AI bot for a simple game

Question

There is a relatively simple game with such rules:

There is a safe which needs to be unlocked.

Code to the safe is a 4 digits number without repetitions(1234, 4867, 1092, etc., code like 1231 isn't possible in this game).

The game gives 5 attempts to guess the right code.

Let's say I start a new game and on the first try I test code like 0123.

The game responds with 2-1 . 2 means that code 0123 has 2 right numbers which I need to use in the final unlock code. 1 means that one of those 2 numbers is at the correct position already.

After this I have 4 more exact same steps where I try different codes based on the previous tested numbers and responses from the game.

The goal is to get final code, let's say 9135(based on the prev 0123 try) and response from the game needs to be 4-4(4 right numbers, 4 in place) . The earlier it happens - better.

I know that this can be solved using combinatorics just by excluding some combinations but I don't know how to choose the most weighted combination for the next try and hope AI can do it better.

I'm a frontend developer and an absolute beginner in AI. I don't really understand how complex code will be to solve this problem and what effort it requires. I will really appreciate if you can explain to me and share some links/code examples(lang doesn't matter but would be good if it is JS or Python) of similar solved tasks, so I can solve my problem based on this.

Feel free to tell me if my explanation wasn't clear, I will try more simple words then:) Thanks!

Answer 1

Your game sounds similar to Mastermind , only with numbers instead of colored pegs.

Googling "Mastermind AI" leads to eg this implementation using a genetic algorithm to solve Mastermind, which you could probably look at for inspiration.

Answer 2

While @AKX is correct that this is a variant of Mastermind, a genetic algorithm might not be the first place to look, as this is probably more complex that simpler approaches.

Donald Knuth is famous (among many other things) for working out a solution to the game. There is a good overview of this approach on the Puzzling Stack Exchange site, and if you look at the other answers on that question, there is also a discussion of how to code the solution.

In your case, the simple approach is to write a function that iterates from 0000 to 9999. These are all potential answers. But, when you iterate through the numbers you want to remove (1) all numbers with duplicate digits and (2) all numbers that are inconsistent with the guesses so far. Any other numbers can be put in an array or list storing potential answers. From these remaining numbers, you can just guess any number and then continue the process.

A more complicated approach would be to make the next guess using an algorithm similar to ID3 to try to find the guess that maximizes the information gain you get from the response. But, given how much information you get from each guess, this is unlikely to be needed.