“I will be honest: We keep finding things that are mysterious, even unsettling,” he said. “We find structures that mirror results from human neuroscience. We find evidence of introspection. We find internal states that functionally mirror joy, satisfaction, fear, grief and unease. I don’t know what that means, but I think it warrants ongoing discernment.”
Chris Olah works on "Mechanistic Interpretability" (<a href="https://en.wikipedia.org/wiki/Mechanistic_interpretability" rel="nofollow">https://en.wikipedia.org/wiki/Mechanistic_interpretability) and therefore makes absurd claims based on mere parallels (if any).
See also A Comprehensive Mechanistic Interpretability Explainer & Glossary for some detailed background - <a href="https://dynalist.io/d/n2ZWtnoYHrU1s4vnFSAQ519J" rel="nofollow">https://dynalist.io/d/n2ZWtnoYHrU1s4vnFSAQ519J
"Consciousness" in Humans is a form of "I Know It when I See It" threshold condition (<a href="https://en.wikipedia.org/wiki/I_know_it_when_I_see_it" rel="nofollow">https://en.wikipedia.org/wiki/I_know_it_when_I_see_it) since it is based on self-experience and then extrapolated to other members of our own species.
Whether that same attribute can be applied to other living species (from our pov) which share our planet is not accepted universally. Aspects of it certainly, but the whole as we experience it is denied to them.
The same is applicable even more absolutely when it comes to AI machines.
As regards Morality/Ethics/etc. uniquely "Human" attributes, they are a function of one's effect on the world and hence in so far as machines are allowed to affect the world autonomously, they must be taught to obey our Morals/Ethics as Rules to be followed i.e. a form of "operant conditioning"(<a href="https://en.wikipedia.org/wiki/Operant_conditioning" rel="nofollow">https://en.wikipedia.org/wiki/Operant_conditioning) and "methodological behaviourism"(<a href="https://en.wikipedia.org/wiki/Behaviorism#Varieties" rel="nofollow">https://en.wikipedia.org/wiki/Behaviorism#Varieties) applied to AI training.
rramadass · · focus · HN ↗
“I will be honest: We keep finding things that are mysterious, even unsettling,” he said. “We find structures that mirror results from human neuroscience. We find evidence of introspection. We find internal states that functionally mirror joy, satisfaction, fear, grief and unease. I don’t know what that means, but I think it warrants ongoing discernment.”
Chris Olah works on "Mechanistic Interpretability" (<a href="https://en.wikipedia.org/wiki/Mechanistic_interpretability" rel="nofollow">https://en.wikipedia.org/wiki/Mechanistic_interpretability) and therefore makes absurd claims based on mere parallels (if any).
See also A Comprehensive Mechanistic Interpretability Explainer & Glossary for some detailed background - <a href="https://dynalist.io/d/n2ZWtnoYHrU1s4vnFSAQ519J" rel="nofollow">https://dynalist.io/d/n2ZWtnoYHrU1s4vnFSAQ519J
"Consciousness" in Humans is a form of "I Know It when I See It" threshold condition (<a href="https://en.wikipedia.org/wiki/I_know_it_when_I_see_it" rel="nofollow">https://en.wikipedia.org/wiki/I_know_it_when_I_see_it) since it is based on self-experience and then extrapolated to other members of our own species.
Whether that same attribute can be applied to other living species (from our pov) which share our planet is not accepted universally. Aspects of it certainly, but the whole as we experience it is denied to them.
The same is applicable even more absolutely when it comes to AI machines.
As regards Morality/Ethics/etc. uniquely "Human" attributes, they are a function of one's effect on the world and hence in so far as machines are allowed to affect the world autonomously, they must be taught to obey our Morals/Ethics as Rules to be followed i.e. a form of "operant conditioning"(<a href="https://en.wikipedia.org/wiki/Operant_conditioning" rel="nofollow">https://en.wikipedia.org/wiki/Operant_conditioning) and "methodological behaviourism"(<a href="https://en.wikipedia.org/wiki/Behaviorism#Varieties" rel="nofollow">https://en.wikipedia.org/wiki/Behaviorism#Varieties) applied to AI training.
No anthropomorphizing and hand-wringing needed.