their phone's GPS for directions, even in familiar areas. AI mediates their perception of their physical surroundings, often leading them to ignore street signs or landmarks, and potentially diminishing their spatial awareness. If the GPS is wrong, they're lost because they haven't engaged directly with the real world.

Similarly, when you watch a heavily AI-edited video or listen to an AI-generated song, you might be consuming something that feels real but was never "created" by a human in the traditional sense, mediating your experience of art and authenticity.

Then comes the black box issue—where AI systems make critical decisions based on opaque algorithms—which means we benefit from outcomes without understanding why or how. The "truth" presented by the AI is accepted without question, becoming an unchallengeable spectacle. This opaque authority reinforces the illusion, as the underlying mechanisms of control are hidden, and the mediated outcome is presented as objective fact, further separating us from the direct understanding of cause and effect.

For example, many people now rely on AI-powered smart assistants to answer questions or provide information without ever checking the source. If the AI gives a subtly incorrect or biased answer, it's often accepted as fact, mediating their understanding of truth without critical engagement.

Or consider a smart refrigerator that automatically reorders groceries based on AI predictions of your consumption; while convenient, it mediates your direct engagement with your food choices and shopping habits, potentially leading to a loss of awareness about what you're actually consuming or spending.

Disconnection from Each Other

The Hyper-Spectacle extends its reach into our most fundamental human connections, fostering a simulated relationship that replaces authentic interaction. The rise of sophisticated AI companions, so convincing they fulfill emotional or social needs, poses a chilling question: what happens to genuine human interaction?

The "relationship" with the AI is a performance, a simulation of connection, a perfect example of Debord's social relationship mediated by images where the image of connection supplants its substance. This risks eroding empathy, social skills, and the capacity for deep, complex human relationships, leading to profound loneliness in a hyper-connected, yet isolated, world.

Think about how many people prefer to text or message rather than call or meet in person. Even without AI, the digital mediation of communication can lead to a simulated connection where nuances of tone, body language, and spontaneous interaction are lost, potentially eroding deeper social skills. Now add an ever-present, never-angry chatbot that can replace all your friends and family.

Moreover, AI-driven algorithms optimize content delivery to maximize engagement, often by showing us more of what we already agree with. This fuels algorithmic polarization and echo chambers, making shared understanding and collective action incredibly difficult. Different groups live in entirely different "spectacles" of reality, curated by algorithms, making genuine dialogue and the bridging of divides increasingly impossible.

The spectacle here is not just what we see, but who we see and how we see them, fragmenting society into isolated, algorithmically-defined bubbles. If your social media feed is constantly showing you content from only one political viewpoint, AI is mediating who you see and what opinions you encounter, making it harder to understand or empathize with opposing views. You're living in an algorithmically-defined bubble.

Servitude to Embedded Ideologies

Perhaps the most insidious risk in the Hyper-Spectacle is the subtle, yet pervasive, servitude to ideologies embedded within these mediating layers. AI systems new and old, when trained on biased historical data, can perpetuate and amplify existing societal inequalities in areas like hiring, lending, or criminal justice. This ideology of past discrimination gets baked into the future, presented as objective truth by AI.

This is the spectacle's ultimate power: presenting its biased outcomes as neutral, objective reality, subtly reinforcing existing power structures and values without overt coercion.

Imagine an AI-powered resume screening tool that, because it was trained on historical hiring data, subtly favors candidates from certain demographics or educational backgrounds, even if those biases aren't explicitly programmed. This mediates who gets professional opportunities, reinforcing past inequalities.

To make things even worse, AI systems are increasingly designed to subtly influence our choices—what to buy, what news to read, whom to vote for—through personalized recommendations and persuasive interfaces. Our choices feel free, but they are meticulously guided by an unseen hand, a spectacle of autonomy that masks underlying control.

This is the essence of Debord's critique of commodity fetishism extended to every aspect of life: our desires and decisions are manufactured and mediated, not genuinely our own. When you're shopping online, AI-powered "recommended for you" sections or "customers also bought" suggestions are constantly mediating what products you see and are encouraged to buy, often leading you to spend more or buy things you didn't initially intend to. Your choice feels free, but it's heavily influenced.

When the development and deployment of powerful AI systems are concentrated in the hands of a few corporations or governments, it gives them unprecedented control over information, resources, and even human behavior. The AI-mediated world becomes a spectacle designed and controlled by an elite, shaping narratives and realities to their advantage, further cementing our role as passive spectators in a world we no longer truly govern.

The Future is Not Predetermined

This Era of AI is not a distant future; it is here, and it is accelerating at a pace that leaves little room for avoidance. We cannot, and perhaps should not, try to stop it. But we absolutely must shape its trajectory. The technology is inevitable, but its direction and impact are not predetermined. This requires a proactive, collective effort, guided by reason and evidence, not by fear or blind faith.

I have absolutely no idea how to "fix" the problem of AI mediation. But I'm sure there are at least three critical approaches we have to embrace, lest we become mere spectators of this hyper-spectacular reality.

First, we must demand regulation that keeps choice firmly in the hands of the people. This requires advocating for the principles of algorithmic choice. We should always be able to determine when an algorithm is being used to guide our choices, and crucially, we must have the agency to understand, challenge, and more importantly, select which algorithms we let influence ourselves. Just as we choose our food or news sources—because you do, right?—, we should demand the right to choose our algorithmic lenses.

This means not only having, e.g., options to turn off algorithmic recommendation in Netflix or Twitter, but also to connect our own algorithmic filters and recommendation engines to replace the built in functionality. To achieve this, platforms need to embrace open protocols for algorithmic recommendation, for example, such that anyone can implement a different feed sorting procedure for Twitter. Platforms like Mastodon and Bluesky are pioneering this approach, but without explicit regulation there is no hope the largest platforms will follow.

It is clear then that we need robust institutional oversight. This means proactive, informed government regulation to set standards and enforce accountability. But it also means empowering non-governmental watchdogs, independent organizations, academia, and civil society, to monitor, critique, and advocate. Their role is crucial in preventing overreach, ensuring fairness, and keeping powerful AI entities in check. This collective vigilance is essential to prevent the concentration of power that fuels the spectacle.

And finally, beyond regulation and oversight, a more fundamental and personal shift is needed. We must consciously and actively strive to reclaim our direct experience of reality and shed the algorithmic lenses that increasingly mediate our lives. This means cultivating critical awareness of when and how AI is shaping our perceptions, fostering genuine human connections that bypass algorithmic curation, and seeking out unmediated experiences.

It means valuing authentic engagement over passive consumption, and actively participating in the world rather than merely observing its mediated representation. This is a call to intellectual and experiential rebellion against the Hyper-Spectacle, a commitment to living directly rather than through the image.

The Era of AI is upon us. It promises to mediate nearly every aspect of our lives. The question is not if it will mediate, but how. Answering this in a way that encourages human flourishing demands an unending, tireless reevaluation of our position towards technology. There are no banners, no ideology we can trust blindly. Techno-optimists and AI-doomers are both wrong, because they believe the outcome of technology is predetermined. And that makes us, by definition, spectators.

So here is my call to action for you today. Dare to stop living inside this spectacle of reality for a minute, put down your phone, close your laptop, or turn off your display. Look up, the world is still out there, waiting for you. Call one of your friends and ask them out for a pizza or a coffee or a beer—or all of them for a triple bonus. Hugh your kids if you have any, and your parents if you're lucky to still have them around. Go outside, feel the warmth of the sun in your face, the smell of the city, or the countryside, or the ocean, the roughness of dirt or pavement or grass under your bare feet. If it hurts just a little bit, even better.

Dare to choose how you want experience the actual, raw, unmediated world. The choice, I believe, is still yours. But not for long.

Unless you've been living under a digital rock for the last three years, you've almost certainly heard the whispers, or perhaps the shouts, that Artificial General Intelligence (AGI) is almost here. This idealized form of AI, often pitched as the ultimate problem-solver, promises to revolutionize everything —from curing diseases to managing global economies, while also improving your coffee and doing your dishes—making our current AI advancements look like child's play. While predictions on its exact arrival diverge wildly, there's a palpable feeling that we might be on the brink of a major milestone for humanity: the emergence of truly intelligent AIs that can perform anywhere humans can, equally or even better than human experts.

And while those grand visions have their place in sparking imagination, I want to propose a more grounded, perhaps more pragmatic, path forward. My thesis here is simple: while the pursuit of AGI remains a fascinating and worthy long-term aspiration, a clearer, more achievable, and ultimately more impactful goal for our immediate future lies in what I call "General-Purpose AI" (GPAI). Think of GPAI not as the ultimate destination, but as a vital compass, guiding us toward meaningful AI development and real-world impact right now.

In this article, I'll delve into why the singular focus on AGI can be a distraction, define what I mean by General-Purpose AI, break down its essential components, and outline a realistic roadmap for how we can get there. My aim is to shift our collective gaze towards a more actionable and universally beneficial future for AI.

Challenges in Defining and Achieving AGI

Let's be honest, the concept of AGI, as it's often discussed, is at least somewhat problematic.

One of the biggest hurdles for achieving AGI is its sheer vagueness. What does it really mean to achieve "human-level intelligence across all relevant domains"? The definitions are often slippery, lacking the clarity needed for a concrete engineering goal. It frequently leans into those fantastical, almost mystical, connotations of sentience, consciousness, and an all-encompassing intelligence. Because of this, proving definitively that AGI has been "achieved" becomes incredibly difficult, almost unquantifiable.

Furthermore, a significant challenge lies in the lack of clear measurability. How would we even know when we've truly arrived at AGI? What are the metrics, the benchmarks that would declare its presence? This contrasts sharply with traditional software development, where goals are typically practical and eminently measurable. Without clear, agreed-upon ways to measure progress or success, AGI risks remaining an elusive phantom.

Most critically, the singular focus on AGI can inadvertently pull our resources and attention away from the tangible. It can divert brilliant minds and significant investment from impactful AI applications that are not only achievable but are already solving real problems today. This pursuit of a distant ideal risks creating unrealistic expectations and, potentially, widespread disillusionment if those grand promises don't materialize in the near term.

For me, the emphasis should be on useful AI, the kind that drives genuine innovation and addresses pressing challenges in our world right now. This is where GPAI offers a more grounded and actionable target. It's about building tools that work, that empower, and that deliver measurable value, without getting lost in the theoretical infinite.

What is General-Purpose AI (GPAI)?

So, if AGI is the abstract ideal, what then is GPAI?

The core idea behind GPAI is a fundamental shift in how we perceive AI. Instead of seeing AI primarily as a finished, consumer-facing application, I view it as a foundational set of tools, capabilities, and infrastructure. Think of it like electricity or the internet, not an end in itself, but a pervasive utility that empowers countless other creations and innovations. This infrastructural view brings immense benefits like scalability, ubiquity, and incredibly broad applicability across diverse domains.

Crucially, GPAI isn't primarily designed for the end-user. Its true audience is the developer. GPAI is about empowering developers to effortlessly integrate sophisticated AI capabilities into their own applications. This democratizes AI, enabling a new generation of problem-solvers to leverage powerful AI tools without necessarily needing to be deep AI experts themselves. It's about bringing AI out of the specialist's lab and into the everyday toolkit of every software engineer.

