Quote from: IchoTolot on April 10, 2020, 05:09:24 PM
Yes, these are learning algorithms that improve with every match they played until they could compete.

Couldn't we just use those algorithms here, too?

Quote from: IchoTolot on April 10, 2020, 05:09:24 PM
The difference here is there is no statistic that shows how good an attempt is in lemmings. In DOTA you got the objectives, kills, positioning, items....

What about time spent in game, amount of lemmings saved, amount of skills used?

Quote from: IchoTolot on April 10, 2020, 05:09:24 PM
There is no clear optimal way in lemmings compared to these games where you can make out certain variables (there is a clear matagame)

I'm not sure whether I understand what you mean by this, but I think every level in lemmings has a theoretical optimal solution and I think A.I. is able to find that solution one day in the very near future.

Let's keep our hopes high!

A very interesting concept!!!

I have not much experience with A.I. terminology at all, but I've heard there exists A.I. (for example OpenAI?) that win complex games like DOTA2, GO and Starcraft against pro players... These seem like very very very complex games to me, many times more complex than a lemmix game, so I'm curious to see a solution!