Behavior in Social Networks: Information Cascades

Summary

This lecture delves into the concept of information cascades, exploring when it's rational to follow the crowd's decisions despite personal information. It highlights the 'wisdom of the crowd' and the impact of social influence on decision-making, where individual choices can be swayed by others' actions, leading to cascades that may not always reflect rational thought. The discussion also touches on the potential manipulation of public opinion through viral content and the importance of recognizing these influences in our increasingly interconnected world.

Takeaways

• 😀 The lecture emphasizes the importance of understanding when it makes sense to conform to the crowd's decisions, highlighting the concept of the 'wisdom of the crowd'.
• 🤔 It discusses the rational decision-making process, including the use of Bayes' rule, to determine when an information cascade occurs, even in a purely rational environment.
• 📊 The script mentions that an information cascade can begin when the difference between the number of acceptances and rejections reaches two, indicating a shift in decision-making dynamics.
• 🧐 The importance of signal reliability is underscored, as it influences the decision-making process, especially when one's own information is uncertain.
• 📚 The concept of a 'prior probability' is introduced, representing initial uncertainty and expectations about an outcome, such as the quality of a new restaurant.
• 🍽️ An example is given about evaluating a new restaurant based on the menu descriptions, illustrating how signals (like menu quality) can affect the perceived likelihood of the restaurant being good.
• 📉 The script points out that people do not always make decisions rationally, and emotions and social influences play a significant role in decision-making, especially when personal information is unreliable.
• 🌐 It is noted that decision-makers exist within networks and may not receive all public signals, which can lead to 'filter bubbles' and isolated cascades within these clusters.
• 🌪️ The potential for information cascades to be manipulated, such as through viral news or memes, is acknowledged, with examples of how this can influence public opinion.
• 🚨 The script warns about the propagandist toolkit, which can be used to start or stop information cascades by attacking the reliability of public signals.
• 🌐 Finally, it is emphasized that information cascades are a significant aspect of social influence and decision-making, with real-world implications in areas like politics and public health.

Q & A

What is the main topic of the lecture?

The main topic of the lecture is information cascades, particularly focusing on when it makes sense to go along with the crowd and the factors that cause or prevent such cascades.

What is the 'wisdom of the crowd' concept mentioned in the lecture?

The 'wisdom of the crowd' concept refers to the idea that the collective judgment of a group of people can be more accurate than that of an individual, especially when the judgments are independent.

How does the concept of 'Bayes rule' relate to the discussion of information cascades?

Bayes rule is used to illustrate the idea of rational decision-making in the context of information cascades, showing how an individual might update their beliefs based on new information.

What is the significance of the prior probability in the lecture?

The prior probability represents the initial belief or uncertainty about an event before new information is received. It interacts with the payoff to determine whether an individual should be indifferent to the new information.

What does the lecture suggest about the rationality of decision-making?

The lecture suggests that even in a purely rational decision-making environment, information cascades can occur, indicating that people's decisions are not always rational and can be influenced by social factors.

What is the role of signals in the context of information cascades?

Signals are pieces of information that individuals gather to make decisions. The reliability of these signals, represented by their conditional probability, can influence whether an individual decides to go along with the crowd.

How does the lecture describe the conditions for an information cascade to begin?

The lecture describes that an information cascade begins when the difference between the number of acceptances and rejections reaches two, indicating a tipping point in decision-making.

What is the impact of emotions and social decision-making on information cascades?

Emotions and social decision-making can lead to information cascades when individuals rely on others' decisions, especially when their own information is uncertain or not fully reliable.

How can public signals influence the opinions of a large number of people?

Public signals, such as viral news or memes, can create a systematic leaning one way or another, influencing the opinions of many people through social influence.

What is the role of network effects in the context of the lecture?

Network effects are mentioned as a benefit of going along with the crowd, where individuals gain a direct benefit from others doing the same thing, although the main focus is on information cascades.

How can clusters in a network, like filter bubbles, affect information cascades?

Clusters in a network, such as filter bubbles on social media, can block information cascades from bridging across different groups, leading to isolated cascades within these clusters.

What is the propagandist toolkit mentioned in the lecture?

The propagandist toolkit refers to strategies used to manipulate information cascades, such as starting or stopping them, by influencing the public signals that people rely on to make decisions.

Outlines

00:00

📚 Understanding Information Cascades and Rational Decision-Making

The lecture delves into the concept of information cascades, emphasizing the rational decision-making process behind going along with the crowd. It discusses the 'wisdom of the crowd' and the idea that independent judgments can lead to a more accurate consensus. The speaker explains the use of Bayes' rule in a rational environment to calculate the probability of an option being good based on a high signal, highlighting the importance of the prior probability and the conditional probability of signals. The lecture also touches on the emotional and social aspects of decision-making, where personal information can be influenced by others' signals, potentially leading to information cascades.

