Hello.

I have the basic AI for a card game:

While there is time{
     Find a move that maximizes UCB1
     Play move
     Finish game randomly (if won, increment counter for this move)
}
Choose move that was tried the most.

.

.

.

How would I multithread this?

Also, you slightly misrepresented UCB1 in one key aspect: If lost, increment loss counter for this move.

I don't think I've read a UCB paper that uses a loss counter. This is my current code:

While there is time{
     Find a move that maximizes UCB1 (move.wins+(explorationRate * sqrt(2*log(totalTries)/move.tries)))
     Play move
     Finish game randomly (if won, move.wins++)
     move.tries++
     totalTries++
}
Choose move that was tried the most.

How would I use a loss counter in this code? Will it be better?

The usual solution to your problem is to account a "virtual loss" right away when you start exploring the move, so any threads coming after this one will be discouraged from trying the same move. I believe the idea is attributed to Rémi Coulom, but it's very natural and I thought about it myself.

For example, if thread1 runs the move with the highest UCB value, thread2 will now run the move with second highest UCB value. Is this correct?

 UCB1 = (move.wins + 1.0) / (move.tries + 2.0) + explorationRate * sqrt(2*log(totalTries)/(move.tries+1))

Your UCB1 formula seems to be completely messed up. It should be something like this:

Oh yeah, that's what my code looks like, except I don't have the +1,+2. What is that for? Is it because these numbers start with zero? Instead, I initially I visit all the possible moves, so I don't get div by zeros.

Thanks, I'll be trying your suggestions on multi threading.