Logo

ICS Fall Semester 2022 Colloquia Zoom Meeting - Shared screen with speaker view
Alba Tuninetti
17:44
Hi all, thanks for joining over Zoom! If you have any questions for Chris, please feel free to pop them in here and I can pass them onto him. He’ll be answering questions at the end of his talk. Thanks!
Althea Wallop
01:08:34
How much tech support do the teachers at summer school tend to need?
Althea Wallop
01:08:56
Do you see any patterns in the type of teachers who need more or less support?
Lane
01:13:40
One clear advantage of teachable AI is that it can explain its actions in terms of sensible, interpretable operations on a sensible, interpretable state space. How do you envision those state spaces and operation sets being built in more complex environments, like open-world video games or even the real world? Will they be engineered by humans, or will the AI be able to create and/or update its ability to decompose "raw" world data into abstract state spaces and operation sets?
Roman Khamov
01:15:40
You mentioned that at one point you passed data to the model visually via pixels.What's the current thinking on the effects of how a model receives data?A human receives a torrent of visual and auditory data and has to choose which data to attend to.Do models also have to show such discretion, or are they fed selective data?I would imagine the data we choose to show a model has a significant impact.
Martha Palmer (Colorado)
01:16:48
Great talk. Herb Simon is cheering from the grave! My first question was going to be about comparing this to RL, so thanks for addressing that. I’m still not clear on where the strategies come from that the agent tries out for solving the problems. Could you please say a little more about that?
Donna J Caccamise
01:17:47
+1
Jim Kukla
01:20:42
+1
Samuel Friedman (RedShred, he/him)
01:22:31
Really enjoyed the talk. Have you looked at multi-agent learning / teachable AI?
Samuel Friedman (RedShred, he/him)
01:44:33
Thank you
Jim Kukla (RedShred, he/him)
01:44:49
Thank you!
Alba Tuninetti
01:44:51
Thanks everyone!
Lane
01:44:57
Thanks, Chris!