A Musing on the Difference Between Humans and Machines

The more I have learned, the less convinced I have become that there is anything special about the human brain which cannot eventually be replicated in a machine. If one wants to claim that humans are intrinsically more special then ANYTHING which could ever be engineered, I believe the burden of proof is on them to demonstrate this claim.

In other words, I do not believe the question is IF generalized artificial intelligence (whatever that means) CAN be created. I believe the question is how do we identify and solve all the engineering problems necessary to search for, or design, on algorithm capable of satisfying our definition.

This seems so obvious to me that I'm deeply surprised when people expressed skepticism about the possibility of general artificial intelligence. Of course, maybe our species will lack the ingenuity to do it. Maybe we will superheat our planet before were able to the project. But the idea of some barrier which makes this not to a difficult problem but an impossible problem does not seem to exist in my observation.

Roger Penrose would disagree with me. A proper response to his ideas truly belongs in its own post. If I had to sum it up, I would say his argument is a physical one. He expects new physics to emerge which clarifies the separation. If and when such new physics emerges, it will be very interesting to see.

In the meantime, I was wondering what a convincing argument against general artificial intelligence might look like. I had a novel thought that I wanted to share. I'm not entirely convinced by this argument, But I think it's worth a little debate if people are interested in sharing their thoughts in the comments section.

What if the key to AGI required some sort of witness string given to an algorithm in order to manifest the intelligence. This part should be uncontroversial. It's basically the same thing as saying give I description of A touring machine to universal Turing machine. But what if you want to take away the witness string? Tell the algorithm to compute its own witness string from first principles. That seems perfectly reasonable.

What if the calculation of that witness string was provably in I'm intimidating complexity class such as nondeterministic exponential time . If anyone cares to check the complexity zoo and propose a worse class to be in, I'd be thrilled to update with your suggestion.

If it was impossible for AGI to occur without being given such a witness string, and the witness string was provably in efficient to discover, that one could argue the human who possesses the right artificial intelligence algorithm but does not possess the magic witness string would be unlikely to ever discover the necessary string.

If so, is there any equivalents in human beings? Given my limited knowledge of biology, I will not speculate too much, but perhaps something about our genetic code or some emergent process based on it has a key like this embedded. This key, by definition, was difficult to come by. So how did nature end up computing it? Since we have one example of abiogenesis, one could perhaps argue for tremendous good luck.

I am not entirely convinced by this line of thinking. Nor am I dissuaded by it. I'd love to hear any thoughts from readers in the comments.