HACKER Q&A
📣 MichaelRazum

How to construct a simple game for humans but difficult for AI?


Assume you want to create a simple game where players can play against each other online for money. BUT they shouldn't loose against bots. Is this possible and what would be the easiest way to do so?

PS: Of course you could think of poker, but that is nearly solved by fb https://ai.facebook.com/blog/pluribus-first-ai-to-beat-pros-in-6-player-poker/


  👤 captn3m0 Accepted Answer ✓
I've been working on finding boardgame-research and this is based mostly on research I've been skimming through (and not finding).

1. Hidden State. Not necessarily by itself, but the less the player knows about what the state transitions look like, the harder it is.

2. Visual interactions. Dixit is a really cool game with a known deck of cards, but I haven't seen any bots tackle it yet. The game involves giving a phrase about a specific card which isn't too specific.

3. Game Complexity. Easy to explode by having card effects, such as in MTG, which becomes Turing complete. If you have cards that change the game state too much, the game state becomes too chaotic to easily account for.

4. Social interactions. Some research exists on Mafia, but most of it is rudimentary and doesn't take into account the hundreds of different characters and emergent character behavior.

5. Games which force you to model your opponenet's strategy. For eg, some games have a mechanic where if you play the same card as your opponent, the action cancels out (actions are picked and revealed simultaneously). Unless you can build a model of your opponent, it becomes hard to win.

6. Random start states (and implement something like the Pie rule to ensure fairness)

Try to stay away from games which have a small state space (Blackjack), or have easy state-transitions, or are perfect information.

[0]: https://github.com/captn3m0/boardgame-research


👤 lumpy28
Backgammon is a very interesting money game. A professional player I used to play with told me that it's 70% luck and 30% strategy.