Succinct and motivating overview. Really satisfying to read.
Question - no mention of Active Inference/Free Energy Principle-based approaches? Hipshot you'd classify them as 'statistical', which is arguably correct but..they are more, allowing explainability, adaptivity, and speed unlike any large-scale number-crunching stat-based approach. Statistical (bayes, e.g.) and numeric to be sure, but also symbolic and experiential in a way LLMs will never be.
Perhaps these are the merger of the two approaches you write of...?
Thanks! Interesting point, for sure :) I wasn't aware of these approaches, so I actually had to google the terms, thanks for pointing it out. It seems to me, yes, it would somehow fit into the, let's face it, somewhat reductive idea of merging symbolic and statistical AI that I try to push at the end. (Being an article for the general reader I have to make some necessary simplifications that I hope are not too much oversimplifications :) So yes, from what I can grasp by a quick online read, I think these methods kind of fit this narrative but as you say, a deeper dive would necessarily break this simplistic dualism of symbolic vs statistical. Something I have to keep in mind for a future version, for sure!
intro book Active Inference by Friston, et al. starts with basics and powers up.
The NL-based team of Lazy Dynamics is doing *big* things and currently in US for fundraising, networking, research visit. Very active and responsive team!
There's a new IEEE standard, very recently ratified, for Spatial Web and HSML, HSTP protocols defined. IEEE-2874.
Succinct and motivating overview. Really satisfying to read.
Question - no mention of Active Inference/Free Energy Principle-based approaches? Hipshot you'd classify them as 'statistical', which is arguably correct but..they are more, allowing explainability, adaptivity, and speed unlike any large-scale number-crunching stat-based approach. Statistical (bayes, e.g.) and numeric to be sure, but also symbolic and experiential in a way LLMs will never be.
Perhaps these are the merger of the two approaches you write of...?
Thanks! Interesting point, for sure :) I wasn't aware of these approaches, so I actually had to google the terms, thanks for pointing it out. It seems to me, yes, it would somehow fit into the, let's face it, somewhat reductive idea of merging symbolic and statistical AI that I try to push at the end. (Being an article for the general reader I have to make some necessary simplifications that I hope are not too much oversimplifications :) So yes, from what I can grasp by a quick online read, I think these methods kind of fit this narrative but as you say, a deeper dive would necessarily break this simplistic dualism of symbolic vs statistical. Something I have to keep in mind for a future version, for sure!
Cool - thanks.
Some resources I found useful...sharing:
intro book Active Inference by Friston, et al. starts with basics and powers up.
The NL-based team of Lazy Dynamics is doing *big* things and currently in US for fundraising, networking, research visit. Very active and responsive team!
There's a new IEEE standard, very recently ratified, for Spatial Web and HSML, HSTP protocols defined. IEEE-2874.
An excellent recapitulation! For a look at what the future may bring and how our relationship with AI may evolve, take a look at https://open.substack.com/pub/jonathanblake/p/are-we-going-to-worship-artificial