Development of a new serious game utilising reinforcement learning to enrich high school mathematics

Abstract

A serious game is developed in this study to aid South African grade 9 learners with mastering algebraic mathematics. The game is a two-player zero-sum adversarial card game that uses cards to represent different mathematical symbols/objects. The second player is an AI opponent that relies on a machine learning technique known as reinforcement learning to learn how to play the game.

The AI opponent is able to ”learn” basic mathematical concepts such as addition, subtraction, multiplication, division, and functions by playing the game against itself. A comparison between the SAC and PPO reinforcement learning algorithms showed that PPO outperforms SAC in the game. The inferior performance of SAC is attributed to the random elements of the card game that had adverse effects on the entropy-seeking characteristics of SAC.

The serious game was also tested with real learners at a school in Potchefstroom and Ikageng. The majority of learners did not proceed past the first level of the game and only played the game for around a week. The dire performance of the serious game is likely due to the mediocre video tutorials of the game. It is expected that learners’ response to the game can be improved by enhancing the game tutorials, and incorporating the game into classes or extracurricular activities. Adding multiplayer functionality, where players duel each other, can also potentially increase player motivation. This source of motivation stems from the positive peer pressure among students.