Finding Wordle Strategies That Can be Mastered by Humans
DOI:
https://doi.org/10.54097/hset.v39i.6633Keywords:
Wordle; Puzzle Games; Reinforcement Learning; Q-learning; Human-computer Interaction.Abstract
Wordle is a popular web game. Players use strategies and hints to figure out the correct answer of a daily puzzle. In general, there are several strategies to solve it efficiently, but which is the best depends on the game’s current states. This paper designed 4 individual strategies which is learnable for both computers and human players, and trained an AI based on Q-learning to solve the game automatically with a win rate of over 80%. The decision this AI made is only based on the word list, not including other information (e.g., frequency of appearance on the Internet). Based on the analysis, it can give players advices in various kinds of situations and offer useful combos. By using this AI players, it is feasible to figure out what is the best choice and improve their skills. However, it still cannot solve some rare situations efficiently, which may require more training or a new way of setting the game states.
Downloads
References
Heinrich, J., Silver, D.: Deep reinforcement learning from self-play in imperfect-information games. arXiv preprint arXiv:1603.01121 (2016).
Mnih, V., Kavukcuoglu, K., Silver, D., Graves, A., Antonoglou, I., Wierstra, D., Riedmiller, M.: Playing atari with deep reinforcement learning. arXiv preprint arXiv:1312.5602 (2013).
Brown, N., Bakhtin, A., Lerer, A., Gong, Q.: Combining deep reinforcement learning and search for imperfect-information games. Advances in Neural Information Processing Systems, 33, 17057-17069 (2020).
Rikters, M., Reinsone, S.: How Masterly Are People at Playing with Their Vocabulary? Analysis of the Wordle Game for Latvian. arXiv preprint arXiv:2210.01508 (2022).
A Wordle Hack, 2022, Retrieved from: https://thamara.blog/a-wordle-hack-aa83912cd979.
de Silva, N.: Selecting Seed Words for Wordle using Character Statistics. arXiv preprint arXiv: 2202. 03457 (2022).
Anderson, B. J., Meyer, J. G.: Finding the optimal human strategy for wordle using maximum correct letter probabilities and reinforcement learning. arXiv preprint arXiv:2202.00557 (2022).
Omidshafiei, S., Papadimitriou, C., Piliouras, G., et al.: α-rank: multi-agent evaluation by evolution. Scientific reports, 9(1), 1-29 (2019).
Liu, C. L.: Using Wordle for Learning to Design and Compare Strategies. arXiv preprint arXiv: 2205. 11225 (2022).
Bonthron, M.: Rank One Approximation as a Strategy for Wordle. arXiv preprint arXiv:2204.06324 (2022).
Downloads
Published
Issue
Section
License
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
