Induction of Social Contagion

Key Question Being Explored in This Insight:

What is social contagion? How does friendship-nomination targeting enhance the spread of health education and behaviors in social networks?

View the page of: Nicholas A. Christakis

Institution: Yale University

Explore more insights in:

Leading Figures Medicine Professional Public Health Qualitative Health

Media Literacy Toolkit

Faculti's Media Literacy Toolkit helps viewers critically engage with academic insights by analyzing the research context, identifying perspectives, and encouraging thoughtful evaluation.

Critical Questions to Consider ▼

What assumptions does the research make?
Are there alternative perspectives not explored?
What are the limitations of the research method?

Bias and Perspective Awareness ▼

This research comes from Yale University . Reflect on how the institution's academic focus and research partnerships may shape the questions being explored.

Further Reading ▼

Read the Study

What is social contagion? How does friendship-nomination targeting enhance the spread of health education and behaviors in social networks?

Image courtesy of interviewee. December 18, 2024

Log-in or Sign-up to Faculti
Currently viewing this subject insight as a guest. You have insight(s) remaining for this month. Login to view 8000+ figures on the platform.

Generating Workspace...

Export as PDF

This video presents a comprehensive discussion on human social connections and their implications for behavior change and intervention strategies. The speaker emphasizes that human fates are intertwined, particularly illustrated through the recent COVID-19 pandemic, which showed how health issues can travel through social networks.

Three primary types of interventions are proposed:

1. **Connection Manipulation**: This video involves altering social connections, such as fostering introductions or severing ties. This video is considered complex in real life but can be easier in online environments.

2. **Contagion Interventions**: This video method leverages existing social connections to introduce behavior or information changes to selected individuals, which then spreads through their social networks to influence others.

3. **Position Interventions**: This video focuses on the arrangement of individuals within a network to optimize outcomes, such as strategically seating students in a classroom to enhance learning.

The speaker details a study conducted in Honduras involving maternal and child health interventions in isolated villages. The research targeted influential individuals within social networks, demonstrating that selecting individuals based on their social position rather than personal attributes can lead to greater adoption of health practices across the community.

In addition, the text explores how artificial intelligence (AI) agents interact within social frameworks and can influence human behaviors, emphasizing the need for studying the effects of AI on interpersonal relationships.

Ultimately, the speaker aims to understand the unique phenomenon of human social networks and apply this knowledge to create desirable societal changes, leveraging insights from social contagion to improve public health, economic behavior, and cooperative interactions within various contexts.

Language:

Export as PDF

Export as PDF
Here are five key points from the video:

1. **Interconnectedness of Human Fates**: The video emphasizes that human beings are connected through social networks, and events or behaviors in one person's life can significantly impact the lives of others, even those who are only indirectly connected. This interconnectedness is exemplified through the COVID-19 pandemic, illustrating how an infection can spread through social ties.

2. **Methods of Intervention**: Three primary methods of intervening in social systems are described:
- **Connection Manipulation**: Rewiring the social network by introducing or severing connections.
- **Contagion Interventions**: Selecting individuals within an existing network to introduce behavior changes, which can then ripple outward to influence others.
- **Position Intervention**: Optimally arranging individuals in a network to enhance group outcomes, such as seating arrangements in classrooms to promote learning.

3. **Empirical Research on Network Interventions**: The video discusses a study conducted in Honduras where researchers mapped social networks of villages and targeted structurally influential individuals to introduce health interventions (e.g., maternal and child health practices). The results demonstrated a significant positive spillover effect on the broader community, showcasing the effectiveness of social contagion.

4. **Implications for Public Health and Policy**: Understanding the structure and function of social networks allows for strategic interventions that can enhance the effectiveness of public health measures or social programs, leading to greater overall impacts. The research suggests that careful selection of initial individuals for intervention can dramatically increase the number of people who adopt new practices.

5. **Human-Machine Interaction**: The video also addresses the role of AI agents in social networks and their influence on human interactions. Experiments show that AI can be designed to facilitate cooperation and enhance social dynamics among humans, providing valuable insights into how the presence of machines can affect social behavior and relationships.

These points collectively highlight the importance of social networks in shaping behaviors and outcomes, and the potential for targeted interventions to produce positive changes in communities.

Export as PDF
Here are some key quotes extracted from the video:

1. **Connection of Fates**:
- "Human beings are connected, and so their fates are connected."

