If I knew how to write an AI that were on par with top human warlight players, I would be a millionaire, probably even a billionaire. It's much more complex a task than what you would think.
That aside, writing an AI that can beat other AI's is much easier.
Well, my comment was about Warlight, not about chess, nor about some other random game. Chess and Warlight aren't comparable because Warlight's complexity is exponentially higher than that of chess. This is especially obvious in the endgame of chess, where a computer can often calculate the objectively best move within seconds.
If you want to make software that is on par with top human Warlight players (on all maps and settings, not just a single one), you have the following options: 1) Hard-code every possible board situation into your program, 2) Make the program think holistically, like a human being, and 3) Try to imitate human-like thinking in your program.
1) Is impossible because there are billions over billions of game situations that could arise. Hard-coding all of them is not feasible. And you'd have to do it again for every new map and setting.
2) Nobody in the world has been successful in this yet despite numerous efforts.
3) I won't say it's impossible to imitate human thinking to a degree, but making code that could beat a top player at Warlight by merely imitating would be very difficult, with no guarantee of success. There are numerous challenging taks, such as estimating enemy income and activity, or evaluating random game positions. Also, the human player could exploit knowledge of the AI's style and play in a way that is extremely effective against it. Making the AI to react to that in a sensible way is very difficult.
All in all, my prediction has held well until now. The top player of the AI ladder admits that his current bot could probably not fight on par with even a mediocre WL player. Even on such a simple map as Small Earth. Still, it has no trouble beating other AI's.
