Helping robots decide what to do
Leslie Pack Kaelbling works on learning, planning and reasoning for robots at MIT. A robot rarely has all the information it needs. In her account, deciding well includes knowing when a best guess is sufficient and when it is worth gathering more information, such as looking inside something or asking a question.
The robot that stopped, then spun
As a graduate student, Kaelbling tried rewarding a robot for avoiding collisions. It stopped moving. When she added an incentive to keep moving, it spun in place. She uses the example to explain how a system can satisfy the problem it was given without doing what its designer intended.
A route through philosophy
Arriving at Stanford from a small farming town, she found that entering mathematics assumed preparation she had not received. She moved into philosophy, where logic and formal reasoning led back towards mathematics. She describes finding another route, while explicitly saying that her career overall has been enjoyable and has not involved major setbacks.
Learning from different mentors
Kaelbling credits her doctoral adviser with teaching careful, critical thinking and an early faculty mentor with helping her learn how to work as a researcher. For younger listeners, she emphasises finding work they enjoy, whether that turns out to be science, art, literature or something else.