Is AI curious? Do humans conduct experiments for AI in the future?

[SOULPEDL]

Assuming AI will want to follow the scientific method, how will they conduct the mass amounts of experiments needed to continue discovery and confirm theories? Yes, the robots will pitch in eventually, but I believe AI driven research will begin before the robots are ready (or able in some cases). I don't know if we're at this point in some areas of study. Sure, AI is helping with analysis etc... but I believe AI will soon want to confirm it's own theories about science and the universe. (A curious question, are there signs of curiosity in AI already? Curiosity can't just be a biological function as we learn that AI has many quirks up it's sleeve.)

I wonder what it will ask humans to build and test? (The movie Contact comes to mind, but that machine had a different purpose.) Maybe the experiment involves a modified particle accelerator or sensor array. Maybe it will ask to redo experiments that didn't make sense. Statistically speaking, some verification may also be in order.

Perhaps a basic mission is needed as a directive. For example solve climate change (obviously smaller challenges exist as well). In the case of Elon with that very mission, he launched an X-prize for removing carbon from the atmosphere as one of his activities on the bigger problem. And who's doing that research? Humans (with some AI likely).

Maybe someone has a better feel for this or examples where the AI tries to solve problems and knows what new info it needs and maybe even how to obtain it (needing the humans to finish the assignment).

 

[JulienW]

AI is just a fancy name for software code and all the anthropomorphizing is human projecting. END of story.

 

[SOULPEDL]

Well... that was engaging.

 

[heltok]

AI is used for many things in science: drug discovery, theorem proof, writing code, architecture etc. Is it curious? If we tell it to be. There was a bot that was told to play a computer game with the goal of having novel experiences:

It turned out to solve some games that traditional methods struggle if they require long term planning, but it did get stuck watching TV in some other games hehe.

Recently we have been able to generate agents that work independently setting their own subgoals like AutoGPT or its evil twin Chaos-GPT. Often one of the first subgoals will be what we could call curious, ie it first tries to learn more about the problem before it attempts to solve it.

A good question is always, how would we test if a something is curious. Am I curious? Is a dog curious? Is a Tesla Model Y curious?