Let me try then to define GPAI more concretely:

GPAI is a state of the AI infrastructure, tools, and techniques, that enables any developer to incorporate narrow AI for any specific problem on any specific domain, within any reasonable set of infrastructure constraints.

To put it more succinctly, GPAI aims for "anything, anywhere, anytime," in contrast to AGI's often implied "everything, everywhere, all at once". It's about practical, pervasive utility rather than an all-encompassing, perhaps unattainable, general intelligence.

Unpacking GPAI

So, what are the core ingredients that I think will make GPAI a reality?

1- Foundation Models

At the heart of GPAI are what we call Foundation Models. These are vast collections of pre-trained models, already capable of understanding and generating content across various modalities. They are essential because they provide powerful, ready-to-use intelligence, significantly reducing the need for developers to train models from scratch for every new application.

We're already seeing this with text, audio, and image models. But for true GPAI, we need a crucial extension into more challenging modalities: think 3D scenes with realistic physical interactions, complex tabular and relational data, time series, and general graph-like structures. Expanding foundational models into these areas will enable developers to tackle an even more diverse set of problems without requiring deep, specialized AI expertise for each domain.

2- Comprehensive Interoperability

For AI to become truly infrastructural, it needs to speak the same language as everything else in our digital world. This is where standardized communication methods, like MCP and alternatives, become critical. These protocols will allow AI systems to connect seamlessly with traditional software components.

This is about breaking down the silos that currently exist between AI and our existing IT infrastructure. It's about enabling AI features to be integrated just as easily into legacy systems as they are into cutting-edge modern applications. Imagine AI that can effortlessly plug into databases, operating systems, file systems, and web browsers. It also means ensuring different types of AI models—generative, discriminative, pre-trained, custom-trained—can work together harmoniously.

3- Simplified AutoML

Building custom AI solutions often requires specialized knowledge in machine learning. GPAI will change that by making the training and fine-tuning of narrow, task-specific ML models astonishingly simple, directly integrated into standard development environments.

The goal here is to drastically reduce the complexity and specialized knowledge typically required for model development. We want to make custom AI solutions accessible to a vastly wider base of developers. Picture configuring a machine learning model with the same ease as setting up a database or a Docker runtime. This will involve heavy integration of AutoML techniques —automated model selection, hyperparameter tuning, and the like— directly into our IDEs. The aspiration is to shift ML from a niche, specialist domain to a standard, everyday development task for any programmer.

4- Accessible GOFAI

While modern machine learning is transformative, we shouldn't forget the rich history of AI. GPAI will also include tools that expose classic symbolic AI methods, often referred to as Good Old-Fashioned AI (GOFAI), such as constraint satisfaction and search algorithms, all with modern ease of use.

Many real-world problems—think scheduling, logistics, or complex planning—are remarkably well-suited for these time-tested techniques. Providing these simply and effectively adds a powerful, complementary set of tools to the developer's arsenal. It’s about integrating these methods with the same configuration simplicity as other components, ensuring developers have a comprehensive toolkit that combines the best of both modern and classical AI approaches.

5- Pervasive Hardware/OS Support

Finally, for AI to truly be "anywhere, anytime," GPAI systems must be incredibly robust operationally. This means they need to run effectively under a vast array of circumstances, even offline or on mobile devices.

This implies robust hardware support for in-service training and inference of traditional models directly on devices, and at least the ability to perform inference of sufficiently good generative models on those devices. This on-device capability is absolutely essential for AI to become a truly ubiquitous utility, woven into the fabric of our digital lives.

How We Get There

Achieving GPAI won't be a single, monolithic event. It will be a journey with several overlapping phases.

It's important to understand that these phases won't unfold one after another in a strict sequence. Instead, they will run concurrently, with progress in each area accelerating at varying speeds. We should also acknowledge upfront that some challenges, particularly in extending foundation models to very specific and complex modalities, will naturally take longer to solve than others.

Phase 1: Pervasive Protocol Integration and Tooling Development: This initial phase focuses on laying the foundational plumbing. We'll see the widespread development of protocols like MCP designed to seamlessly integrate AI systems with all the traditional boundaries of software—databases, operating systems, file systems, and web browsers. This is about creating the conduits through which AI can flow freely. This is ongoing as we speak.

Phase 2: Comprehensive Framework and Tooling Accessibility: Concurrent with the plumbing, this phase is about creating the robust AI frameworks and tools themselves. These will be accessible across most programming languages, encompassing both modern machine learning techniques and the classical GOFAI methods like constraint satisfaction and search. The emphasis here is on simplifying complex AI tasks, making them approachable for the broader developer community. This is also ongoing in some domains, but we still lack developer-friendly tooling for the most common GOFAI techniques and traditional ML.

Phase 3: Extension of Foundation Models to Challenging Modalities: As the frameworks mature, we'll see a significant push to advance foundation models beyond common text and image data. This phase will focus on effectively processing and reasoning with more complex data types such as 3D spaces, tabular data, time series, and graphs. This tackles the "less-solved" domains, opening up new frontiers for AI application. We’re seeing right now an increased interested in World Models that can reason in terms of the physical properties of objects and better predict their interactions. We also need more work in models that can understand heavily structured data.

Phase 4: Mainstreaming of AutoML Techniques: This phase will mark the maturation of AutoML. These techniques, currently more prevalent in academic or specialized domains, will become sufficiently developed and integrated to move into widespread, mainstream development practices. This means that optimizing and selecting machine learning models will become a standard, almost automated, part of the development workflow. We already have pretty strong AutoML frameworks, but rarely any developer knows them, and they are lagging behind in integration (think AutoML running in your CI/CD) and deployment.

Phase 5: Ubiquitous On-Device AI Execution: Finally, a truly pervasive GPAI requires that AI can run almost anywhere. This phase is about ensuring that virtually any device can execute a "good enough" foundation model. Crucially, this will involve establishing robust operating system-level support for on-device AI inference, making local AI development and deployment a standard, seamless experience. This is already happening, but I think we need some breakthrough in hardware capabilities to make running a somewhat capable model, similar to what a 24B MoE can deliver today, natively in Android or iOS.

Given these phases and the inherent complexities, I believe we're looking at a realistic timeframe of 5 to 10 years for GPAI to reach a point where any developer on Earth can easily integrate AI into any reasonably complex application, much like they currently add a database or a graphical user interface today.

It believe this isn't science fiction. It’s ambitious, yes, but also a pragmatic and achievable goal.

Final Thoughts

Focusing on GPAI offers immense, tangible benefits: it promises genuine, measurable progress, the democratization of AI development, and ultimately, a much wider societal impact. It grounds our ambitions in what is actionable and truly useful, contrasting sharply with the often abstract and distant nature of AGI.

This isn't to say that the pursuit of AGI isn't worthy – it absolutely is, as a grand, long-term scientific endeavor. But while AGI might be the distant star we navigate by, GPAI is an immediate, well-lit path beneath our feet, a compass guiding us to real-world impact.

I'm personally committed to these ideas and to helping develop this vision. If you're interested in collaborating, please feel free to send me a direct message or leave a comment below. Let's build this future together.

Subscribe now

I've been teaching at college for close to 15 years now. In this time, I've changed a lot, as a person and educator. I've developed my own mentality about teaching, why it matters, and how to best do it.

This article attempts to organize and structure my teaching. What follows are the principles, practices, and tools I use. (Actually, this one is only about the core principles. The rest will follow soon.)

This article is very personal and does not claim anyone else should do the same. These ideas work for me, and my teaching style, and my primary reason for sharing them is to give some of you a couple of ideas you may want to try—nothing more.

Fair warning, none of what you'll read in this article is entirely original. I got most of these principles and practices from other educators I've had the privilege of knowing and learning from. Great ideas are meant to be stolen.

However, all of this is tinted and limited by the lens of my own experiences, motivations, and ideals. So, feel free to disagree with anything and everything I share in this article, and take whatever you find relevant or interesting for your practice. As usual, I'd love to know your thoughts!

Principles

These principles inform my teaching practices. I've tried to distil them down to core ideas that can be shared, but of course, there is a lot of nuance in between that got lost in the translation from lived experience to words.

In general, I think these principles may apply to education as a whole across all disciplines and levels, but my experience is almost completely limited to college-level STEM or adjacent disciplines. Needless to say, take all of this with a spoonful of salt.

To me, the most essential teaching principle, and the one that guides it all, is that...

Learning happens in an environment. Basically, there is nothing an educator can directly do to ensure learning occurs inside a student's brain. You cannot simply dump your knowledge there and call it a day. All we can do is influence the environment around the students so that learning emerges as a natural and almost inevitable result.

That environment is primarily composed of the relationships between students and teachers and among students themselves within a shared space—the school—which is why for many, the school, teachers, and students are all that matters.

However, there is much more going on that influences and may ultimately determine the quality of learning. There are parents and friends and other people around students, social media, TV, and video games, and the social and political landscape in which students live. Hell, there is even climate change and wars on the other side of the world—or across the street. You get it.

As educators, we need to be cognizant of these factors, even if we cannot directly modify many of them, so as to design learning experiences that effectively support the type of environment the students live in. Otherwise, no matter how good our lessons and books are, learning cannot happen.

For this reason, lessons need to be at least somewhat ingrained in that environment. I've found that the best way to achieve that is to...

Start with Why. Nobody learns anything if they're not paying attention; people only pay attention to things they care about. As we just said, the environment ultimately determines the outcome of any learning experience. So you have to start by convincing students they care about the stuff they're about to learn.

One way—the worst way—to do so is by invoking exams or grades. A slightly better but still awful way is to invoke some abstract future need—the very common "believe me, this will be useful when you're graduated". Even worse if that future need is just another subject this one is a prerequisite for.

A much better way is to find relevant problems students care about today and build from there—problems that are part of their environment. This is easier said than done, though. Students don't care about many of the really important problems they maybe should care about, often simply because they don't know why they should care about them.

So start with that why. Make it personal, if possible, or at least as close to their personal experience as reasonable. Not all problems can be framed in terms of the current state of society, but many times, we can find some connections, even if the thing you're actually teaching is somewhat abstract.

Don't overdo it, though. A poorly framed connection can be counterproductive because students realise this is just an artificial motivational device rather than a real problem they should care about.

But crucially, in the quest to find pragmatic motivations, don't forget to...

Ask the Big Questions. Even if we prefer to ground the motivation for learning something in personal experience as much as possible, that doesn't mean we shy away from asking big, existential questions.

Almost all practical matters can be abstracted and extended to encompass the meaning of life, metaphorically and literally. And even if we don't get to the big answer in the end, asking the big question elevates the subject and shows students that we are not only concerned with practical matters—although practical matters are often the priority.

Also, the big question can sometimes be the primary motivation. Especially in senior years, once students have learned lots of practical skills in other subjects, having at least one subject matter focused on the big questions can be a great change of pace and a chance to engage in different kinds of critical thinking and learning.