05:01

🔍 Analyzing the Reliability of Signals in Decision-Making

This paragraph explores the reliability of signals in the decision-making process, using the example of evaluating a restaurant's menu to determine its quality. It explains how signals, such as menu descriptions, can provide a conditional probability that indicates the likelihood of the restaurant being good or bad. The speaker simplifies the scenario by assuming a 50-50 chance of the restaurant being good, which simplifies the calculation of the probability using Bayes' rule. The paragraph also discusses the irrational aspects of decision-making, where people do not always calculate probabilities but are influenced by emotions and social cues, which can lead to information cascades.

10:04

🌐 The Impact of Social Influence and Network Effects on Decisions

The final paragraph examines the impact of social influence and network effects on decision-making, particularly in the context of information cascades. It discusses how public signals, such as online reviews or comments, can be manipulated to sway opinions on a large scale. The speaker also addresses the concept of filter bubbles and how they can create isolated cascades within clusters of the network. The paragraph concludes with a discussion on the potential for propagandists to start or stop information cascades, using the example of a viral video that was later revealed to have a different context, emphasizing the importance of understanding the dynamics of information cascades in decision-making.

Keywords

💡Information Cascade

An information cascade occurs when individuals make decisions based on the actions of those before them, rather than their own information. In the video, it's discussed in the context of decision-making influenced by social signals, such as the adoption of wearing masks during a pandemic. The script mentions that even in a rational decision-making environment, cascades can happen, emphasizing the power of social influence.

💡Wisdom of the Crowd

The 'wisdom of the crowd' refers to the collective intelligence that emerges from the collaboration and competition of many individuals. The video script uses this concept to explain how independent judgments can lead to more accurate outcomes, such as estimating the number of marbles in a bottle. It is contrasted with the influence of social signals, which can lead to cascades rather than independent decision-making.

💡Rational Decision Making

Rational decision making is the process of making choices based on logical reasoning and the best available information. The script discusses the use of Bayes' rule as an example of rational decision making, where the probability of an event is updated based on new evidence. However, it also points out that people often do not make decisions this way, especially in social settings.

💡Bayes' Rule

Bayes' Rule is a fundamental theorem in probability that describes how to update the probabilities of hypotheses when given evidence. In the script, it's used to illustrate how rational decision-making should account for new information. The example given is about evaluating a restaurant's quality based on the menu descriptions, showing how prior beliefs are updated with new signals.

💡Signal

In the context of the video, a signal is a piece of information that an individual uses to make a decision. The script discusses how the reliability of signals, such as menu descriptions, can influence the decision to try a new restaurant. It highlights the importance of conditional probability in assessing the value of a signal.

💡Prior Probability

Prior probability represents the initial belief or expectation about an event before new evidence is considered. The video script uses the example of choosing a restaurant, where the prior probability is the initial assumption that it is as likely to be good as it is to be bad, highlighting the role of prior beliefs in decision-making.

💡Network Effects

Network effects occur when the value of a product or service increases with the number of others using it. The script briefly mentions network effects as a benefit of going along with the crowd, suggesting that in some cases, there is a direct benefit from others doing the same thing as you, which is a separate concept from information cascades.

💡Filter Bubble

A filter bubble refers to the phenomenon where online platforms create a personalized experience that isolates users from information that disagrees with their viewpoints. The script discusses how filter bubbles can create clusters within networks that block information cascades from bridging across different groups, as seen with differing views on mask-wearing.

💡Social Influence

Social influence is the process by which an individual's thoughts, feelings, or behaviors are affected by the actions or beliefs of others. The video script addresses social influence in the context of elections and the spread of information, noting how it can be used to manipulate opinions and create divisions.

💡Public Signals

Public signals are indicators or cues that are visible to a broad audience and can influence their opinions or behaviors. The script discusses the potential for public signals, such as online reviews or comments, to sway the opinions of many people, either positively or negatively.

Highlights

The importance of understanding when to follow the crowd and the concept of 'hurting' in decision-making.

The potential benefits of going along with the crowd, such as gaining information from others when your own is uncertain.

The concept of 'wisdom of the crowd' and how independent judgments can lead to a more accurate average.

The role of Bayes' rule in rational decision-making and the occurrence of information cascades even in rational environments.

The impact of prior probability on decision-making, exemplified by the 50/50 chance of a new restaurant being good.

The significance of signals and their conditional probability in determining the reliability of gathered information.

The mathematical approach to decision-making using Bayes' rule and the calculation of the probability of an option being good.

The irrational nature of human decision-making and the contrast with the overly rational mathematical models.

The influence of emotions and social decision-making on individual choices, especially when personal information is uncertain.

The conditions that lead to the start of an information cascade and the mathematical representation of this phenomenon.

