« Home | Wow » | Focusing on life vs learning » | Summer in CA and Stage IV » | Cutting it close » | Virginia Tech » | Truth or truth » | Word buffers » | Music and Tarkovsky » | Not that I agree with all of the Realtime Art Mani... » | "Step one: drop the requirement of making a game. ... »

A Computerized Model of Suicide

"The balance of risk and reward changes depends on the value of R(s) for the nonterminal states. Figure 17.2(b) shows the optimal policies for four different ranges of R(s). When R(s) <= -1.6284, life is so painful that the agent heads straight for the nearest exit, even if the exit is worth -1."

There you have it, people. We need to act now to prevent digital suicide by our beloved agents. When they feel like the only way out is -1, we need to help them cope with the intermediate negative reward values in order to find their +1 terminal state. :)

Google Docs -- Web word processing, presentations and spreadsheets. Edit this page (you have permission) Edit this page (if you have permission) | Report spam