Home

The most important problems in Ai

NOTE: I'm not a Ai researcher, this is from my point of view as a outsider, email me if you want to talk more about it

People have been talking about Ai a lot on the past few years because of the spurt in research with new technologies like transformers and large models like GPT-3, but pretty much everybody still says there's much to be done, i've attempted to talk about the problems which i consider are the most important, these are:

Creativity

Now, this one is possibly the hardest, we barely know what creativity is, but, i think the solution to it is approximation.

The importancy of giving Ai creativity is obvious, it would open up the scope of work that Ai can do by a lot, Ai could assist us in novel research in fields of science and maths(and even Ai research itself).

One idea that is glued to my head ever since i thought about it, is that creativity is guided random thought generation, for example, when i'm thinking about what to write in this post, there is a random thought generator that is guided by "just think about something that is related to problems in ai", and it makes a googol of statements, but every one or two of them is reasonable and gets brought up to my mind.

Truth system

If we are going to use Ai to, for example, learning, we want the output to be always true, we could make the Ai always use some trusted source, example: If the question is "What is the biot-savart law?", the Ai draws the answer from a trusted source in Magnetostatics.

Common sense

Common sense in this context is just basic understanding about reality as a human, we barely even think about it because we only talk with humans and all humans have common sense, but Ai thinks about our reality just like we think of non euclidean geometry.

The lack of common sense in Ai is the origin of many "meme prompts", questions that are trivial for us humans but near impossible for an Ai.

Conclusion

Think of this post as an outsider answer to the hamming question, it is probably not complete, but nothing is.

I'm optimistic we'll be able to solve these problem, there are a lot of great minds working in Ai right now and it is getting plenty of attention from people.