That's a solid argument, that thinking doesn't require consciousness and the process is similar in AI. However, this is blurring the term "thinking" and fitting AI inside of the blurred picture. Are calculators also thinking? WHy or why not?
From my perspective, thinking should have a set of base requirements. Perhaps it's a computation that requires:
- real-time information processing of sensory data
- recurring state that also combines in real-time to produce infinite combinations
- in-place updating of recurring state to allow for real-time learning
I'm... thinking... about this in real-time and producing something completely new based on past information and having read the article. I'm iterating on information in ways that are changing my current internal, recurring state. There is also a component of logic, but as I just mentioned, I'm improvising this on-the-fly, so there could be other requirements, this is just an example.
Whatever it is though, I think AI lacks most of these potential requirements. AI does not update its weights in real-time, it does not constantly learn, and judging by the enormous amounts of resources the frontier models require, they are not efficient in generating new "thoughts". It's also why they have "jagged intelligence" and can't seem to reason from first principles, despite Chain of Thought techniques being surprisingly effective.
I'm not convinced AI is thinking anymore than a calculator is, but that isn't to say we won't have an AI that does think, in the proper sense, in some near future.
That's a solid argument, that thinking doesn't require consciousness and the process is similar in AI. However, this is blurring the term "thinking" and fitting AI inside of the blurred picture. Are calculators also thinking? WHy or why not?
From my perspective, thinking should have a set of base requirements. Perhaps it's a computation that requires:
- real-time information processing of sensory data
- recurring state that also combines in real-time to produce infinite combinations
- in-place updating of recurring state to allow for real-time learning
I'm... thinking... about this in real-time and producing something completely new based on past information and having read the article. I'm iterating on information in ways that are changing my current internal, recurring state. There is also a component of logic, but as I just mentioned, I'm improvising this on-the-fly, so there could be other requirements, this is just an example.
Whatever it is though, I think AI lacks most of these potential requirements. AI does not update its weights in real-time, it does not constantly learn, and judging by the enormous amounts of resources the frontier models require, they are not efficient in generating new "thoughts". It's also why they have "jagged intelligence" and can't seem to reason from first principles, despite Chain of Thought techniques being surprisingly effective.
I'm not convinced AI is thinking anymore than a calculator is, but that isn't to say we won't have an AI that does think, in the proper sense, in some near future.