episode
stringlengths
45
100
text
stringlengths
1
528
timestamp_link
stringlengths
56
56
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
but you write in your paper an automated mechanism design
https://karpathy.ai/lexicap/0012-large.html#00:43:27.600
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
that I quickly read.
https://karpathy.ai/lexicap/0012-large.html#00:43:31.440
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
So mechanism design is designing the rules of the game
https://karpathy.ai/lexicap/0012-large.html#00:43:34.000
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
so you get a certain desirable outcome.
https://karpathy.ai/lexicap/0012-large.html#00:43:37.880
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
And you have this work on doing so in an automatic fashion
https://karpathy.ai/lexicap/0012-large.html#00:43:40.200
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
as opposed to fine tuning it.
https://karpathy.ai/lexicap/0012-large.html#00:43:44.920
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
So what have you learned from those efforts?
https://karpathy.ai/lexicap/0012-large.html#00:43:46.720
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
If you look, say, I don't know,
https://karpathy.ai/lexicap/0012-large.html#00:43:50.680
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
at complexes like our political system,
https://karpathy.ai/lexicap/0012-large.html#00:43:52.280
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
can we design our political system
https://karpathy.ai/lexicap/0012-large.html#00:43:56.200
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
to have, in an automated fashion,
https://karpathy.ai/lexicap/0012-large.html#00:43:58.560
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
to have outcomes that we want?
https://karpathy.ai/lexicap/0012-large.html#00:44:01.800
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Can we design something like traffic lights to be smart
https://karpathy.ai/lexicap/0012-large.html#00:44:03.360
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
where it gets outcomes that we want?
https://karpathy.ai/lexicap/0012-large.html#00:44:09.000
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
So what are the lessons that you draw from that work?
https://karpathy.ai/lexicap/0012-large.html#00:44:11.800
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Yeah, so I still very much believe
https://karpathy.ai/lexicap/0012-large.html#00:44:14.840
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
in the automated mechanism design direction.
https://karpathy.ai/lexicap/0012-large.html#00:44:17.240
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Yes.
https://karpathy.ai/lexicap/0012-large.html#00:44:19.400
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
But it's not a panacea.
https://karpathy.ai/lexicap/0012-large.html#00:44:20.840
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
There are impossibility results in mechanism design
https://karpathy.ai/lexicap/0012-large.html#00:44:23.000
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
saying that there is no mechanism that accomplishes
https://karpathy.ai/lexicap/0012-large.html#00:44:26.520
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
objective X in class C.
https://karpathy.ai/lexicap/0012-large.html#00:44:30.240
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
So it's not going to,
https://karpathy.ai/lexicap/0012-large.html#00:44:33.920
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
there's no way using any mechanism design tools,
https://karpathy.ai/lexicap/0012-large.html#00:44:36.120
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
manual or automated,
https://karpathy.ai/lexicap/0012-large.html#00:44:39.000
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
to do certain things in mechanism design.
https://karpathy.ai/lexicap/0012-large.html#00:44:41.000
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Can you describe that again?
https://karpathy.ai/lexicap/0012-large.html#00:44:42.800
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
So meaning it's impossible to achieve that?
https://karpathy.ai/lexicap/0012-large.html#00:44:43.800
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Yeah, yeah.
https://karpathy.ai/lexicap/0012-large.html#00:44:47.480
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
And it's unlikely.
https://karpathy.ai/lexicap/0012-large.html#00:44:48.320
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Impossible.
https://karpathy.ai/lexicap/0012-large.html#00:44:50.440
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Impossible.
https://karpathy.ai/lexicap/0012-large.html#00:44:51.280
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
So these are not statements about human ingenuity
https://karpathy.ai/lexicap/0012-large.html#00:44:52.120
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
who might come up with something smart.
https://karpathy.ai/lexicap/0012-large.html#00:44:55.240
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
These are proofs that if you want to accomplish
https://karpathy.ai/lexicap/0012-large.html#00:44:57.120
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
properties X in class C,
https://karpathy.ai/lexicap/0012-large.html#00:44:59.880
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
that is not doable with any mechanism.
https://karpathy.ai/lexicap/0012-large.html#00:45:02.480
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
The good thing about automated mechanism design
https://karpathy.ai/lexicap/0012-large.html#00:45:04.880
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
is that we're not really designing for a class.
https://karpathy.ai/lexicap/0012-large.html#00:45:07.080
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
We're designing for specific settings at a time.
https://karpathy.ai/lexicap/0012-large.html#00:45:10.840
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
So even if there's an impossibility result
https://karpathy.ai/lexicap/0012-large.html#00:45:14.160
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
for the whole class,
https://karpathy.ai/lexicap/0012-large.html#00:45:16.720
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
it just doesn't mean that all of the cases
https://karpathy.ai/lexicap/0012-large.html#00:45:18.240
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
in the class are impossible.
https://karpathy.ai/lexicap/0012-large.html#00:45:21.360
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
It just means that some of the cases are impossible.
https://karpathy.ai/lexicap/0012-large.html#00:45:22.560
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
So we can actually carve these islands of possibility
https://karpathy.ai/lexicap/0012-large.html#00:45:25.080
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
within these known impossible classes.
https://karpathy.ai/lexicap/0012-large.html#00:45:28.240
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
And we've actually done that.
https://karpathy.ai/lexicap/0012-large.html#00:45:30.920
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
So one of the famous results in mechanism design
https://karpathy.ai/lexicap/0012-large.html#00:45:31.960
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
is the Meyerson Satethweight theorem
https://karpathy.ai/lexicap/0012-large.html#00:45:35.160
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
by Roger Meyerson and Mark Satethweight from 1983.