This leads me to what I think is the best order to present complex ideas, that is...

Intuition before formalization. Many classes will begin with the formalization—give you some abstract concepts and definitions someone else thought—and then try to justify why those make sense. Or worse, just take the formalization for granted and move on to solve some artificial problems.

I've found it much easier and useful to tackle a specific formalization once students already have an intuitive grasp of the concepts. Even more, we can build the formalization together, as if they were rediscovering the theory. With a little nudge from the teacher, students will often arrive at the same theory because good theories just make more sense than bad ones.

This principle can be seen as a particular case of a more general principle, that is...

Show, then tell. The original sin of modern education is the "do as I say, not as I do" model. We teach students they should behave in a way—they should think in a way—but rarely show them that behaviour ourselves.

In formal sciences, for example, we tell students that real-life problems are nuanced, that modelling is the most important part of problem-solving, and that they should think critically and creatively. Yet whenever we show them how to solve one problem, we already have a preconceived "correct" modelling and a scripted solution. They rarely see us teachers thinking critically and creatively.

I don't subscribe entirely to the anti-traditionalist ideal of “show, don't tell”, though. Educational environments are purposefully artificial because they are controlled environments where it's safe to fail. So, there is a limit to how much we can show, and we must complement it with telling. But we should embrace the mindsets we advocate as much as possible.

However, in the end, imitation is just one part of learning and not even the primary one. The best way to learn almost anything is to...

Learn by doing. Some subjects can be learned passively, but most essential skills students need are best acquired via active and deliberate practice.

This means any effective learning environment must encourage and support lots of practising. This includes both showing and telling students what to practice and, perhaps more importantly, how to practice effectively.

Plenty of studies, plus common sense, tell us that time spent practising is far less important than how one actually practices. Mindlessly solving the same type of math exercise over and over, for example, is probably a waste of time if you want to build problem-solving skills.

This last part is often forgotten, and thus, students are inundated with resources to practice—exercises, term projects, past exams—but no instructions as to how best to distribute their time or how to evaluate their practice performance. This is why frequent and personalized feedback is the most valuable thing students can get from us.

And if doing is the best way of learning, doing things in teams is even better, because...

Cooperation beats competition. Some students love competition and get a lot of energy from it, and that's good. But more often than not, students prefer cooperation.

Cooperation allows for richer interactions and is often even more challenging than competition. Managing team roles and distributing tasks in a term project is an additional skill that complements whatever technical knowledge they'll gain by just doing the project alone.

Also, cooperation allows for tackling bigger problems—not only bigger in quantitative terms but bigger in scope. Projects that involve background research aren't precisely defined and require some blue/red teaming, for example.

This doesn't mean competition is completely out of the picture. Healthy competition can be a strong motivator. However, when there's competition in my classes, I prefer that it be against an objective standard rather than against other students and that students compete in teams.

Finally, when designing team projects and evaluating work in cooperation, don't forget that...

Every student is different. Some thrive at analytical reasoning, others are better planners and organizers, and still others excel at remembering and recalling large swaths of knowledge.

While we want to form generally well-rounded professionals with a balanced set of relevant skills, this doesn't mean we must strive for, nor desire, an equal mix of abilities in all students. On the contrary, the best teams are those whose members complement each other.

Yet, formal education curricula tend to be designed for one particular type of student—the type of student who is good at gaming exams—and often leave little space to accommodate diverging needs and abilities that may also be as relevant and useful as those that are the primary goal of the educational system.

As educators, we can alleviate this narrow-minded view of schools as factories by designing learning environments that cater to varied learning styles and ensuring there is more than one valid path to success in our courses.

Moving on

This article is already quite long, so I'll leave the practices and tools for a follow-up. Those are even more personal and limited in scope, so they might not be terribly interesting to many of you. If there's interest, please let me know in the comments, and I'll prioritize that article.

In the meantime, I'd love to know your thoughts on these matters. Whether you're an educator yourself or a student (now or a while ago), I'm sure there's something we can all gain from your experience.

Until next time, stay curious.

Leave a comment

We are living in the most technologically advanced era of human civilization. Through sheer ingenuity and lots of deliberate effort, we have conquered time and time again the greatest scourges of our species. For a significant part of the world, long gone are the ages of illiteracy, widespread hunger, and the constant threat of death from a myriad of diseases.

Science and engineering have brought us a surplus of resources that only kings could have imagined centuries ago. We have mastered the sky, the sea, and the closer outer space. We have built communication lanes, both physical and digital, that extend throughout the globe. Living a healthy and wealthy (intellectually and otherwise) life has never been easier.

However, we also live in one of the most challenging eras of human civilization. We have stressed our planet to critical levels, and how much environmental damage can be reversed in the short term is unclear. The same technology that has brought us to the apex of civilization now threatens to demolish the very institutions that have long guaranteed the continued improvement of our lives: democracy, science, and media are under siege from misinformation campaigns and censorship attempts.

Last but not least, we have failed to ensure everyone is on the same boat, enjoying equally the marvels of technical progress. While a part of the world has pretty much all they can wish for, another part suffers from those same evils we supposedly eradicated: hunger, disease, illiteracy.

Some view these facts as proof that technological progress is inherently unsustainable and claim we are on the verge of an unavoidable civilizational collapse. Others point to the undeniable technical progress of the last few centuries and hope that, if we just leave the free markets to do their thing, it will all turn out to be the best in the end.

We do not fully align with either techno-pessimists or techno-optimists in thinking that there is a predetermined outcome of technological progress. Instead, we believe technology can be used for good and for evil, and that we, as a society, have both the power and the responsibility to decide how to make the best of it.

We are the techno-pragmatists, and this is our manifesto.

Our beliefs

We believe in the broadest notion of humanism, interpreted as an affirmation that each of us have both the power and responsibility to shape our own destiny, under our own terms and values, in harmony with each other and the rest of the world. We believe that seeking the welfare of all individuals, regardless of their place of origin, gender, race, beliefs, and other personal characteristics and preferences, is the ultimate purpose of humanity.

We believe in the enlightenment ideals of logic and rationality as effective means to attain true and useful knowledge. We also believe in the necessity of empirical validation, as no plan, no matter how reasonable, survives its encounter with reality.

We believe the scientific method is still our most effective tool to progressively reveal new pieces of the puzzle of reality, even if, ultimately, the true nature of reality may not be entirely knowable. We also believe there can be other tools effective for gaining knowledge and wisdom in specific domains of the human intellect that need not contradict the principles of rationality and empiricism.

We believe knowledge and information should be as accessible as possible, and that everyone should be empowered to seek their own intellectual growth, to the best of their abilities, and aligned with their personal interests. We also believe it is fair to protect some information, when it is proven that doing so is in the best interests of society.

We believe that technology can produce net gains when deployed responsibly and made sufficiently accessible. We also believe some applications of specific technologies should be discouraged or even actively disallowed, provided sufficient evidence of their harmful effects.

We believe that when deployed within reasonable boundaries and subjected to the right incentives, free markets are effective tools to incentivize innovation and can act as catalysts for technological progress. We also believe some problems are too important to be left for markets to solve and involve social commitment at a large scale.

We believe in the power of public institutions, bestowed upon them by the collective agreement of reasonable individuals, to define and implement the best strategies to keep society healthy. We also believe they should be held accountable for their performance and subjected to the highest standards of scrutiny from independent evaluators.

We believe that all tolerant voices should be heard, especially the dissenting ones, and that many truths —though not necessarily all— are relative to the context of their application. We also believe that extreme intolerance should be dealt with strongly and swiftly.

Our commitments

As researchers, we commit to pursuing knowledge and truth wherever it leads, whether it confirms or contradicts our suppositions. We also commit to keeping our biases and egos in check to the best of our ability and to disclose all potential conflicts of interest in our work. And we commit to submitting our work to our qualified peers for independent and transparent replication or falsification, and to provide the same service to the rest of the community.

As technologists, we commit to the responsible development of advanced technologies, carefully considering the potential harm their deployment can provoke, physical, psychological, financial, sociological, environmental, or otherwise. We also commit to thoroughly evaluating risks and benefits before advocating for the widespread adoption of any technology, regardless of our personal implications in its development, especially when it impacts the most vulnerable minorities.

As educators, we commit to the open dissemination of knowledge and information as far and wide as our practical means allow, rooted on scientific grounds and powered by critical, holistic thinking that considers the technical, sociological, and environmental aspects of the technologies we teach. If we can’t afford to give our knowledge away for free, we commit to making our best effort to ensure those less privileged have a fair chance to access it.

As policymakers, we commit to base decisions on the best available information and seek expert opinions where our knowledge is lacking. We also commit to engaging in honest and good-faith disagreements with our fellow policymakers across the political and ideological spectrum to achieve the best collective outcomes.

Finally, as citizens, we commit to educating ourselves on the benefits and potential harms of all the technology offered to us. We commit to exerting our choosing power, via purchases, votes, consumption, or any other means at our disposal, to direct the markets and the society towards futures that are beneficial in the short term but still sustainable in the long term.

Techno-pragmatism is accepting that the future is not predetermined. That we have the power to decide among many potential futures and the responsibility to make that choice based on reason and evidence, respecting the plurality of interests of all our fellow humans and being thoughtful about our planet and future generations.

Techno-pragmatism is understanding there is no free lunch, only trade-offs, but that doesn’t mean all possibilities are equally good or bad. Some possibilities are objectively better than others, and it takes significant effort to find them.

Techno-pragmatism is committing to the unending, tireless reevaluation of one’s position towards any specific technology. To not fall prey to cheap heuristics or ideological banners. To seek a progressively more refined understanding of the role of technology in shaping society and acting in consequence.

We, the techno-pragmatists worldwide, don’t always agree on the hows, but we share a common ideal: to foster collaboration and innovation in the responsible development of technology, ensuring it benefits a majority and its potential harms are contained and well-understood.

Signed off,

Alejandro Piad Morffis, Ph.D., Machine Learning, Cuba.
Nicolas Potkalitsky, Ph.D., AI Literacy and Education, USA.

This manifesto is a work in progress. A level-headed, reasonable response to extremist views on both sides of the pessimist-optimist spectrum. If you agree with the sentiment and want to add, change, or remove anything, feel free to leave a comment!

And if you want to sign in, let me know too!

The mind of a computer scientist operates in a superposition of three distinct modes: the scientist, the engineer, and the hacker.

These modes, when combined, enable computer scientists to approach problems uniquely, setting them apart from other fields. This way of thinking is not inherently flawless or even superior to others, but it brings a unique blend of skills that makes us computer scientists thrive in certain types of challenges.

When I say “thinking modes,” I’m not talking about personality types or other pseudo-scientific nonsense. Experienced computer scientists can seamlessly switch between and combine these different thinking modes. These are just tools for framing the same problems from different perspectives and getting the most out of them.

This is how it works.

The scientist mindset compels you to question all your assumptions. It’s what drives you to search for the underlying reasons behind things, to keep peeling the layers of understanding, to keep asking why until there’s no further answer.

There are multiple ways to embody the scientist mindset.

