| Session Name: | AI Summit: Virtual Happiness and How to Measure It: Utility AI Revisited |
| Speaker(s): | Aaron C Gaudette |
| Company Name(s): | aaronsee.media |
| Track / Format: | AI Summit |
Did you know free users get access to 30% of content from the last 2 years?
Get your team full access to the most up to date GDC content
| Overview: | Utility AI has a reputation of being difficult to control compared to alternatives like behavior trees or FSMs. Designing and tuning utility architectures is sometimes perceived as more of an art than a science. What does the value of a utility score really mean? What concrete effect does changing a utility score have? How many "apples" is an orange worth?nnIntuitively, it would make sense to provide utility with a standard unit of measurement. This idea has struggled in the real world, since it is difficult to imagine how one might "measure" happiness. However, it is plausible in a game environment, where designers have perfect information.nnIn this session, a new "unit utility" AI is proposed. This approach challenges incumbent methods and is informed by concepts in statistics, probability theory, and reinforcement learning. |
Recommended
