What I was more interested in was the first third of his talk, where he discussed some observations about the trends of video game development. He observed that in his career as a game developer his teams grew as he became successful, but also that with each new team the ratio of artists to programmers grew.
In part he pointed at the invention of the CD-ROM, which allowed games to include so much data that people couldn't even create enough junk to fill the CD. He extrapolated the trend of growing game development team sizes, claiming that in 2050 a Grand Theft Auto equivalent would take, y'know, a million people and 250 million dollars to build.
(I think more recently there are opposite trends in play too, though: see the success of tiny Flash-based games, which are often much more creative and interesting than their commercial counterparts.)
In completing the Sims 2, the art team had outgrown the development team, and Spore was even larger in scope. So for Spore they've put a lot of emphasis two escape paths: the first is procedural generation, from game worlds and player models to animations. The second is user-generated content. The game pushes the data you create up to a central server and pulls down data from other players. He offhandedly mentioned using collaborative filtering to have the game tune itself to your play style -- if you preferred cute characters it ought to find other cute characters, or if you were playing a peaceful trade-oriented game it'd pull parameters for other peaceful trade-oriented civilizations.
Both of those have a lot of potential, but I am skeptical they'll be able to pull them off successfully. I've only seen one or two instances of collaborative filtering that was actually useful. And randomly generated content often feels, well, random, and empty in a way that is hard to pin down. At the same time, there are instances of random processes generating data that is still interesting to explore, such as the game of life or fractals. I wish them luck.
(Google people can view the talk by searching for "Spore" in the talks database.)