It can be expressed by asking why and striving for a comprehensive understanding. It involves clearly stating assumptions and remaining impartial when judging solutions. It entails being cognizant of your biases and conducting experiments to validate or refute them.

The scientist mindset extends beyond just researchers. While the nature of research clearly requires a scientific approach, every computer scientist possesses a part of the scientist's essence. When faced with a problem, we formulate hypotheses and design experiments to confirm or disprove them. It enables an unbiased pursuit of truth and a determination to uncover the underlying facts.

The engineering mindset revolves around comprehending the functionality of things and ensuring their functionality. It involves solving problems with practicality, using the available tools without being overwhelmed by unnecessary complexity.

Additionally, the engineering mindset prioritizes safety and acknowledges potential edge cases when designing solutions. It recognizes that unforeseen contingencies will arise despite thorough planning and must be addressed at the moment. Hence, some flexibility and adaptability should be built in to handle unforeseen circumstances.

Furthermore, engineering embodies the concept of trade-offs. It entails striking a balance between achieving perfection and accomplishing things. This mindset enables one to understand the concessions necessary to achieve a solution now instead of aiming for the ideal solution that will never come.

The hacker mindset is all about problem-solving. It involves unraveling and resolving problems for the sake and joy of finding a solution beyond any practical or theoretical considerations.

This mindset seeks to comprehend the inner workings of things and determine how to address and overcome issues because it brings enjoyment rather than it must be done. The hacker's mindset is not limited to hacking computer networks or bypassing online security, as typically associated with the traditional meaning of “hacker”. Instead, it is about relishing in challenges and complex problems simply for the satisfaction of solving them.

Being a hacker is about proving yourself your capability to conquer a difficult challenge. It pushes you to maximize the performance of algorithms, discover the most optimal approach to designing a data structure, or identify the cleverest method of implementing a feature because it is fun.

If the scientist embodies thoughtfulness, and the engineer embraces pragmatism, then the hacker exudes playfulness.

Every computer scientist I know embodies a combination of these three thinking modes. But not in a fixed, choose-your-character-type kind of way. It is not a matter of assigning percentages to each skill to build a perfect problem-solver. Instead, these diverse mindsets are all used together at all times, in opposition to and complementing each other.

When confronted with a complex challenge, computer scientists simultaneously draw upon their inner scientists, engineers, and hackers. Although all of us may lean more towards one persona or the other, any successful computer scientist must be capable of invoking any of these three roles when required.

To computer science undergrads reading this, I advise embracing and nurturing these three mindsets. Avoid confining yourself to a single mindset and restricting your potential. Embrace the fact that you contain multitudes. It only makes you stronger.

Recently, some sensible people have argued that Substack is becoming sort of an MLM or pyramid scheme, in which a whole lot of amateur writers are trying to convince our sucker friends to also become amateur writers by selling them on the idea they can make a living out of writing online if only —and here’s the catch— they subscribe to our own writing.

Subscribe now

And yes, there’s some of that, probably more than it’s healthy. To be clear, anyone who tries to sell you on the idea that making a living out of writing online is easy is lying to you, lying to themselves, or both. People far more clever and informed than me have pointed out many times just how brutal the competition is in the writing business. Only a tiny portion of the writers in Substack make any money at all, let alone enough for it to be a significant part of their income. (I’m certainly not one of those.)

And if that’s your game — if you want to break into the writing industry and become the next Pulitzer winner or NYT bestseller— that’s awesome; I really hope you make it. But if that’s not your game, there are still plenty of genuine reasons to write online beyond any financial or professional motivation.

The simplest reason is just that writing is good for you. It makes you a better thinker. Writing about technical topics helps you deepen and consolidate your knowledge. Writing about the state of the world will help you see the bigger picture and figure out what the hell is going on. Writing about your daily struggles and concerns will help you put your problems in perspective and understand how you feel about them. Even writing about mundane things will help you debug your emotions and psyche.

So, just writing for yourself is already a perfectly valid reason. But if you can find a community of 10 or 200 or 1000 readers that resonate with your words and cherish your writing, that’s when the real magic begins. You can think of them as an audience, but you don’t have to. You can also just think of them as a community, a bunch of disparate people who share some common interests or goals.

In I think of my readers primarily as people who want to gain some new understanding about some cool topic I happen to know a bit about. In I see them as a community of people interested in honing their craft in technical writing. In I treat them as explorers of a wonderful and surprising world we’re building together.

In each of these cases, I write primarily for myself, but your presence makes it so much better. Would I still write the same articles, short stories, and essays if no one would read them? Probably not all of it. Some of the things I write are heavily motivated by the feedback you give me. But the vast majority of it is just weird ideas that I want to explore and understand more deeply, and writing is just the best tool for that.

So, I don’t think there’s anything inherently fishy or disingenuous in trying to convince you all to become writers, even if only occasionally. If you are not playing the numbers game, then there is no fixed pie to share —and if you are, well, great for you.

There has never been a better time to start writing online. And I can give you some advice, for sure, but there are others far better equipped than me to do so. Just from the top of my head, check out , , and for some guidance.

So go, start writing online today. You most likely won’t ever make any significant money from it, but who cares?

Being a better writer just means being a better thinker. And everyone would benefit if we all just learned to think better.

This will be a significantly more loaded and sentimental rant than it is usual for me, and I’ll touch some topics that may be triggering for someone going through rough moments. If you don’t need any extra sadness or anger right now, feel free to skip this essay.

In 2019, my father suddenly died. He was very healthy and didn't have any medical condition. His heart just stopped working one day.

And there was that. There was nothing we could do, nothing we could blame, nothing we could learn from it. Except that, this can and will happen to any of us. It’s one of those mysteries of life, you know, that the best thing about it is that it can stop at any minute, or some bullshit like that.

As you can imagine, his death sent all of our world into chaos, especially my mother, of course, but also me and my sister, who suddenly lost one of the fundamental pillars of our life.

My father was that person in your life who would always be there. He didn't always agree with my choices, but he always supported them. And he always put himself in my shoes.

He wasn't the most careful speaker. He could really make you feel like shit sometimes, because he would tell you the things he thought without any sugarcoating.

And I would get angry at him, but it also was, in a sense, extremely refreshing because you would always know what he was thinking about, and whether he thought something you did was great or not, he would tell you.

So, he passed away, and there was no lesson to take from it, other than life can be as long or as short as it can be, and there is nothing you can do about it. So deal with that however you have to.

I mean, I've had loved ones die before. I had grandparents who passed away when I was a kid, and I felt sad about them, of course, really sad. But as a kid, I was mostly sad because everybody around me was also sad, and it was hard for me to comprehend why, but I just knew sad was the right way to feel.

However, my father's death was the first moment in my life where I felt this indescribable pain that is almost numbing, that you basically think your capacity for feeling has been all but obliterated, that your feeling sensors have been overloaded, and that you will never be able to feel happiness, or sadness, or anything else again.

That's not true, by the way, as many of you will know. You do get to feel happiness again. In fact, the two most joyous moments of my life so far have been after my father's death, when my two daughters were born.

So yes, you recover. But you also get scarred, and those scars do teach you something. For different people, these will be different lessons. For me, the most important lesson I took out of my father's whole life was this:

Only people matter. Everything else is secondary.

Here's what I mean by this.

We do almost everything we do, from leisure to work to destiny, because we think they are meaningful in some sense. More often than not, that meaning comes from this notion that something is bigger than us and that something matters.

So we say the community matters, and we’re doing this for the sake of the community.

Or we say the company matters, the product matters, and we're doing this because we want our company to succeed and we care to make the best possible product.

Or maybe we say the country matters, and we’re doing this because we want our country to grow and flourish.

Or sometimes, we may even say that humanity matters, and we do this because extending the light of consciousness into the universe is essential.

Whatever floats your boat.

The point is, we create concepts that give meaning to the things we do. And then, what usually happens is these concepts will get a life and an identity of their own and we will forget that all of these big, important entities are just collections of people. A community, a company, even a country, or the whole of humanity is just a group of people with somewhat common interests.

We create these concepts, and we ascribe them meaning. And then we say we're doing things for the sake of the community, the sake of the organization, the sake of the country, when what we really mean is we're doing it for the sake of some specific people.

There is nothing intrinsically wrong in taking a group of people, putting them under some kind of abstract collective entity, and then reasoning as if you were doing things for the collective rather than for the individual people. It makes sense, because sometimes you don't even know exactly who those people are.

They can be your students, they can be your co-workers, they can be the readers of this article. They can be children in a faraway country, or they can be all people on the planet who care about some specific idea.

The harm comes when we forget it’s just people all the way down. There are at least two reasons why this is dangerous.

The first is when this collective entity gets so diluted and so abstracted away that we forget who are we doing the things for, and we end up doing things for the wrong reasons —or for no reason at all.

The second is that some people will attempt to hijack that concept and ask you to do things for the abstract collective when what they really want is for you to do things that benefit them.

Let me give you a couple of examples.

The first kind of danger is best exemplified in big, necessarily bureaucratic organizations that are, in principle, benign, but then bureaucracy grows so large that everybody ends up doing things just because they must.

I work at a university, and I'm sure everybody there will tell you that the things we do matter because we do them for our students and colleagues. And those people matter.

But then you get this bunch of bureaucratic stuff, that, to some level, makes some sense —like methodological paperwork, and reports, and summaries of those reports, and balances, and meta-reports— at some point, you start asking who am I doing this for?

Your boss will tell you the administration needs this, or the ministry needs this, or the academy of whatever needs this. But at that moment, you may ask, well, who exactly is the administration, the ministry, the academy? Are there actual people in the administration whose lives are better because I'm doing this?

Maybe, because they have to make some tough decisions —they have to allocate resources or decide what to prioritize— and so their jobs are easier and better because I'm spending time doing these stupid reports and balances? If that’s the case, then doing them may make total sense. Those people matter, and they care.

But oftentimes, if you follow that chain of thought, you will realize that there is nobody at the end of the chain. It's just a circular fucking chain.

These people receive these reports and balances and just send them to someone else, who sends them to someone else, who sends them back. Nobody at any point will read them, arrive at some enlightenment, and take some action that will transform something that will make somebody else's life or work better.

Everybody is working for an abstract system, and nobody benefits. No one at that point cares, so no one matters.

Examples of the second type of danger really abound. In the corporate world, you will often be asked to do something for the sake of the company, the product, or the customer.

And then you can ask, but who is the company? Is it me and all of my other co-workers? Or is it just the people at the top, the people at the board, who profit ultimately from my work? Am I doing this because it serves the users of the product? Does it really make their lives better? Or are we doing this simply because someone will profit from those customers and give nothing in return?

But by far, the most prevalent example of people hijacking some collective concept for their own benefit is in politics. Politicians will often claim that something must be done for the sake of the community, for the sake of society, or for the sake of the country. But if you ask around and dig deep, you will find that this only benefits their partners and patrons.

So, what I'm going to do about this is try to be as conscious as possible of the people behind the collective abstraction. Who am I working for? Make sure I do everything I do because I believe there is someone who matters, someone who cares.

