FWIW I always felt my contribution would have been towards “real” AI rather than the current pseudo-AI stuff that is all the rage. In addition to being a programmer, I also have a background in evolutionary biology and neuroanatomy. The “neural networks” that are part of current AI were originally based on the understanding of how neurons function that existed in the 1970s, and they were (and are still) a poor implementation of that level of knowledge. IMHO true AI can be achieved by more accurate simulation of how real neurons and real neural networks function and how they have come about in the real world through hundreds of millions of years of evolution by natural selection.
That being said, “real” AI is probably going to be even worse for humanity than this current stuff is, so at least I’m glad that blood won’t be on my hands.
I always viewed it as an engineering challenge, without really seeing or caring about the literally existential threat that it represents. At this point in my life, however, I would not want true AI, as I mentioned in my comment.
I really don’t understand this perspective, and it’s my biggest gripe with Chollet and his ilk as well. What, in your opinion, makes neuro-symbolic AI inherently superior to (the currently better performing) statistical methods?
This is honestly a strange thing to say. The only examples we have on the planet today of what we consider true intelligence are living animals. Why wouldn’t it at least be possible that the properties of the signaling units that comprise these existing, observed intelligences (i.e. neurons) are fundamental to the task of replicating these intelligences? It seems to me it’s the “statistical methods” that have yet to prove they’re capable of producing true intelligence.
I will retreat to the fact that I said “true AI can be achieved by more accurate simulation of how real neurons and real neural networks function” and not “can only be achieved”. Time will tell which approach works better.
I don’t know who Chollet is but I’ll look him up. Thanks for the reference.
FWIW I always felt my contribution would have been towards “real” AI rather than the current pseudo-AI stuff that is all the rage. In addition to being a programmer, I also have a background in evolutionary biology and neuroanatomy. The “neural networks” that are part of current AI were originally based on the understanding of how neurons function that existed in the 1970s, and they were (and are still) a poor implementation of that level of knowledge. IMHO true AI can be achieved by more accurate simulation of how real neurons and real neural networks function and how they have come about in the real world through hundreds of millions of years of evolution by natural selection.
That being said, “real” AI is probably going to be even worse for humanity than this current stuff is, so at least I’m glad that blood won’t be on my hands.
Honest question. Why would you want “true AI”?
This is like saying “true mega bubonic-AIDS can be made if we do XYZ”.
Stop, stop making mega bubonic-AIDS
I always viewed it as an engineering challenge, without really seeing or caring about the literally existential threat that it represents. At this point in my life, however, I would not want true AI, as I mentioned in my comment.
I feel for you. Going into machine learning was my dream to. Until this. I wanted to work on true artificial intelligence based on the human brain.
I really don’t understand this perspective, and it’s my biggest gripe with Chollet and his ilk as well. What, in your opinion, makes neuro-symbolic AI inherently superior to (the currently better performing) statistical methods?
This is honestly a strange thing to say. The only examples we have on the planet today of what we consider true intelligence are living animals. Why wouldn’t it at least be possible that the properties of the signaling units that comprise these existing, observed intelligences (i.e. neurons) are fundamental to the task of replicating these intelligences? It seems to me it’s the “statistical methods” that have yet to prove they’re capable of producing true intelligence.
I will retreat to the fact that I said “true AI can be achieved by more accurate simulation of how real neurons and real neural networks function” and not “can only be achieved”. Time will tell which approach works better.
I don’t know who Chollet is but I’ll look him up. Thanks for the reference.
Edit: the downvote was predictable. Nice job.
they can’t read and, as Sartre said: the slopper knows the things it says are absurd…