Making that scuba video was fun! As long as I’m playing with video I might as well dust off an old one.
The last time I worked with video was six years ago, at the MIT Media Lab, when I cut together Waiting for Goatzilla.
Goatzilla is the autonomous character that offered a proof-of-concept for the predictive learning I developed for my graduate thesis at the lab. By providing virtual characters with the ability to learn about causality in their environments, the new cognitive model let Goatzilla innovate and arrive at new ways to satisfy his drives.
In the clip, captured entirely from the Synthetic Characters behavior system, Goatzilla is performing actions that he has discovered satisfy his drives: hunger, pain avoidance, the desire to dominate other creatures, and so on. When the “mouse pointer of Fate” comes in and cranks up his hunger drive, Goatzilla is stymied. He’d like to eat sheep and there aren’t any left. But his time/rate learning mechanism has taught him that kicking the shed reliably leads to the appearance of sheep. So he kicks the shed anticipating the sheep, out they come, there’s a feeding frenzy, and goatzilla rejoices.
But why did Scott and I create a bizarre sheap-munching beastie instead of something more… conventional? Here I’m going to blame Ireland. We’d been working on an autonomous sheepdog (and had even watched One Man And His Dog to learn the basics) for the opening of the MediaLabEurope project. For that lab’s inauguration, we appeared on Irish national television teaching the Taoiseach (Irish Prime Minister) how to herd virtual sheep. We were sick to death with sheep by that stage and having Goatzilla devour them was… cathartic.