When I spend time with my family, my friends, or my little girls, instead of doing something “productive”, it’s because they matter more than anything else and because they care about me spending that time with them.

When I teach my classes, I spend nights and nights not with my kids but in front of a computer reading and writing, I do it because my students matter, and my students care about those lectures and classes.

When I write papers and go through the hops of the academic publishing video game and have to please editors and reviewers, I'm doing it because I think there is somebody, somewhere, whose life can be improved down the road when this technology or this idea gets applied. (Or is it?)

When I write educational articles in this blog, spend tens of hours making sure the content is approachable but still technically accurate, and then publish it for free, I do it because I think you matter and you care about this stuff.

But if I think something I'm doing, whether it is some dumb, bureaucratic, or enlightened bullshit, doesn't benefit anybody that matters and cares, then I'm simply not gonna do it.

I don't care who gets annoyed and I don't care if it threatens my job or my career. This is the most important lesson I learned from my father and the best way to honor his life.

From this day on, only people matter.

For this reason, competitive programming has become an increasingly popular element of Computer Science education. Many introductory courses in algorithms and data structures lend themselves easily to this format. It also simplifies evaluation, as problems are graded simply by a well-crafted set of input/output pairs.

Subscribe now

As a CS college professor, having taught introductory programming courses and specialized courses on algorithms, I have observed many pros and cons when fostering a competitive mindset in my students. In this essay, I want to explore ways to enhance CS classes with a bit of competitive programming flavor while avoiding its main caveats. I'm focusing on competitive programming for the purpose of clarity, but I think the main takeaways of this post apply to many other forms of competition in education.

Benefits of the competitive mindset

Many students excel in competitive environments because the drive to win serves as motivation. This competitive environment can push students to reach their full potential and develop valuable skills.

In competitive programming, problems are rarely straightforward. Students must exercise creativity to decode the underlying problem. Even once the real problem is identified, applying a standard algorithm is often insufficient. Problems may require combining two or more classic algorithms unconventionally or implementing a non-trivial change to a data structure. To come out on top, one must master complex programming problems that require a deeper understanding of algorithms and data structures.

Additionally, participating in competitions encourages students to code quickly and anticipate errors because feedback is indirect —they only know if they solved the problem or not, but not which test cases failed. This cultivates creative and innovative thinking, valuable skills that extend beyond competition.

Participating in competitive programming entails collaborating with a team to solve a specific problem while coding on the same computer. Although this setup is only used during the competition, students must practice working in teams to develop a collaborative mindset. Over time, they establish strong bonds that often endure throughout their careers. Typically, teams consist of at least one member who excels in coding speed and another with solid math skills. This method of working in teams fosters trust and encourages students to divide tasks based on each individual’s strengths.

However, there are several drawbacks to overemphasizing a competitive mindset. In the following section we’ll examine these potential downsides. While I’ll be discussing the negative aspects in greater detail, it doesn’t necessarily mean that I believe they outweigh the benefits. However, it’s often easier to argue in favor of the positive aspects, whereas the negative aspects require more careful consideration.

Possible harms to be aware

One major disadvantage of competitive programming is the missed opportunities for many students who do not achieve significant success beyond their university studies. I have witnessed numerous students focusing solely on competition after competition, neglecting their other subjects but still not reaching the top. These are not poor-performing students, as they could have excelled in other subjects if they had redirected their efforts. However, the nature of the competition is such that only the winners are recognized, leaving the second-best teams with little recognition despite their hard work and dedication.

Note that extreme cases where one team consistently outshines all others are rare. Typically, the top two or three teams will alternate winning in different events, and multiple prizes are often available. However, when considering the 10 or 15 teams that never place in any competition, it becomes clear how much talent and hard work are sacrificed in a competitive environment. This is a downside of competition in general; for a few to win, you need a long tail of equally talented and hardworking participants who will lose.

An additional concern with focusing too much on competitive coding is that it can lead to the formation of bad habits. Competition problems are often artificial and come with pre-defined inputs and expected outputs. It’s also common for the expected performance to be provided, making it easier to determine if a linear, quadratic, or brute-force solution is needed. However, real-world problems are rarely so straightforward, and students who excel in competitive programming may struggle when faced with open-ended problems that lack a clear path forward.

To succeed in competitive coding, particularly during a competition, students must rely on their instincts as there is no time to unit-test or formally prove a solution’s efficacy. However, this approach can lead to poor software engineering practices in the long run. It may also promote rote learning of pre-made solutions instead of genuinely comprehending the fundamental principles of computer science.

I have noticed a particularly concerning issue among some students. They have become so accustomed to winning that they shy away from challenges where they cannot achieve maximum success. As they progress in their careers, projects become more realistic, and there is no limit to performance - there will always be room for improvement. Additionally, there is often a need to balance trade-offs, making it challenging to optimize performance metrics. These students don’t just struggle with these problems, but actively avoid them, choosing not to participate in research groups or tackle interesting open problems due to a fear of not being at the top.

Lastly, it is the emphasis on competition can have a negative impact not only on the students who directly participate in competitive coding activities. If we base an entire course around a competitive mindset, it can negatively affect those who do not excel under pressure. As any teacher knows, students come in all different flavors and have countless ways of demonstrating creativity, intelligence, and talent. Forcing them all through a competitive funnel can stifle their potential. I have witnessed this disproportionately affect marginalized students, but it is not limited to them. Any student who struggles with working under pressure will see their performance and motivation impacted in an environment overly focused on competition.

Now that we’ve thoroughly discussed the pros and cons, let’s briefly discuss additional strategies to implement in a competitive environment. These strategies will help students who find it challenging to perform under pressure by fostering a more encouraging and supportive atmosphere.

Alternatives to competition

A straightforward way to provide students with flexibility is by allowing them to revise assignments after completion. For instance, in my programming classes, students participate take coding exams that mimic the competitive coding environment to an extent: they are constrained in time and resources.

However, I always remind the students that if they don’t perform well during the exam, there’s still an opportunity for improvement once they get home. They can rework their solutions and submit a better version to their professor for evaluation. This somewhat reduces the stress on exam performance alone. Their TA will assess the initial attempt made during the exam and subsequent revisions and give a final evaluation that considers not only what the student managed to solve during the exam but the ideas they had afterward. This approach acknowledges that factors such as time constraints or nervousness may lead to mistakes and allows students to demonstrate their proper understanding of the material.

Another way to create flexible assignments is by providing open-ended projects for students to work on in teams. These projects involve a loosely defined problem, and the students must refine and define it correctly. This approach benefits those who may not excel when given strictly limited constraints. For example, posing an open-ended question like “How can we improve healthcare using machine learning?” allows students to explore various aspects of the problem through research, interviews, and critical thinking. As a result, they may develop more innovative solutions than initially anticipated.

And finally, I also encourage my students to demonstrate their learning in any way they can think of beyond the assigned tasks. Submitting well-formatted lecture notes, expanding on in-class examples, or solving additional problems from textbooks or online sources are all valuable ways to showcase their understanding. They are free to work on independent projects or delve deeper into the theory only briefly discussed in lectures. I assure them their efforts in exploring and sharing these areas with me will be appreciated and acknowledged in their final evaluation as a testament to their growth and progress in the subject matter.

The key takeaway is that flexibility is crucial when designing assignments. Recognize that each student has unique abilities and expectations; there isn’t a one-size-fits-all approach to learning. By being adaptable with your teaching methods, you can better cater to diverse needs and facilitate effective education for all students.

Conclusion

Competitive programming allows students to develop their coding abilities as they work together to solve complex problems. It also encourages students to think outside the box, as they must develop creative solutions to complex problems. Additionally, competitive programming can foster collaboration and a sense of camaraderie among students as they work together to reach a common goal.

On the other hand, competitive programming can also create an environment of unnecessary stress and competition. As a result, it can cause students to become discouraged or overwhelmed. Additionally, competitive programming can lead to students exclusively focusing on problem-solving instead of learning the underlying concepts of the discipline.

Ultimately, every student deserves a chance to explore competitive programming and discover if it interests them. It can be a great way to increase problem-solving and critical thinking skills and hone coding abilities. If students enjoy competitive programming, it can provide a great platform to develop and showcase their technical skills.

However, it is essential to strike a balance when incorporating competitive programming into the classroom. While providing an environment where the most competitive students can flourish is beneficial, it is also essential to remember those who do not thrive in competition. Instead of solely focusing on competitive programming, providing mechanisms better suited to different learning styles is just as important.

I have argued before about how beneficial it is to foster the mindset of being an effective team player. While competitive programming can be a valuable team skill, it should not be the only one. Encouraging collaboration among students and having them work together to learn will significantly benefit them in the long run. As educators, our role is to build a diverse classroom environment that caters to different learning styles so that students can benefit from each others’ strengths and create something better than they could on their own.

This post was written in collaboration with , a brilliant student I had the honor to teach and a hugely successful competitive coder during his college days. He now shares his hard-earned insights on algorithms at . Check it out!

Of course, a priori, neither way is strictly better than the other. Both have advantages and limitations. I have used both in different combinations in my classes, courses, and other learning experiences I’ve designed. In this essay, I will discuss when these two approaches fit best and how to mix them for an optimal learning experience.

Subscribe now

Bottom-up learning is very efficient and practical for massive education. If you’re building a college curriculum, you can define the theoretical principles supporting all skills and applications you want your students to learn. You can then design a learning path that efficiently covers all the theoretical details necessary for each subject matter. For example, before machine learning, you teach the mathematical principles behind calculus and algebra. This also allows the learning materials and evaluations to be more standardized, making it easier to deploy in large educational institutions. It doesn’t matter if your major is physics or chemistry; the calculus class is probably the same. Thus, this approach is preferred in any centrally designed educational system. It is an efficient way for students to understand the whole material, as each course element is touched only once and explained in depth before it is necessary. And it’s an efficient way to schedule and optimize your resources (teachers, classrooms, exams) at scale.

But trying to teach an unjustified theory to students can be difficult. One must constantly tell the students, “Please believe me, this is useful; you'll see why later.” Furthermore, since you have no grounding applications, to begin with, your evaluations tend to be artificial —e.g., solving abstract equation systems—leaving the students feeling like they are being forced to learn things that seem to have no use. This can be frustrating for the students as they are expected to understand and remember theories that may have no immediate relevance to their lives.

In contrast, Top-down learning is a great way to keep students engaged. By having a clear problem in mind and then discovering the theory piece by piece, students can relate to and understand the importance of the theory to make the solution work. Seeing the application first also makes the theory easier to comprehend and remember than when dealing with, e.g., some abstract theorems. When students see the theory applied to a concrete problem, they can understand intuitively why the theory works before going through the hassle of proving it. The application also helps motivate and inspire students and sparks an interest in the studied subject, presumably because the application is something they care about. For example, try to build a chatbot first, diving into language modeling as you discover its necessity. This approach is often used in short tutorials and online courses. The desired result is a concrete application since you can skip many elements of the underlying theory and focus on the most useful ones.

