Steve Rabin, Michael Dawe, Brett Laming, Robert Zubek, John Funge
I got a few notes from this, a session more on what to teach to people to learn AI – from and to educators really.
This panel for what to teach students, and also get the industry to have a coherent vision. An example syllabus from John is to build up a game AI NPC from scratch, going through the overview, input and outputs of the AI – perception and movement, reactions, memory, planning and learning. Get to pathfinding pretty late compared to normal AI courses.
An alternative from Steve is to work on AI Architecture, then movement, pathfinding, agents and animation, tactics and strategy then learning.
From Robert is lessons from robotics is a great thing to learn. A great correspondence between both fields, for autonomous moving robots that is (not factory ones).
Michael comments that Steve’s course is pretty comprehensive but you can’t learn everything in the realm of AI in one course, it can take years to learn.
Question about getting students off machine learning – answered by John by encouraging them to try them, but consider the analogy with a master craftsman – you can’t be given a chisel and hammer and just immediately make great crafts, it takes major time to do it. Michael says they are not entirely useless – such as in racing games. Students should learn when they are appropriate and when they are not.
Question on what to use to teach the basics – Steve says a game engine is too complicated (and expensive to licence) – and he uses a DirectX demo sample – Multi Animation. Brett adds understanding the integration of it into game engine theory is a good idea.