Michael Mateas: Appreciating the complexity of AI development

Center Director Michael Mateas was our speaker today during our IFOG: Speaker Series, and his take on the "Photoshop of AI" debate was fascinating!

In 2008, Chris Hecker posited that game developers needed a "Photoshop of AI" equivalent in order to compete with the strides made in graphics development over the last few decades. Michael posited that while a grand idea, it's not feasible, as it's based on the assumption that the problem of representing 3D space is comparable to the problem of creating AI in games. But it's not.

There is only one goal in tackling the 3D representation problem - does it look like a 3D representation? AI is far more complex, and encompasses a number of different fields: character sensory, physics of movement, conversation, character learning, pacing and dynamic difficulty adjustment, modeling of city dynamics, and the modeling of common sense reasoning, to name a few. No one system can address this.

Graphics are the statics of what you see, but AI is gameplay - everything else! AI is conditional, and therefore much too complex to expect an equivalent instant feedback program like Photoshop to be an option. But Michael does agree with Chris in one area - AI *must* be about authoring. And creating leverage (where quality and variability is greater than the effort put in) is an author-centric way to focus on the future of AI.

And make sure to check out the Senster, an amazing example of the aesthetics of behavior - built in 1970!