filmov
tv
2019 09 19 Stuart Armstrong Research Agenda Online Talk
Показать описание
Humans have a (roughly) shared theory of mind that allows them to model the preferences of other humans from their behaviour. Getting this theory of mind into other agents is highly non-trivial.
Stuart Armstrong shows how this is a consequence of the No Free Lunch result in value learning (you cannot deduce the preferences of a potentially irrational agent by observing its behaviour; and simplicity doesn't help), and sketches out his research agenda for learning human preferences despite this impossibility result.
In short, humans have a (roughly) shared theory of mind that allows them to model the preferences of other humans from their behaviour. Getting this theory of mind into other agents is highly non-trivial.
Stuart Armstrong shows how this is a consequence of the No Free Lunch result in value learning (you cannot deduce the preferences of a potentially irrational agent by observing its behaviour; and simplicity doesn't help), and sketches out his research agenda for learning human preferences despite this impossibility result.
In short, humans have a (roughly) shared theory of mind that allows them to model the preferences of other humans from their behaviour. Getting this theory of mind into other agents is highly non-trivial.
0:00:24
0:00:20
0:00:27
0:00:12
0:00:07
0:01:37
0:01:36
0:14:56
0:18:17
1:06:18
0:11:08
1:13:00
0:01:35
0:15:22
0:02:20
0:54:19
0:15:23
0:31:24
0:04:46
0:02:24
0:00:27
0:03:51
0:01:01