Can my homemade A.I. beat me at Warhammer 40k?

TL;DR

Learn how the creator used machine learning techniques to teach a computer how to play the tabletop game Warhammer 40,000 Kill Team.

Transcript

thanks to curiositystream and nebula for sponsoring this video stick around to the end to get more than 40 off warhammer is a game that needs at least two people to play and this year that's been a bit difficult covid has put a complete dampener on my tabletop gaming activities this year and yes you can do stuff through tabletop simulator but it's ... Read More

Key Insights

• 🎮 Training a computer to play a tabletop game like Warhammer Kill Team involves converting the game's rules into a form that the computer can understand and interpret.
• 👾 Machine learning algorithms, such as reinforcement learning and genetic algorithms, can help improve the computer's decision-making abilities in the game.
• 🧡 The computer considers various inputs, such as distance, skills, and weapon range, to determine the probability of each action.
• 💄 The computer's decision-making process relies on matrices and vectors, with probabilities determined by the values in these matrices.
• 💯 The creator made simplifications in the code to focus on the core mechanics of Kill Team and train the computer more effectively.
• 👾 The computer's performance in the game depends on the training it has undergone, with iterations and improvements over multiple generations.
• 🎰 The creator received advice from a machine learning expert, Jordan Harrod, to enhance the effectiveness of the model.

Questions & Answers

Q: How did the creator convert the rules of Kill Team into a form that the computer can understand?

The creator wrote Python code to represent the rules of Kill Team, allowing the computer to read and interpret the game's mechanics.

Q: What factors does the computer consider when making decisions in the game?

The computer takes several inputs into account, such as the model's distance from the target, its ballistic skill, weapon range, and number of attacks. These inputs are used to generate a probability for each possible action.

Q: How does the computer decide which action to take?

The computer ranks all possible actions based on their probabilities and generates a random number. If the random number is less than the probability of the action, the computer takes that action. This process is repeated until an action is taken or all possibilities are exhausted.

Q: How did the creator train the computer to play the game more effectively?

The creator used reinforcement learning and a genetic algorithm. The computer was trained to find a policy that maximizes the reward in the game. The genetic algorithm involved breeding models that performed well and adding permutations to explore different areas of the parameter space.

Summary & Key Takeaways

• The creator used machine learning techniques to teach a computer how to play Warhammer 40,000 Kill Team, a small-scale skirmish version of the tabletop game.

• The rules of Kill Team were converted into Python code to enable the computer to understand and play the game.

• The computer makes decisions based on inputs and probabilities generated by machine learning algorithms, simulating battles and ranking actions from most to least likely.