KEN FOREMAN
Re: Does AI need a “body” to become truly intelligent? Meta thinks so I completely agree with this.
I’ve seen several respected luminaries argue that LLM is not “true AI” or “Strong AI” since it’s based on large learning sets and predictive behavior. They argue that humans and animals are not taught on such large language models or data sets.
What are education and experience, if not Large Learning Models based on the teaching of schools, universities, and books?
It’s been argued by multiple sociologists and psychologists that Language defines Learning. If your language does not define a word, concept, color, skill, or experience, you have no way to understand or communicate it. This is the very basis of movies like “Arrival” (Ted Chiang’s Story of Your Life) and novellas like Ted Chiang’s Lifecycle of Software Objects.
Likewise, it is our physical experiences that teach and reinforce our learning. Touching a hot stove teaches us or reinforces us not to touch a hot object again.
Bodies and physical interaction will drastically accelerate AI, much as the original textual GPT and LLM did.
TIM KELLOG
Really? definitions always come after understanding. You absolutely have to understand a word before you can define it. Like “yeet”, there’s no way on earth someone sat down and made a definition for it before using it. It seems like language is more of a linguistic representation of experience (which, imo, is kinda what LLMs are doing too, in a way). Also, not sure what any of that has to do with “true” or “strong” AI
KEN FOREMAN
How would you communicate “yeet”?
To a non-English speaker, how would you communicate “yeet” without shared language and concepts?
Without a framework of language, you can have all the experiences you want and completely unable to share them with others.
Newton was the first to “understand” and define the law of universal gravitation, but F = G(m1m2)/R2 doesn’t need to be understood to be shared and communicated.
FERALROBOTS
what they are if not that is much, much more than that.
“I am a stochastic parrot and so are you” [to quote Sam Altman] is a RADICALLY insufficient way of understanding not just human intellect, but also the intellect of any animal we routinely interact with, too.
KEN FOREMAN
How would you communicate “yeet”?
To a non-English speaker, how would you communicate “yeet” without shared language and concepts?
TIM KELLOG
with motions
KEN FOREMAN
Also a good answer, but an equal justification for why we should give AI/ML a physical body for physical expression, interaction, and communication?
TIM KELLOG
tbh i’ve never thought about it, but my intuition says yes. if language is all that’s needed, then you could teach a child to ride a bike just by telling them how. but that’s not how it works — a lot of our knowledge isn’t represented in language.
KEN FOREMAN
If human muscles or locomotion had an instruction set, then it might be possible. However, we can’t communicate while riding a bike because we have no language for locomotion control.
Mathematics works as a shared language but also as an instruction set.
The same might also be true of LLMs as applied to physical motion and interaction?Not sure we’ll know until we try.
TIM KELLOG
oh! there was a paper a couple weeks ago that i forgot to save that talked about how LLMs, when asked to translate french to chinese, will first translate to “english”. not actually English, but a “instruction set” heavily based on english. so, LLMs actually do derive an “instruction set” before learning more complex concepts, and it’s something you can observe