This is a minimum pause time, and the AI might spend more time thinking than this value.
AI Strength
AI Strategy
The expectimax search algorithm, searching to a certain depth
The expectimax search algorithm, searching all options with a probability above a cutoff
The minimax search algorithm, which assumes that tiles are placed adversarially
Weak search algorithm that uses random trials to simulate tile placements
Good strategy that tests the next move by simulating many games from the following position
Weak strategy which always moves up or left
Weak strategy that tries to move left, otherwise up, then right, then left
Very weak strategy that picks random moves
Very weak strategy that moves left, then up, then right, then left
Temporal difference learning with n-tuple network based on this paper.
Note that this requires downloading and unzipping a 28MB model file. Random moves will be played while waiting for the download.
Heuristic
Strong heuristic based on tile values, with higher weights for tiles in the corner
Strong heuristic that penalizes rows/columns that aren't in sorted order
Decent heuristic which compares boards in lexicographical order, going along a wall with penalties for inversions
Decent heuristic which compares boards in lexicographical order, going along a wall
Decent heuristic that compares boards in lexicographical order, going along a wall with a gap
Decent heuristic, similar to Corner Building but with different weights
Weak heuristic which uses an approximation of the in-game score
Weak heuristic which counts number of empty tiles on board
How to play: Pick a strategy for the AI to use.
You can also turn off the AI and use your arrow keys to move the tiles.
Press R to restart.
Click here for page where you can edit the board to test the AI.
AI players written by Stanley Zhong. (Source code for AIs and website)
Game mechanics from the original 2048 game.