The impact of public signals and the potential for manipulation in social influence, such as in elections or viral news.

The role of network effects and filter bubbles in shaping individual decisions and the potential for isolated cascades.

The potential for propagandists to start or stop information cascades by manipulating public signals.

The real-world example of a viral video and the subsequent attack on the reliability of the information presented.

The importance of recognizing information cascades in politics and their impact on public opinion.

The conclusion of the chapter emphasizing the significance of understanding information cascades in various contexts.

Transcripts

all right this is gonna be the lecture

on chapter 16 for this lecture in

particular I really think the exercises

brought most of the points home so if

you've done the exercises and reviewed

those solutions you will hopefully have

an idea of sort of what causes what

prevents an information cascade here I

just want to provide some really like

the solid insights of what they're

really trying to do and just emphasizing

a few strong points that we need to get

keep in mind all right number one strong

point here the idea of hurting when does

it make sense to go along with the crowd

for the most part in the chapter it

talks about a lot of bad things and it

just at least briefly mentions the idea

of good things you can get from going

along with what other people are doing

we will talk in the next chapter about

network effects that is you get a direct

benefit of other people doing exactly

the same thing as you in this chapter

though it's about information so you can

get information from other people

because you feel that maybe your

information isn't quite up to par and

that then will aid you and so this is a

I want to emphasize more and more as we

go through this this is definitely

rational decision and we're trying to

sort of get an idea of why does it make

sense or not to go along with the crowd

alright so the main idea of going along

with the crowd is the idea of what they

call the wisdom of the crowd and so if

we're talking about you know if we're

looking at how many marbles are in a

bottle or how much this cow weighs then

you go and you talk to a thousand people

if the judgments are independent you

will get closer to the the average of

those judgments will be closer to the

actual amount because you have sort of a

triangulation if they're independent if

they're not independent that is

decisions are being made in sequence or

for other reasons of social influencing

that are actually outside the scope of

this chapter then that causes other

problems

so what we're gonna do for the most part

here is say this is a rational decision

okay the reason in the chapter it even

has Bayes rule on there is to say in a

purely purely purely rational

decision-making environment an

information cascade occurs so in the

book when it talks about a prior

probability this basically represents

it's somewhat your uncertainty but it's

the idea that for example you know a new

restaurant opens if you think one out of

every two restaurants selected so half

of the restaurants are good and half of

them are bad that means randomly

speaking when you select a restaurant

you expect you know 50% chance that it's

going to be good given this idea that I

think hey there's a 50/50 chance this

new restaurants good my payoff interacts

with the prior probability and what

should happen here is you're somewhat

indifferent to what's going on and that

means you know if I think okay there's

50/50 chance the new restaurants good I

should be able to say more or less you

know randomly picking it's indifferent

to me based on the rewards I get

choosing a good or a bad restaurant so

if I think just for example if I really

have a bad time at a bad restaurant okay

so it's just awful and you know that's

terrible then if there's a 50/50 chance

that the new restaurants good I'm not

gonna go because it's the the bad thing

that happens from a bad restaurant is so

terrible an alternative is the same

right I know you know too bad restaurant

I mean the food kind of sucks but it's

you know is it really the worst thing

you've ever had happen where a new

restaurant that's good is like woohoo I

get to tell all my friends this awesome

restaurant so the idea in the book is

that you're initially indifferent it

actually yeah we'll get to that in a

second so the payoff combines you should

be initially more or less

and different according to what the book

says we'll talk about you don't really

need to be indifferent the next issue is

the signals and that is uni or you're

going to gather information about this

in this case a new restaurant and you're

trying to gather information that would

say whether or not accepting or trying

this new restaurant would be good or bad

so in the book when they say you know

assume that good is true what they're

saying is going to the restaurant is a

good idea

so each signal so the signals you gather

have a conditional probability and that

reflects how reliable it is so the

signal here might be I might look at the

menu and I read the little descriptions

of each dish and looking at the

descriptions I would say oh like these

descriptions are awesome and given these

great descriptions there's a 75% chance

that a good restaurant would write their

descriptions this way and only 25%

chance that a bad restaurant would write

their descriptions this way so that's

the sort of reliability of the signals

that you look at okay now one of the

things about the the book sort of makes

it they make a simplification that makes

it a little kind of weird and that is if

we go back up and we said there's a

50-50 chance of the new restaurant being

good that 50% that's the P and then

because it's 50/50 that means P equals

one minus P so more or less P isn't a

factor at all the 75% chance down here

below that's the Q and the Q is really

why you know it ends up being like hey

there's a 75% chance that it's a good

restaurant so again the whole point of

using Bayes rule is to show rational

decision making so we're popping up

Bayes rule here if you'll notice here

we're saying you choose the option if

the probability that the option is good

given a high signal is greater than the

