Human-like AI for decision making in games (Summer school 2019)

In recent years, there have been a number of significant breakthroughs in game playing AI that have hit the headlines, for example, Deepmind’s Atari AI, AlphaGo and OpenAI five. Some of these breakthroughs have demonstrated that a game playing AI can defeat professional human players, most notably the recent 10-1 defeat of professionals by Deepmind’s AlphaStar on Starcraft 2.

However, although these game playing AIs are exceptionally good at winning, do they actually exhibit human-like behaviour? The goal of human-like AI isn’t necessarily to have an AI that can win a game, but to have an AI that is enjoyable to play alongside or against and would play the game in a similar manner to a human, for example, making similar mistakes, exploring in a similar manner, etc. AlphaStar won the first 10 games against humans playing with beyond-human skill. When it played with human-level capabilities, it lost the final game. 

During the internship there will be opportunities to contribute to the existing research activities in DC Labs through the development of tools, frameworks, algorithms and software for developing, analysing and evaluating human-like behaviour exhibited by AI algorithms.

If you are interested in this opportunity, please contact James Walker or Victoria Hodge for more information.

Required skills

We are looking for two advanced undergraduate, masters or PhD students interested in human-like AI in games.

  • [Essential] Experience of AI and/or machine learning
  • [Essential] Experience of software engineering
  • [Essential] Knowledge of Unity programming environment
  • [Essential] Knowledge of C#

How to apply

For more details on the summer school application process (including eligibility and funding) please see the overview page here.

Supervisor(s)

James Walker
Victoria Hodge