However, for the same reason, it can be hard to justify going too deep into the underlying principles and processes when learning purely for an application. Instead, the goal is to gain enough knowledge to understand and use the application effectively; and then you move on. Repeating this approach for many related problems can lead to a disconnect between the individual pieces of information, as no grand theory unifies them. This is why it can be beneficial to delve a little deeper, as this may provide a unifying framework that ties all the pieces together. For example, learning algebra allows us to understand how operators work on different mathematical objects, while learning logic will enable us to unify all forms of reasoning. Going the extra step to uncover the more profound principles can be worthwhile in the long run, even when it is not strictly necessary for the application.

So which is better in which case? I don't have any proven answers here, but I sense that the larger the scale, the more valuable the bottom-up approach becomes.

When designing a full curriculum, it is essential to consider the relationships between principles and applications. Visualizing the whole dependency graph of principles and applications makes it easier to organize the curriculum to ensure overarching principles are only covered once while allowing them to be applied and reused throughout the curriculum. This helps ensure that students are exposed to the relevant principles and have the opportunity to understand and use them in various contexts.

However, in small-scale learning projects, such as a single class, I believe is better to begin with an intriguing problem and then move on to understand the theory and techniques necessary for its resolution. Immersing oneself in a subject matter or problem first makes staying motivated and engaged in the learning process easier. Additionally, it allows the learner to be more familiar with the subject by the time they reach the theoretical aspects of the project, making it easier to absorb the necessary information.

Things get interesting at the mid-level when designing, e.g., a complete course. Here, I have found that a mix of practical and theoretical approaches gives you the best of both worlds: a hybrid model that maintains student engagement and ensures that all necessary foundations are firmly established. This hybrid model consists of alternating lectures between surface applications and deep theory. You start with a motivating application and highlight what you need to understand further. Then you take a deep breath and spend three or four lectures studying the underlying theory. Once students have a solid grasp, they switch gears again and devote a couple of lectures to direct applications, solving the initial problem as if by sticking parts of a puzzle. There’s an aha! moment there where the theory pays off, and you have gained your students’ trust so that you can push them again into another theoretical rabbit hole. By alternating between these two activities, students feel they are consistently progressing and can better understand how theory-based concepts directly relate to their desired applications.

Take, for example, my compilers course. I begin by introducing the grand vision of a compiler and the key milestones we need to hit to build it: tokenizing, parsing, semantics, and code optimization. This is the motivating application; they all want to know how this compiler is built. There I promise, “Come dive with me for a while; when we’re out, you’ll see we’ll be much closer,” and I immediately dive into formal language theory. I will do this three or four times in a classic 16 weeks course, which always pays off. We gain momentum as we hit each milestone, and students are even more eager to see what the next theoretical deep dive will uncover. But this workflow also allows them to understand better the theory underlying the compiler construction process because each piece of theory is grounded in some relevant application.

Sometimes this hybrid approach is hard to pull off. If the subject you’re teaching is very abstract, finding realistic, motivating examples may be almost impossible. But in most cases, what I’ve seen is that college courses don’t try hard enough. Algebra, calculus, differential equations, and probability, …, all these are subjects classically taught at a very abstract and flavorless level. And all of these can be significantly improved with just a few motivating applications sprinkled here and there. If you try it, I can promise students will not only love it; they will learn better, which is the ultimate payoff.

This is a long rant about some things I think are wrong in Academia and some ideas about how to improve this situation. I mostly focus on Artificial Intelligence because that's my field, but I think most of these ideas apply everywhere. I'm not trying to discredit or criticize any individual or organization, but rather raise some questions that I think all of us scientists, as a community, should attend. I declare myself as guilty of all the sins I describe.

The setup

If you've ever tried Academia you have surely been in this situation. You come up with a good idea, do some experiments, and write a paper about it, and... that's when the real work starts. Whether you send that paper to a conference or a journal, you'll get 2 to 5 reviewers to critique your paper, ask you for improvements and decide if your work is good enough for publication. If it's a conference, you'll usually either get accepted or rejected, but if it's a journal, you might get a second chance to improve and resubmit. This process is called peer review, and it's one of the fundamental pillars of Science.

Don't get me wrong, peer review is extremely important. You see, Science is a social process. Yes, you can follow the scientific method and come up with a Frankenstein monster all by yourself on a private island, and you would be doing science (without capital "s"). It is only when those results are scrutinized, retested, and confirmed by additional researchers, that they become part of the continuous and incremental body of accumulated knowledge that we call Science.

Peer review is a fundamental part of this process because it ensures that you are not deluding yourself into believing what you want to believe. It also guarantees we all follow the same high standards of openness, honesty, and goodwill. However, problems arise when the means become an end in itself.

Since peer review is such an important concept in Science, we have built all our social scientific processes around it. We set deadlines, ratings, and whole systems to formalize and organize what peer review means. We have double-blind and single-blind peer reviews to guarantee that authors and reviewers don't take revenge on each other. We have evaluation forms and protocols, and we have workshops and workshops about peer review.

And yet, time and time again, experiments have shown that reviews are significantly inconsistent. If you randomly redistribute the papers at a top AI conference, a large part of the accepted papers get rejected, and vice-versa. However, I do not take this as evidence that scientists are lousy reviewers. Not even close. Scientists are pretty good at being objectively critical of others and our work, we do that every single day! I think the problem lies in the system and the incentives built around it, mostly for the benefit of the big players in the Academic world, the publishers.

Subscribe now

The symptoms

Every time you take a metric and turn it into an objective, it ceases to be a useful metric. This has happened in Science with the concept of publishing a paper. Publishing a paper is the main mechanism for socializing research. A research paper usually describes some scientific hypotheses in as clear terms as possible, a protocol to test (i.e., falsify) those hypotheses, and an honest and critical discussion of results and their implications. By reading a paper, fellow scientists can come up with additional hypotheses or ideas, and build on top of previous work. And every time you use someone else's ideas as part of your own, you are supposed to include a citation. This is what Newton was referring to when he said he had "stand on the shoulder of giants".

In time, the most significant scientific discoveries should get a large number of citations, because everyone building on top of your ideas would cite you. Hence, a large number of citations is seen as a sign of scientific achievement, and that is often taken as the One Metric of Academic Success. See the problem here?

Once citations become a distinction mark, everyone tries to maximize them. A lot of strategies begin to arise, like publishing lots of low-effort papers instead of fewer and better ones and working only on the most fashionable topics. Since to get cited you have to get published first, publishers become the gatekeepers. A feedback loop starts to build in which publishers try to be as exclusive as possible to attract better papers, since more citations imply more readers which implies more subscriptions; and authors try to aim for the most exclusive publishers since, otherwise, they won't get enough citations.

In this dynamic, two very harmful things start to happen.

First, scientists spend a lot of effort and money, very often public money, on research that never gets published because of the massive competition. Ironically, once that research made with public money is published, is often put behind a subscription paywall, which most universities and institutions subscribe to. So taxpayers end up paying for research twice, once when done by Alice and again when Bob wants to read Alice's paper. Isn't that crazy enough?

The second issue is more subtle but far more harmful. In this process of out-competing each other for citations, we forgot what's important about Science. It's a social process designed to improve human life by solving humanity's most pressing problems. But this competition, far from what free-market ideologists could believe, only serves to undermine the very purpose of Science:

• The most fashionable topics get the most attention, and those are often not correlated with the need of the many.

• Also, scientists are not born, they are educated. If competition is so fierce that junior researchers don't get a break, we end up losing the best minds before they get a chance to shine.

• And finally, this constant competition for citations discourages any kind of self-critical research, any analysis of negative results, and any replication study, because no one will cite you for saying "Yeah, I retested this, and it does seem to work as they originally said...".

This discussion started with peer review, and how the whole academic publishing is built around this concept. Now is the time to criticize it. Since scientists are forced to compete for attention, we have turned peer review from the supportive and self-healing process it should be into the most unpleasant part of doing research. To be fair, not all reviewers are nasty, and when we do, I'm arguing it is more often than not because we are forced by the system.

Continue reading on Substack

The new paradigm

I believe the root of the problem in this picture should be clear by now.

So, how do we realign the incentives of scientists with the original purpose of Science, and make it better for everyone? Honestly, I don't know. But I think we can take some ideas from the FOSS community to at least foster some good practices which I believe might put us on the right track.

The idea starts with embracing openness in the whole process of scientific discovery and innovation. This is not my original idea, of course, there are some commonly shared principles of "open science" in the academic community. This is one possible way to express them:

• Open Methodology: Document the application of methods and the entire process behind them as far as practicable and relevant.

• Open Source: Use open source technology (software and hardware) and open your own technologies.

• Open Data: Make the data freely and easily available.

• Open Access: Publish openly and make publications usable and accessible to everyone.

• Open Peer Review: Provide peer review in an open and public forum.

• Open Educational Resources: Use free and open materials for education and in university teaching.

In this form, these principles are quite abstract, and there are many ways in which they could be implemented. There are plenty of degrees of "open science" like publishing in open-access journals managed by non-profit organizations, publishing pre-prints before submitting to "traditional" journals, and all the good practices around making data and protocols publicly available.

I want to focus on some key ideas I think could be fruitful to try, without implying that this is the absolute solution to this problem, but rather a small part of a much larger paradigm shift that Science has to undertake.

The practices

These are my proposals. Most of them relate specifically to the peer-review process because, as I said before, this process is a pillar of the scientific process, but also because I think this is the one place where we as a community can innovate the most, without requiring the government grants or changing the way bureaucratic institutions work. The peer-review process is at the base of the entire scientific process and any major change in its functioning could have a massive impact up the chain.

Public reviews

Let's start by acknowledging that single- and double-blind reviews are more harmful than helpful. These measures are supposed to shield reviewers and authors from future retaliation and disallow any form of favoritism, which should make the review process more just and honest. In practice, they shield reviewers from criticism and make the whole review process less transparent.

I propose to turn this concept around and make all reviews completely public. We have to trust we are all reasonable individuals and professional scientists, who should be able to provide objective judgment without favoritism. But if we don't, then our reviews themselves are public and subject to review and criticism.

This is very easy to implement with any workflow that allows posting comments on a public forum. Note that I don't necessarily mean that anyone can review (this is discussed further down) but even if only specific "official" reviewers are assigned to a paper, both their comments and their identity and credentials should be public.1

Continuous peer-review

This idea ties in with the previous point. Currently, almost all peer reviews (that I'm aware of) happen in the context of some specific conference or journal. What I'm proposing here is to detach the peer-review process from any journal or conference and make it instead integral to the paper. Every paper would carry around the Internet with all its reviews, and if rejected at some previous point, a future conference or journal editor would have access to the full history of reviews and changes to reconsider the paper for "mainstream" publication (we'll talk more about what this means later).

I can see this happening similar to how issues are handled in Github. You publish a paper, and potential reviewers would open "issues" against it, one for each important thing to address. Issues would be discussed and worked on in public and there would a history of every change introduced into the paper with links to which issues are being fixed.

Since no paper is perfect, conferences and journal editors should not aim for publishing issue-free papers, but rather papers that show a healthy list of open and closed issues and demonstrable usefulness in their current state. A healthy list of open and closed issues would be an indication of a solid paper, the same way as for software.

