Chess Agentic AI in C++

[modern grotto]

Hello guys. Currently I'm in an AI course and my project idea for that course is to do an AI that play chess, so I tell the idea to professor and he said project can't be classic AI so I need to do an Agentic AI (like autonomous AI). I don't know much about it, I was thinking about making a simple NN that adjust to the player chess level, but sorry if I saying bullshit , just crazy ideas, so Can you recommend resources or textbooks to implement this? The professor isn't cool because we can't use too much frameworks and things like that.

• professor expects a model of an AI agent that can learn and evolve and make decisions by itself, not just apply classical AI algorithms like just making an implementation of minimax, also i need to "express" the used model mathematically.
• AI course is starting right now, but course include topics like basic search, llms, fuzzy logic, genetic algorithms, neural networks...
  Thank you for advice!

[cloud sentinel]

When your question is answered use !solved to mark the question as resolved.

Remember to ask specific questions, provide necessary details, and reduce your question to its simplest form. For tips on how to ask a good question use !howto ask.

[umbral oyster]

Can you elaborate more on what kind of AI the prof expects?

And also what kind of AI class this is?

[modern grotto]

yep, I've edited the post

So I asked the prof and said me like I can't just implement classical algos. I need to implement something more complex but also not tooo complex because the course is four months

[umbral oyster]

professor expects a model of an AI agent that can learn and evolve and make decisions by itself, not just apply classical AI algorithms like just making an implementation of minimax, also i need to "express" the used model mathematically.
Well, then use one of the approaches you learnt though the course.

AI course is starting right now
Wait, the course is only just starting? I thought it was like a final project for it. It must have "previous requirements" then, because there's no way to do all of that without any.

[modern grotto]

Yep. The thing is that the project must be proposed now so I don't want to put something achievable 🙁

[umbral oyster]

So, basically this is not an AI course, but a machine learning one?

[modern grotto]

Oh man, I don't know the name of the course is AI but don't know

[umbral oyster]

Can you check if the course has some prerequisites?
Also, were you given any resources to help you decide on a project?

[modern grotto]

Prerequisites are formal languages and Discreet math. No there's no resources for helping me chose

[umbral oyster]

Okay, damn. That sucks. Well, I'm probably not the right person for that question tbh, I more or less just wanted to verify what kind of AI you were talking about.
Maybe look at some (older) CodeBullet videos for inspiration?

Yeah sorry, machine learning is not my area of expertise at all.
I'll leave it for someone else to answer (question is in original message, mainly OP's asking for resources/textbooks)

[modern grotto]

Oh thanks by the way

I will check if a NN is possible to do

[cloud sentinel]

@modern grotto Has your question been resolved? If so, type !solved :)

[zinc aspen]

you can also try asking in another discord server, like Engine Programming

though most of the people there are doing alpha-beta pruning, and a few are doing MCTS

still, there are bound to be a few who can help you

you can easily train a neural net that's far stronger than you are

so I think the goal doesn't have to be strength alone

this idea to "train a net that plays like a human would" already exists, btw: https://www.maiachess.com/

though it's really hard to quantify what this means

[modern grotto]

Ok, thank you. I'll see what I can do. Some modifications xd

[cloud wasp]

professor expects a model of an AI agent that can learn and evolve and make decisions by itself, not just apply classical AI algorithms like just making an implementation of minimax, also i need to "express" the used model mathematically.
this description is weird..
I wouldn't view "minmax" as an "AI algorithm"
and an "AI agent" has recently gained a quite different meaning.. mainly LLM related and make decisions based on a users request
and then there are traditional machine learning models, which are trained to do one specific thing with very specific inputs, like recognize numbers from an image

99.999% of all chess engine currently employ the latter, together with minmax, but ai isn't a requirement for minmax

[umbral oyster]

I wouldn't view "minmax" as an "AI algorithm"
Of course it is, assuming we use the classical definition of "AI".

and an "AI agent" has recently gained a quite different meaning.. mainly LLM related and make decisions based on a users request
Really? I know the term "(AI) agent" as an AI that is able to make decisions and perform actions on its own, in order to achieve a certain goal (like deleting your entire production database)

99.999% if all chess engine currently employ [ML], together with minmax, but ai isn't a requirement for minmax
Stockfish uses an optimized minimax together with a ML model that judges the final position in the path, but for example Leela and AlphaZero don't use a Minimax variation at all.

[cloud wasp]

1. I mean in today’s age… it’s not the sort of “ai” that’s been growing in hype, but if people classify that algo as AI already I’m ok with that after all it does mimic basic intelligence
2. I agree with that but chess bots are very limited in their capabilities and I wouldn’t use the term ai agent in this domain
3. thanks for explaining stockfish to me 🤣🤣 I mean yea there are other alternatives which also use MCTS for example but usually new people in this domain go the traditional way, which stockfish went, alpha beta (PVS) etc, and only later they explore other ways, the number I gave was just exaggerated a bit