After Jesse’s introduction he gave a presentation about long term risk from artificial intelligence. We learned that there have been many advancements in AI but no systems have been developed to radically change the real world yet.
Some risks he talked about where the fairness and reduction of bias in AI and the case where AI causes catastrophic outcomes because it cares about the wrong things. We learned that it may be hard to design AI to have goals aligned exactly with the creator’s goal and that the AI may do harmful things in order to maximize efficiency to achieve its goals. An example of the first case we saw was in a race game simulation the goal was to collect the most coins with the intent of having the AI be fastest in completing the race to get the coins at the end, however it ended up circling over respawning coins in the middle of the race the whole time. It was very interesting to see the problems AI posed in simulations to better examine the risk of more powerful AI.