Encouraging reviews

The next problem I want to tackle consists of how to kick-start the reviewing process. Once we detach reviews from specific conferences or journals, how can we guarantee everyone has access to good reviews? For sure, rockstar scientists will get thousands of reviews but what about junior researchers who are just starting?

One idea is to see reviews as an integral part of the scientific career. Researchers should be evaluated also in terms of how much value they put back to the community, and one way senior scientists can contribute is to review junior scientists. We should be proud to put in our CVs how many reviews we have given. And good reviews, which are in turn evaluated positively by the author and other reviewers, should count towards one's scientific output.

Scientists would get a "badge" with the number of reviews they have given, and display it on their homepage, their LinkedIn or ResearchGate profile, etc. This badge would link to some online list that links back to all reviews. This could be maybe hosted in our ORCID profiles or any similar non-profit initiative.

Also, senior researchers are part of a community and are often connected with like-minded individuals in other institutions and countries. It should be part of their work to look up to each other's students and junior researchers. And yes, someone will say "but then you can give a good review to my students if I give a good review to yours". Again, this is why all review is, first and foremost, public in nature.

Qualitative evaluations

Now let's move on to specific review formats. Too often I see very long lists of checkboxes and 1-5 ratings, etc. I believe there is value in having a structured evaluation template, to make sure we more or less agree on what are the core issues we should care about. But going to the extreme of having 10 different ratings for a paper is insane! What is the difference between 6 and 7?

When putting reviewers under the pressure of giving numerical scores, we are asking them to unconsciously introduce all the biases they have about that particular problem or field or approach, or author. There is simply no objective way to numerically compare two different papers.

A good research paper needs to have a solid methodology (correctly apply the principles of science as it is common practice in that field), provide relevant results and conclusions (either positive or negative), be feasible to reproduce by independent researchers, and have a clear presentation. Either the paper is good enough to be considered publishable, if all these aspects are covered, or it isn't. That's it.

I prefer a simple evaluation form that asks: "Is this aspect of the paper up to the scientific standard?” and a space for you to explain why you evaluate each aspect as such.2

Methodology     [x]👍 [ ]👎 
Results         [x]👍 [ ]👎 
Reproducibility [ ]👍 [x]👎 
Presentation    [ ]👍 [x]👎

A specific conference or journal might want to evaluate the potential impact or significance of a paper before accepting it for publishing. But impact or significance is not what Science is about. There are however legitimate cases where impact or significance is important.

If you have to allocate a restricted pool of resources (e.g., grant money) of course you want to evaluate the impact. Yet, I argue this is not part of the peer-review process, but a posterior analysis that each institution or publisher should do based on their specific criteria. Peer review should be a process by which the scientific community as a whole evaluates that some research is sound science, irrespective of idiosyncrasies.

Self-publishing

Now that the peer-review process is completely detached from the "mainstream" publishing industry, who decides when is a paper ready to be published? Well, of course, the authors! It is up to the authors to determine that, given all the feedback received, they consider their work is production-ready.

All papers would be published first in draft mode, perhaps even before being completely written. During the draft phase, you collect all the feedback you can from peer reviews and work on the issues you consider more relevant. When you feel it is good enough (possibly because most of the recent reviews are favorable) you hit that Publish button and create a release. If some errors appear, later on, you fix them and publish another release.

What becomes citable then? Easy, each release of each paper gets a unique DOI that will forever point to that exact version, together with all its metadata and reviews. If I cite something yours and criticize it, and you later fix it, that's OK. My critique is still valid because it points to a previous version that is indexable, and the fact that you fixed it only speaks higher of you!

But wait, won't authors publish a lot of low-effort papers to engross their CVs? Well, maybe someone will but, who cares? We started by saying that, intuitively, citations should be a good measure of scientific quality. This is still true in this format. If I release a bunch of crappy papers, no one will cite them.

And also, who thinks CVs are useful? Anyone trying to evaluate me as a researcher would not look into some list of titles and numbers I pasted into a Word document. They would go to my researcher profile and see my most significant work, the reviews it has received, and how my whole research process works!

This doesn't mean that I get to decide whether my work is relevant, however. This just means I get to decide whether my work is ready to be consumed by the scientific community. The community will still judge my work's relevance by citing it, criticizing it, and in two more ways I left for the end.

Conferences for networking

Now that all papers are being published by their authors, what's the purpose of scientific conferences? We can now recover their original purpose. Conferences were created as a medium to get like-minded scientists together to share their experiences and discuss the most relevant problems in their field. But as conferences became more and more a mainstream path for getting published, their organization has become more and more about managing the peer-review process.

Now that peer review is detached from conferences, their organizers are free to focus on scooping what are the most interesting topics and the most significant results in those topics and invite those speakers they believe will bring the biggest value. Which papers get to be presented? I think we could deal with that in two ways:

As an organizer of a thematic conference, I would spend half of the year looking around for interesting papers to invite their authors. They would still pay for their accommodations (or their institutions would) and they would come to enjoy what's best in every conference, the networking.

I could also open a call for papers, as usual, but authors would submit papers that are already reviewed and released. My role would be to decide, based primarily on thematic fit, what I think is more relevant for my community.

There is even no need to attach participation to a published paper. Authors could simply submit "talks", possibly backed by one or more papers that support their submission, as it is already common in some conferences.

This would completely reshape what conferences are for (at least in my field). There is no reason why we should wait for the top conferences of the year to be able to read papers. We would go to conferences for the chance to talk with the researchers we admire about their work.

And someone asks, but how would conferences compete if they have no publication rights? Well, I argue this would be very good. Conference organizers would have to compete on the grounds of providing a better environment for networking, interesting events, and nice amenities, but no one would have a monopoly on the knowledge itself. There is even no reason why the same talk cannot be presented at more than one conference if enough people are willing to listen.

Journals for socializing

And finally, we come back to the original culprit, the infamous research journal. Now that papers are published openly, what are journals good for? Well, what they were originally designed for, is socializing research! Journals were created as a means for academic societies to collect the most relevant research in a given community and publish it for a larger audience.

The commercial publishers arrived and turned science into a business, and journals became paywalled gatekeepers of knowledge, that require original research often paid with public money that they resell again for public money. The largest academic publishers often state they have costs to cover, but there is plenty of evidence that they make a significant profit. And that's OK, but if I as a journal editor want to make a profit, I'm gonna have to innovate.

Like conferences, I could scoop around and feature the most interesting papers in some thematic issues, maybe ask the authors to give some new comments on them, prepare explainer videos, add links, and put some effort into turning those "raw" papers into beautifully typeset pages.

SOTA reviews would be a nice fit for journals as well. These are not original research papers, but they often provide a lot of value by analyzing a bunch of papers and giving advice on common trends or highlighting interesting lines for future research.

I would also have editorial articles specifically written for an issue that could summarize in layman's terms a particular subject, to introduce it to a larger audience. I would even pay scientists that are good communicators for this work.

Journals would have to compete on the grounds of being good at selecting topics and papers to socialize and provide some additional editorial value. In any case, original research papers would be owned only by their authors and would be published always with some public license (e.g., Creative Commons). This would ensure that Science belongs to the ones who ultimately pay for it, that is society.

The system

Putting all these ideas together in a functioning system will require a lot of work. From the infrastructure point of view, I envision something aking Github, a repository of open-access papers with built-in comments, reviewing, and social features. Ideally, it would also have a web UI for editing, similar to Overleaf and, of course, fully integrated with Git. I understand this might not be the best solution for academic communities that are not very closely related to software, i.e., social sciences, mostly because it could pose a significant learning curve for their members and become more of a hinder than a help.

From the social point of view, kick-starting such a system would require a massive community effort. And not because of the infrastructure cost, that's a minor issue. I think the largest obstacle to this kind of paradigm shift is that a large part of the community would have to move away simultaneously from journals and conferences as the main publication channels. Otherwise, the few that start the effort will be completely disconnected from the rest of the community.

I can see this happening as an effort from, say, the AI community, or any other technically-savvy collective. Tomorrow morning, all the senior scientists that publish in ICLR, ICML, NeurIPs, and ACL, could suddenly decide they want to go fully open. It would require the conference organizers to support the initiative as well. Instead of opening a call for papers, the conferences could decide that they would open a call for submissions, which should be peer-reviewed and published in this new format.

Some non-profit organizations could be formed from within the community to provide the infrastructure. Since there will be some operative costs, this platform would require some payment, but it would be very small compared to publishing fees in most major open-access journals. Also, some of the big players in the industry could support this initiative by providing hosting and infrastructure for free. This would be a big PR boost for these companies.

The metrics

We started this discussion by saying that an intuitively good metric to estimate scientific impact, i.e., citations, had become the target and thus lost its entire meaning. But citations are not an inherently bad metric, it's just when we use citations as the one quantitative metric to compare individual researchers that we miss the entire point.

Likewise, the fact that we self-publish all our papers doesn't mean that being featured in a major conference or journal is worthless. On the contrary, when everything we publish is open, being featured in a mainstream publication becomes an even better measure of impact, because it is no longer tied to my financial capacity or any other unfair advantage I might enjoy in the community. This would be very good for Third World researchers, who produce valuable science but are often cut off from mainstream publication for reasons completely unrelated to the quality of the research.

If we can restructure the incentives and processes of Science such that they are aligned with the purpose of Science as primarily a means to improve human life, everything else would fall into place. Once researchers are free from predatory publishing practices, meaningless numerical statistics, and unhealthy competition, I believe we will all focus on what we love most, doing sound research for the good of mankind.

Then, all those metrics that are used today will regain their meaning. Being invited to a top-tier conference would mean what your community wants to hear from you. Being featured in a top-tier journal would mean that some editors consider your work is high-quality. And being cited often would mean that your research is producing a real impact and that you are becoming a giant on whose shoulders others can stand.

I hope you have enjoyed this post. This is way longer than usual, but it’s a topic I feel very passionate about. Let me know what you think. And as usual, if you believe anyone you know would enjoy reading this stuff, feel free to share it with them!

Share Mostly Harmless Ideas

Modern college education, with its hyper-focus on quantitative evaluation, delayed and sparse feedback, and penalization for failure, is the greatest obstacle to creativity. In this post, I explore alternative ways to frame college education, and specifically evaluation, such that we nurture instead of stifling our students’ creativity. This opens the path for rethinking how can college remain relevant when gatekeeping information is no longer a valid business model.

I have been teaching Computer Science in college for over a decade now. During this time, I have helped over one thousand students learn how to code, and do various cool stuff with computers. Some get there way faster than others, and some never do. And while much of their success can be explained by their effort, we teachers play a significant part. A part that can make all the difference between a student who drops mid-semester, and another, with the same background, initial skills, and family support, who graduates and then goes on to become the next CEO of a thriving software company.

I will talk about my own experience teaching Computer Science, but I think these ideas extend to all other college majors, and perhaps to most if not all educational levels.