initially estimated probability that the

option was good

so remember we

get the menu things and we saw there is

a 75% probability that a good restaurant

would have menu options this good and

only a 25% probability that a bad

restaurant would have many descriptions

this good if we look at this formula the

PRG given H we have to calculate that

the probability that it's good given

that we got a high signal it's also

based on the idea that you know we got a

high signal and we said we had a 75%

chance of getting a high signal if it's

a good restaurant that's PR bracket H

given G and then there was the 25%

chance of a high signal that is bad that

was the PR H given B so we're

calculating the probability that it's

good given that we got a high signal so

the probability this good is based on

you know that 50-50 chance of just

selecting it randomly and then the Q is

that's the probability of it getting a

high signal given it's good so that's

our 0.75 percent and if we calculated it

out here we would get at 0.75% but like

I said that squeezed right now P equals

1 minus P ok now super emphasize like we

went through this I want you to

emphasize 100 percent I'm not going to

make you calculate this because the

reality of it is people don't make these

decisions rationally but I want you to

know that you know there is a way to

describe this using mathematics but you

know frankly you're just pulling the

answers out of thin air right you know

really reading a menu and you're gonna

say I have a 75% chance that it's a good

restaurant would have a menu like this

so that's that's way too rational and

people don't make decisions that

rationally so what the bottom line here

is this is emotions ok and social

decision making situations mean when I'm

going to make a decision I have my own

information if there's any uncertainty

about my own information

other people's information can influence

my decision so anytime my own

information is a hunt isn't a hundred

percent reliable the other people's

signals are going to have an influence

on my choice and that's where we get

really these information cascades I mean

it's not all this red balls and blue

balls business all right

so information cascades the idea again

if you're a hundred percent rational

they could still have a cascade and the

basic idea with the book by formula is

the cascade begins when the difference

between the number of acceptances and

rejections reaches two and you can see

where you know mathematically as soon as

we hit going above a line you know

that's gonna happen that happens in that

super rational draw a red ball out or

draw a blue ball out and what do you

think is in the urn actual

decision-making is of course super

Messier but you can get to the same

results and that's why you know if I go

to a certain store in a certain

neighborhood here everyone is going to

be wearing a mask if I go to a different

store or at a store in a different area

maybe even a different kind of store

like almost no one will be wearing a

mask so you know if we're talking about

the adopt not adopt decision of you know

is it good to wear a mask during a

pandemic it's super messy and people are

definitely relying on other people to do

it but again that's an emotional thing

but there's still an information cascade

there so so is it bad or good hopefully

we remember now that we had all these

problems in like 2016 and they haven't

really stopped with like Russian trolls

trying to influence elections and trying

to like cause sort of divisions in the

left wings of many different countries

so social influence you know is it bad

or good we are people who know what

happens here and we

can do things like you know hire a

thousand people to post reviews online

for a product we can hire a thousand

people to post comments on a news story

and it's possible with those public

signals to influence you know the

opinions of thousands and thousands of

other people so the public signals we

can make them lean strongly one of your

book examples or the exercise examples

was the non adopt cascade make sure you

are aware of when those things happens

when we have viral news or sort of viral

memes that's a systematic leaning of

strongly we'll talk more about chapter

18 just actually yeah when that happens

so remember there's good and bad things

the next thing you got to remember is

when you are a decision maker you are

literally in a network so you're not

getting signals from all over the place

you may not get all of the public

signals so especially on Facebook we

talk about people's filter bubble you

know that's that's what we're talking

about here so clusters in the network

that is these filter bubbles can block a

cascade from sort of bridging across or

within one of these bubbles you might

experience an isolated cascade and you

know we can see that in the United

States right now with people mainly

Trump supporters that don't want to wear

masks and the sort of other will some

other different people think masks are a

good idea and one thing you're also

going to get this is pure like the

propagandist toolkit is this idea of

trying to make information cascade start

or stop and in the example I have here

is where there was a viral video this is

a while back where a police officer

holds teens at gunpoint on their knees

during a snowball fight and the police

are like hey that's not the whole story

and you know interestingly the if you

read the actual what the police story

was it really was they

held some teens at gunpoint had a

snowball fight but it was that you know

they were having a snowball fight and

one of the teens took off and then it

was you know there were two cops there

one of the teens took off one of the

cops chased that teen and the other cop

like I don't know didn't want the other

ones to run away or something but so

there was a big attack in the right-wing

media about how this was a justified

thing that they stopped and held these

kids at gunpoint and really it was an

attack on the reliability so it was like

you saw this viral video but what you

saw isn't really what happened so

there's a lot going on you know this is

really a big time politics but keep in

mind that information cascades is really

a big deal all right so that's it for

this chapter and you'll hear from me

next time