Fundamentally, anything humans can do can be done by physical systems of some kind, (because humans are already such a system), so given enough time I'd bet that it would be eventually possible to make a machine do literally anything that can be done by a human. There might be some things that nobody ever does get an AI to replicate even if technically possible though, just because of not having a motivation to
Since AI is trained by us, using the fruit of human labor as input, it'll have to be something we can't train it to do.
Something biological or instinctual... Like being in close proximity to an AI will never result in synchronized menstruation since an AI can't and won't ever menstruate.
Synched Menstruation is supposed to be a myth now. I have experienced it many times, but I guess it’s mostly considered coincidence, which it could be, I’m not a mathematician.
In their new paper, the five computer scientists prove that interrogating entangled provers makes it possible to verify answers to unsolvable problems, including the halting problem.
How humans think. AI "thinking" will always be different than human thinking. Because human brain is "that thing" that is impossibile to simulate in silico as is. We might be able to have good approximations, but as good as they can get, they'll always diverge from the real thing
I guess a good part also comes from learned experiences. Having a body, growing up, feeling pain, being mortal.
And yes, the brain is an incredibly complex system not only of neurons, but also transmitters, receptors, a whole truckload of biochemistry.
But in the end, both are just matter in patterns, excitation in coordination. The effort to simulate is substantial, but I don't see how that would NEVER succeed, if someone with the capabilities insisted on it. However, it might be fully sufficient for the task (whatever that is, probably porn) to simulate 95% or so, technically still not the real deal.
Funny enough I have the opposite opinion, human brains are the type of thinking we have most experience with - so we've devised our input methods around what we notice most, and so will be able to most easily train the AI.
I also believe that we'll be abke to reduce the noise to a level lower than actual person variation fairly easily, cause an AI has the benefit of being able to scale to a populous size - no human even has that much experience with humans
I use to work on research on microscopic mechanisms of the brain, and I work in AI.
Human thoughts derive from extremely complex microscopic mechanisms, that do not "average out" when moving to the macroscopic world, but instead create very complex non-linear stochastic process that are thoughts.
Unless some scientific miracle happens, human thoughts will stay human.