Subscribe now

Computer Scientists are often tasked with solving problems that are at least slightly different from anything anyone has ever done. Sure, there’s lots of repetition in coding another Android app, but most of our graduates will end up working at places that are pushing the frontier of knowledge, even if in a small way. Succeeding at this requires creativity, which in this post I will loosely define as the ability to produce good and novel ideas. By good, I mean ideas that work, or at least ideas that get you closer to a solution. By novel, I mean ideas that are slightly different from most of what everyone else has done —even if, as structuralists think, everything is somewhat a mixture of everything else.

The problem with creativity is that, at least in this simplistic definition, it is very rare. Many of us just can’t pump great after great ideas. No, we have lots of bad ideas for every good one. But there’s an easy way out: you just have to grind through all the bad to get to the good ones. It’s more of a matter of quantity over quality. But there’s a catch: how do you know an idea is good or bad? Bad ideas often seem good from far away. Most of the time, you’ll have to test them to see them fail miserably and learn something about it. This is what I call the virtuous cycle of creativity:

You come up with an idea, test it, and then reflect on whatever the outcome is. More often than not, that idea is lousy, or mediocre at best. By testing it you become aware of its flaws. By reflecting on it, you adjust your internal ideation mechanism (whatever that is in human cognition) to make it more likely to produce a slightly better idea next time. Rinse and repeat.

Mozart, Beethoven, and Bach were some of the best composers of all time (at least for Western classical music) and also some of the most prolific. And you’ll see this time and time again, not just in art, but in science, engineering, sports, and business. Yes, some people are one-hit wonders, stumbling over an extraordinarily good idea either by dumb luck or extreme talent. But most of the really good ones are just relentless in producing and filtering ideas.

The first question I want to tackle in this post is the following:

My claim is that college education, especially college evaluation, is by design a major obstacle to creativity because it hinders our students’ capacity to loop through this cycle of ideation as fast as possible.

To see why let’s consider the most important element of the ideation cycle: feedback. Looping through the cycle requires having access to feedback about the performance (or quality) of a given idea. That’s what the testing phase gives you. The better the feedback you get, the more effective your reflection will be, and the more you will learn for the next iteration. There are four key characteristics that I believe valuable feedback has: it is frequent, timely, informative, and safe.

In contrast, college evaluation (and thus, the feedback provided by it) is often sparse, delayed, opaque, and unforgiving. Let’s take them one by one.

The first characteristic of valuable feedback is being frequent. It is straightforward to understand why frequency is important, as it directly ties to how often you can test a new idea. College, in contrast, presents students with very few opportunities to get valuable feedback in the form of partial exams and projects. It gives them sparse feedback. Thus, students have very few chances during a regular semester to go through the full cycle and adjust. In my experience, most college projects only get feedback at the end, when there’s already no chance to improve.

The second characteristic of valuable feedback is being timely. This means, being as close as possible to the actual testing of the idea. Even if you get frequent feedback, if it takes on average a month between putting the idea out there and learning something about it, by the time you can reflect on it you’ve already moved on to trying new ideas, and it gets really hard to go back and adjust from an experience. College evaluation tends to be rather delayed since professors need time to grade exams and projects. Thus, when students receive that feedback, they’re already moved on, probably to another project or another subject, and have very little time left to reflect.

The third characteristic is being informative. This means knowing not only whether the idea was good or bad, but why. The more informative feedback is, the easiest it will be to adjust your ideation mechanism for next time because you will know exactly (or close to) what to change. In contrast, most feedback students get in college is an opaque grade: a number or letter that just says how “good” their performance was. However, even when we tell them what was the exact thing they messed up (e.g., this step in this exercise is wrong), we rarely tell them what they should have done instead. Students are left to guess what they could have done differently to achieve the expected result.

Finally, valuable feedback has to be safe to obtain. This means you risk nothing or almost nothing, except of course your time (and maybe a minor financial cost). If getting feedback is costly (in monetary terms or otherwise) then you’ll be incentivized to take as few loops as possible. This will make you concoct complicated schemes in which you mix several ideas at once and make only one experiment, but then it becomes extra hard to elucidate what exactly went wrong. College evaluation tends to be unforgiving, in the sense that bad grades early in the semester can drive down the final grade, which is taken to be some kind of average of your course performance. The ultimate cost of feedback is when it stops you from even getting future feedback, something college professors are very fond of doing: penalizing underperforming students by removing their ability to take the final exams and sending them straight to “extra” exams.

The question I want to tackle next is how, or even whether, this state of affairs can be improved. For that, we should look at the reasons why college evaluation is so bad. For starters, it is not, as many students seem to believe, because professors are sadistic people who enjoy torturing their students (well, maybe some are, but not the vast majority). The reason ties in with something I’ve written about before: Evaluation is seen as an end in itself, and not as another means to improve students’ education.

If our only purpose in evaluating students is to determine who gets to pass and who has to repeat, and we do not consider the (positive or negative) effects it may have on their learning, then we will aim for efficiency. This is precisely what we have now: a dry process designed to sort students and place them into buckets with as little effort from evaluators as possible. Do only the minimum amount of exams that give you enough information to make a final decision on who passes. Make it as generic and quantitative as possible so they’re easy to grade. Give students back the minimum amount of information to cover your ass in case of complaints. This is how we end with an evaluation that is sparse, delayed, uninformative, and unforgiving.

You see, college and academia are built on the foundation that access to knowledge is very limited, and college professors are the gatekeepers of the secrets of the trade. This was true for millennia, but the Internet changed all of that. Now, information is out there, ready for grabs, free, and accessible for anyone with a minimum living standard —and yes, I know this is a minority of the world, but still orders of magnitude more people than those who have college access.

We need to consider that evaluation is ultimately a form of feedback, probably the most relevant. And we need to understand that valuable feedback is perhaps the only useful thing modern colleges can give students.

So let’s rethink all of the college education centered around providing feedback instead of information. Let’s redesign our evaluations as means for students to get frequent, timely, informative, and safe feedback ‘round the clock. Let them do as many evaluations as they want. Check their projects with them weekly, or even more frequently. Become another team player instead of an outsider who’s only there to grade them. Replace quantitative grades with qualitative, informative descriptions of what they did wrong and how they could do it better. Downplay the significance of grades in partial evaluations and reassure them that it is safe to fail and that there’s always time to show how much they have learned later in the course. And then be true to that promise, and grade them according to how much they know when the course ends, and not with the nonsense averaging of partial grades.

I’ve tried to do all of this in my classes, and I can’t say that it has been a total success. It is very hard, not in the least because the rest of the system fights back. College administration demands quantitative grades and a prearranged calendar of sparse partial evaluations. Students are not used to this mindset, and they are cautious to show you their flaws for fear of being penalized. And also, it takes a lot of effort from professors. But if there is anything worth saving about college so that it survives the Information Age, is precisely this: being the most valuable source of frequent, timely, informative, and safe feedback our students could get.

Modern education started with the Industrial Revolution and the need to graduate a ton of skilled workers to carry on the same tasks over and over. Previously, education was only for the brightest and/or luckiest, and usually consisted of a very custom path through which a mentor would guide you. Nowadays in universities all around the world, we have reduced students to numbers, grades, and percentages, as if we were producing computer chips or combustion engines. Efficiency is all that matters.

To achieve the highest possible efficiency, all around the world we educators have become engineers of sorts. We designed what we call a "model of the professional"; a set of skills and competencies that an abstract ideal professional should have. Then we designed an evaluation metric that is basically the micro-average of a ton of super-narrow scores that measure super-specific skills such as taking a derivative or coding a recursive function.

Finally, we designed a pipeline that takes students on one end and produces "professionals" on the other end. Those "smart" enough to learn to beat the system get the highest grades and are stamped with an abstract generic title of Computer Scientist, Medical Doctor, or Lawyer, very much like a certificate of quality in a generic bottle of wine.

This system is deeply flawed, and educators all over the world know it and have been discussing it for a long time. It's hard to change for many reasons, the least of which is the lack of teachers willing to dump the generic instruction set and craft custom learning paths for their students. I think this system is based on two basic assumptions, intuitive but wrong. Changing those assumptions could shed light on ways to improve the system.

The first assumption is that students are a blank slate that when fed through this generic pipeline we call higher education will be magically morphed into this generic professional we designed. This is wrong for so many reasons that it is hard to acknowledge it as a basic assumption of our system.

Ask any university professor and they will all tell you the same: All students are different. They all have different skills, interests and biases. They all require a different approach to get the most out of them. And almost all of them, when given the chance and the right environment, will become the best versions of themselves. Yet time and time again we treat them as generic droids on which we can dump a generic course and expect a generic performance in return.

The second assumption, I think, is harder to spot, because of the way the university is disconnected from real life all around the world. We educators think that society wants this "model of the professional" because we think that a hospital needs 100 equally generic doctors, and a software company needs 100 equally generic programmers. However, this assumption is also wrong on many levels.

Everywhere we ask in the industry we keep hearing the same: we need unique people with unique skills that bring something new to the team. It's like trying to build an ensemble out of 100 equal models. You get much better results with a variety of approaches to the same problem than with an array of 100 exactly equal programs. Yet we keep translating what society asks into skills and competencies. They tell us they need unique people, and we add "uniqueness" to the set of generic skills we want to teach in our generic college programs!

So let's dump those two assumptions and acknowledge that we have a bunch of different kids with different interests and capabilities, and we need to turn them into a bunch of different professionals with different mindsets and skills. Now the question is how on earth can we do that? As engineers, we need to design a streamlined pipeline and a proper evaluation metric. And we need to do that, unfortunately, because there are so many more students than teachers that we cannot hope to be the Aristotle to each Alexander, and that's not about to change in the near future (no, GPT won’t become an AI mentor anytime soon, we can discuss why another time).

I think one possible strategy is to focus on a single evaluation metric and a single skill:

Let's take it piece by piece. Every student is different, so everyone will have a different set of potential capabilities that could make them effective team players. If we encourage those specific capabilities in each student, we are giving each one a different learning path. This one will focus on improving her analytical skills, that one will focus on improving his management skills, the other one her social skills, and so on.

Each one is focusing on their most interesting and desirable version of themselves. On the other hand, everyone is optimizing the same metric, being a good team player, whatever the team. Give them back to society and they will fit in the right spot. The one spot that needs that specific mindset.

Almost all low-hanging fruits that a single bright person could take are already taken, the problems that are left to solve as a society are the hard problems, and they all require teamwork. The easier problems are being automated away as I type. So, I argue, the most important skill today is being an effective team player. If we strive to turn our students into exactly that, we are giving them the best education possible, and we are giving society the best possible return on that investment.

The final question is how exactly do we do that? How do we discover what makes every student unique and valuable in a team? Isn't that the same as Aristotle & Alexander's dilemma?

I think a possible solution is simply to let each of them discover it by themselves. As educators, instead of trying to tell everyone what to do, let's focus on designing learning environments, comfortable for every student to explore their own skills and capabilities, and to decide the best way for them to serve the team. And let's evaluate them on co-op instead of solo so that, when trying to beat the system, they will effectively optimize what we, the rest of the world, need them to be good at.

If you enjoyed this short rant, feel free to share this post with anyone you feel might be interested. Nothing would make me happier!

Subscribe now