Conclusion & further work
Conclusion
The knowledge in each agent's Kripke model is updated after each step in a turn. Each agent checks whether there are inconsistencies between the current state of the game and his current knowledge. If so, his knowledge is updated and relations to Kripke worlds that are no longer possible are removed. If you run the code, it is clear that it is necessary to implement most of the 2nd and 3rd Depth knowledge to come close to human reasoning. However, humans tend to forget knowledge, which isn't implemented in our model. So the agents in the model do a lot better than humans do in the game.
Further work
We have a few ideas which could be implemented in the game, but we didn't had enough time to implement them ourselves. These suggestions could make the game even more interesting for modelling knowledge and could be a starting point for a new project. Some ideas could be implemented in a hour (we implemented our code in such a way that adding new things is easy), but still some ideas could be a whole project on their own:
- More different strategies for each agent.
- Let the agents use bluffing, by which they sometimes perform irrational behaviour, or use multiple strategies together, to confuse the other agents.
- Let the agents guess which strategy the other agents use.
- Estimating when to 'knock' to end the game. At the moment our game ends when a certain amount of turns per agent have passed. It might be fun to implement the actual game rule which states that the game ends when an agent beliefs he has the lowest card values compared to other agents.
- Adding different kind of 'special' cards, like having the ability to take 3 cards of the closed pile and put one of those cards at your own or another agent's cards, or change the peek card, so agents may peek at an other agent's card too, instead of just their own cards.
- Adding new rules to the game, like always letting the agents swap at the end of their turn.