Placeholder Image

Subtitles section Play video

  • OK, so we've been talking about information cascades,

  • and analyzing a model of cascades, where people are imitating

  • each other because they're trying to take

  • into account what these other people know.

  • And we've seen a bunch of things about cascades,

  • so they can start quite easily.

  • They can actually be based on very little genuine information--

  • just a few people who do something-- and as a result,

  • they could actually be wrong.

  • A whole bunch of people doing things, and they're all wrong.

  • And they could be fragile.

  • They could be overturned by the infusion of a little bit of extra information,

  • right?

  • So a nice place to think about all of this

  • is with the Wisdom of Crowds argument.

  • The argument roughly is that crowds are smarter than you would expect.

  • And that argument essentially goes that, as long

  • as people don't start off biased.

  • If you have a bunch of people making independent guesses about something,

  • on average they tend to do pretty well.

  • But what happens with cascades is that that can easily get overturned.

  • Because if people are actually observing others' choices sequentially,

  • then as we've seen, the information gets covered up very fast.

  • And you only get information for a few choices.

  • So sometimes crowds can actually be surprisingly

  • dumb rather than surprisingly smart

  • Right, it's sort of like one of these things where, yeah.

  • In one interesting example to think about

  • is a group of people making a decision.

  • You go around the table and you hear what

  • other people, like, what do David think?

  • What do Jon think?

  • And everyone goes around, says what they think.

  • One aspect we didn't think about in a very simple model

  • is which order you put those people.

  • Like should you put the most informed person

  • first or the less informed person first?

  • It's clear to see that the most informed person first, it's sort of tempting.

  • He knows the most, but he's the easiest to start

  • a cascade, because the next guy, being less informed, you're like,

  • maybe I should listen to that guy.

  • So the order-- the cascade story tells us-- what order might be interesting.

  • It might be Interesting to put the people who are not so mal-informed

  • and get them to express their opinions first.

  • And then have people chime in later on.

  • It's one of these funny things where it's really only once-- in fact,

  • actually once we started teaching cascades-- I

  • thought about this experience.

  • And I mean, anyone who's sat in the room,

  • and you've had a discussion going around the table,

  • and you're like, wow that was more unanimous than I thought, right?

  • And you start to appreciate that, maybe not, maybe people are like,

  • oh, well, I guess I had the other opinion,

  • but I guess everyone else must be right, and then they just agree.

  • One of the things you learn from this is that sometimes it's better to have some

  • mechanism [INAUDIBLE] that people have to express their opinion independently

  • first,--

  • Right.

  • --And then have a way to aggregate and have a discussion.

  • But that initial expression of what they believe

  • can be really useful, in ways that actually-- I

  • have to admit-- I'd never thought about either.

  • Even on simple things like these questions like,

  • we're all sitting here trying to figure out how high is Mount Everest, right.

  • There actually is a right answer-- it's about 29,000 feet high,

  • but the first two people go, I think it's about 18,000 feet.

  • Yeah, that sounds right.

  • And then suddenly, there were people in the room who

  • would have been much better, but they give in to the--

  • and so you only got a little bit of.

  • Whereas if everyone could have like-- it's a heavyweight mechanism--

  • that everyone could write down their guess.

  • You fold it up, and if you average, you could actually do really well.

  • Well that was the original Wisdom of Crowds argument.

  • If you had a bunch of independent guesses,

  • it actually surprisingly did turn out pretty well often.

  • It's just that sometimes it gets used in settings

  • where it actually at this point doesn't apply because of this cascade story.

  • So you see this in lots of settings, so this

  • is kind of a micro setting of people sitting around a table or people all

  • guessing.

  • But then you see it like national politics, right?

  • So if we take the US presidential election system anyway,

  • so the party's nominating people and they have state

  • by state primaries that happen over time.

  • So it's sort of perfect for cascades, because the states move in order

  • and so this Iowan, this New Hampshiran, and some candidate

  • gets off to a good start, it's actually this kind of sequential cascade.

  • It's actually a great example because I'm like, go on around the table,

  • where we can choose to ask the most informed or least informed person.

  • The primary [INAUDIBLE] historical order in which they went, New Hampshire

  • is not particularly a big state, but yet it's going first.

  • So in a way, maybe it shouldn't be so influential.