https://karpathy.ai/lexicap/0012-large.html#00:45:37.360
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
It's an impossibility of efficient trade
https://karpathy.ai/lexicap/0012-large.html#00:45:41.000
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
under imperfect information.
https://karpathy.ai/lexicap/0012-large.html#00:45:43.480
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
We show that you can, in many settings,
https://karpathy.ai/lexicap/0012-large.html#00:45:45.200
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
avoid that and get efficient trade anyway.
https://karpathy.ai/lexicap/0012-large.html#00:45:48.560
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Depending on how they design the game, okay.
https://karpathy.ai/lexicap/0012-large.html#00:45:51.480
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Depending how you design the game.
https://karpathy.ai/lexicap/0012-large.html#00:45:54.160
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
And of course, it doesn't in any way
https://karpathy.ai/lexicap/0012-large.html#00:45:55.880
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
contradict the impossibility result.
https://karpathy.ai/lexicap/0012-large.html#00:46:00.240
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
The impossibility result is still there,
https://karpathy.ai/lexicap/0012-large.html#00:46:01.800
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
but it just finds spots within this impossible class
https://karpathy.ai/lexicap/0012-large.html#00:46:03.920
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
where in those spots, you don't have the impossibility.
https://karpathy.ai/lexicap/0012-large.html#00:46:08.920
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Sorry if I'm going a bit philosophical,
https://karpathy.ai/lexicap/0012-large.html#00:46:12.440
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
but what lessons do you draw towards,
https://karpathy.ai/lexicap/0012-large.html#00:46:14.760
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
like I mentioned, politics or human interaction
https://karpathy.ai/lexicap/0012-large.html#00:46:17.480
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
and designing mechanisms for outside of just
https://karpathy.ai/lexicap/0012-large.html#00:46:20.160
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
these kinds of trading or auctioning
https://karpathy.ai/lexicap/0012-large.html#00:46:24.880
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
or purely formal games or human interaction,
https://karpathy.ai/lexicap/0012-large.html#00:46:26.960
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
like a political system?
https://karpathy.ai/lexicap/0012-large.html#00:46:33.480
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
How, do you think it's applicable to, yeah, politics
https://karpathy.ai/lexicap/0012-large.html#00:46:34.920
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
or to business, to negotiations, these kinds of things,
https://karpathy.ai/lexicap/0012-large.html#00:46:39.160
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
designing rules that have certain outcomes?
https://karpathy.ai/lexicap/0012-large.html#00:46:46.280
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Yeah, yeah, I do think so.
https://karpathy.ai/lexicap/0012-large.html#00:46:49.040
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Have you seen that successfully done?
https://karpathy.ai/lexicap/0012-large.html#00:46:51.360
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
They haven't really, oh, you mean mechanism design
https://karpathy.ai/lexicap/0012-large.html#00:46:54.200
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
or automated mechanism design?
https://karpathy.ai/lexicap/0012-large.html#00:46:56.440
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Automated mechanism design.
https://karpathy.ai/lexicap/0012-large.html#00:46:57.280
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
So mechanism design itself
https://karpathy.ai/lexicap/0012-large.html#00:46:59.000
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
has had fairly limited success so far.
https://karpathy.ai/lexicap/0012-large.html#00:47:01.520
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
There are certain cases,
https://karpathy.ai/lexicap/0012-large.html#00:47:06.440
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
but most of the real world situations
https://karpathy.ai/lexicap/0012-large.html#00:47:07.600
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
are actually not sound from a mechanism design perspective,
https://karpathy.ai/lexicap/0012-large.html#00:47:10.200
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
even in those cases where they've been designed
https://karpathy.ai/lexicap/0012-large.html#00:47:14.680
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
by very knowledgeable mechanism design people,
https://karpathy.ai/lexicap/0012-large.html#00:47:16.920
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
the people are typically just taking some insights
https://karpathy.ai/lexicap/0012-large.html#00:47:20.000
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
from the theory and applying those insights
https://karpathy.ai/lexicap/0012-large.html#00:47:22.760
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
into the real world,
https://karpathy.ai/lexicap/0012-large.html#00:47:25.040
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
rather than applying the mechanisms directly.
https://karpathy.ai/lexicap/0012-large.html#00:47:26.280
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
So one famous example of is the FCC spectrum auctions.
https://karpathy.ai/lexicap/0012-large.html#00:47:29.280
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
So I've also had a small role in that
https://karpathy.ai/lexicap/0012-large.html#00:47:33.520
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
and very good economists have been working,
https://karpathy.ai/lexicap/0012-large.html#00:47:36.880
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
excellent economists have been working on that
https://karpathy.ai/lexicap/0012-large.html#00:47:40.600
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
with no game theory,
https://karpathy.ai/lexicap/0012-large.html#00:47:42.560
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
yet the rules that are designed in practice there,
https://karpathy.ai/lexicap/0012-large.html#00:47:44.040
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
they're such that bidding truthfully
https://karpathy.ai/lexicap/0012-large.html#00:47:47.440
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
is not the best strategy.
https://karpathy.ai/lexicap/0012-large.html#00:47:49.840
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
Usually mechanism design,
https://karpathy.ai/lexicap/0012-large.html#00:47:51.800
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
we try to make things easy for the participants.
https://karpathy.ai/lexicap/0012-large.html#00:47:52.960
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
So telling the truth is the best strategy,
https://karpathy.ai/lexicap/0012-large.html#00:47:56.160
Tuomas Sandholm: Poker and Game Theory | Lex Fridman Podcast #12
but even in those very high stakes auctions
https://karpathy.ai/lexicap/0012-large.html#00:47:58.560