  • It doesn't have so many votes at the final outcome.

  • But yet it's very influential because it's the first state to go.

  • Right, so it becomes this thing where if somebody is successful the first

  • few steps, the cascade might get going.

  • And of course, the whole system conforms around it, so as a result,

  • candidates lavish a ton of attention on New Hampshire

  • residents and Iowa residents.

  • And so in a way they become almost like a focus group.

  • And so maybe that's almost like a self-healing part of the system

  • where they actually invest a lot more energy in it.

  • But it's all because of this notion that they were at the very beginning.

  • But that example also illustrates the fragility of cascades, too.

  • Because if some new information comes along

  • during what would have been a cascade, it

  • seems to be not that hard to overturn it.

  • And that really could be because the cascade actually

  • isn't based on a lot of information anyhow.

  • With the candidates, it's a little harder to tell,

  • but you think back to the simples and models we've had,

  • and you can often have cascades based on just a couple of observations.

  • And so they're actually pretty easy to overturn with some new information.

  • And so you actually consider this thing where someone gets popular,

  • and everyone's like, why is he so popular?

  • And then just evaporates, just suddenly, and because

  • of a little bit of new information.

  • And, I mean, the funny thing, of course, you think like-- I mean all of these

  • have some [INAUDIBLE] design, like if only we could design

  • the right discussion mechanism or primary system.

  • But some of these are very hard to scrap the whole electoral system

  • and do it again.

  • But some things actually just get designed

  • and they could be designed one way or the other.

  • So like, if I think about how people get invited to events, right?

  • Yes--

  • Or parties, right?

  • --Thinking about this evite business.

  • Yeah, so.

  • It's very interesting.

  • Historically, we used to get an invitation

  • and then you either come or not, and you might call up your friends

  • to check, hey Jon, are you coming?

  • And nowadays, people invite this evite and then it's completely public,

  • and you know who is coming and who isn't coming.

  • It's hard to tell what the effect really is.

  • There's some informational effect, but some direct benefit effect

  • that just simply got more information.

  • There's some way the risk cascade because the first few people

  • turned it down, you say eh, that's getting turned down.

  • But you wouldn't have known if it wasn't so public.

  • On the other hand, there is just more information in the system.

  • You simply know who's coming.

  • That's additional information that you simply

  • know that you didn't use to know.

  • And a party is more fun with lots of people,

  • so there's a direct benefit effect.

  • And I guess also-- if you think about the party--

  • once something is established because of a direct benefit effect,

  • now it's hard to overturn.

  • Now it's not fragile anymore.

  • Because suddenly, well, there really are 50 people going to this party,

  • so it will be a fun party, whatever the original plan was, right?

  • In the same way that-- maybe it's accidental

  • that a social networking site gets a zillion people on it,

  • but they're all there now.

  • It really is more valuable.

  • So it's funny that some of these cascades, two cascades

  • get rolling and one of them, you can just sort of knock down really easily.

  • And the other one is really robust now because it sort of locks in it's gains.

  • No, that's a good point.

  • They really can be very different.

  • There can be popularity cascades that aren't necessarily

  • based on much that's real, particularly not much that's a direct benefit,

  • and those are easy to overturn.

  • But the direct benefit ones actually really

  • are hard, even if whatever is succeeding wasn't

  • in some objective-sense the best alternative.

  • Yes.

  • With the direct benefits out there, it actually is better, in fact,

  • than the other alternative because people are using it

  • or because people are doing it.

  • I mean, I feel like it might be sort of almost a strategic aspect to dislike.

  • If you're the owner of a business that got this popularity cascade going,

  • and you're successful, people might come to you with money

  • and advise you, like, it's time to take some steps to kind of make

  • this advantage permanent.

  • To do some things that amplify the direct benefit part of it.

  • Because then, it won't get.

  • That's true, you want to lock your clientele in.

  • However you got them, you want to lock them in now with some kind of benefit.

  • Might be accidental they showed up, but now you

  • want to create the direct benefit effect.

OK, so we've been talking about information cascades,

Subtitles and keywords

A2 BEG US cascade people information informed benefit direct

INFO2040X mod5 roundtable discussion information cascades v1

  • 439 1
    羅紹桀   posted on 2016/03/02
Keywords

Go back to previous version