2. **Impact of Indirect Connections**:
- "Things that happen to me in my life depend on what's happening, not only to the people to whom I'm directly connected, but even the people to whom I'm indirectly connected."

3. **Pandemic Convideo**:
- "We've all just gone through a period with a COVID-19 pandemic where this idea should be very clear to everyone."

4. **Intervention Opportunities**:
- "This key insight gives us opportunities to develop new ways to intervene in social systems to make the world better."

5. **Types of Interventions**:
- "The first way is to manipulate connection... The second category is contagion interventions... The third type of intervention is something called the position intervention."

6. **Contagion Effect**:
- "If you change my behavior, it affects my friends and my friend's friends and my friend's friend's friends."

7. **Friendship Paradox**:
- "This relies on something known as the friendship paradox. This is a deep and fundamental mathematical fact about human social networks."

8. **Structurally Influential Individuals**:
- "We chose them by virtue of where they were located within the social fabric."

9. **Impact Measurement**:
- "Could we foster cascades? Could we create artificial tipping points by thoughtfully picking that 10%?"

10. **Broader Applications**:
- "The same ideas apply to factories, schools, hospitals, and communities."

11. **Human-Machine Interaction**:
- "How does the addition of these AI endowed systems affect human-human interactions?"

12. **Social Impact of AI Agents**:
- "Simply manipulations in how these AI agents act can affect how we treat each other."

13. **Understanding Human Nature**:
- "My most fundamental desire is to just deeply understand human beings."

14. **Formation of Friendships**:
- "We form volitional long-term non-reproductive unions with unrelated conspecifics, namely we have friends."

15. **Importance of Networks**:
- "Understanding networks gives us new levers to intervene."

These quotes encapsulate the main themes of connection, the impact of social networks, opportunities for intervention, and the interplay between human relationships and artificial intelligence.

X-TIMESTAMP-MAP=LOCAL:00:00:00.000,MPEGTS:90000

1
00:00:00.000 --> 00:00:01.350
Human beings are connected,

2
00:00:01.350 --> 00:00:03.165
and so their fates are connected.

3
00:00:03.665 --> 00:00:04.725
Uh, things that happen

4
00:00:04.725 --> 00:00:07.325
to me in my life depend on what's happening, not only

5
00:00:07.325 --> 00:00:10.245
to the people to whom I'm directly connected, such

6
00:00:10.245 --> 00:00:12.725
as my friends and coworkers and relatives and so on,

7
00:00:12.725 --> 00:00:14.925
but even the people to whom I'm indirectly connected,

8
00:00:15.305 --> 00:00:17.285
my friend's, friends or my friend's, friends friends.

9
00:00:17.825 --> 00:00:20.445
And we've all just gone through a period with a COVID-19,

10
00:00:20.545 --> 00:00:22.765
pandemic where this idea should be

11
00:00:22.765 --> 00:00:23.805
very clear to everyone.

12
00:00:24.185 --> 00:00:25.685
You weren't infected with a germ.

13
00:00:25.685 --> 00:00:27.325
Your friends weren't infected with a germ,

14
00:00:27.325 --> 00:00:29.725
but this unknown person to you, your friend's, friend's,

15
00:00:29.725 --> 00:00:31.085
friend, was infected.

16
00:00:31.465 --> 00:00:34.205
And that germ would inexorably follow the course

17
00:00:34.205 --> 00:00:35.925
of the social network and reach you.

18
00:00:36.585 --> 00:00:38.325
So people are connected and

19
00:00:38.325 --> 00:00:39.725
therefore their fates are connected.

20
00:00:39.725 --> 00:00:42.285
And this key insight gives us opportunities

21
00:00:42.745 --> 00:00:45.445
to develop new ways to intervene in social systems

22
00:00:45.585 --> 00:00:46.685
to make the world better.

23
00:00:47.465 --> 00:00:48.765
And there are three basic ways

24
00:00:48.765 --> 00:00:50.205
these interventions can be done.

25
00:00:50.745 --> 00:00:53.445
The first way is to manipulate connection, is to

26
00:00:54.005 --> 00:00:56.165
rewire the network broker introductions

27
00:00:56.185 --> 00:00:58.205
or sever ties between people.

28
00:00:58.705 --> 00:01:01.605
That's a so-called topological manipulation.

29
00:01:01.805 --> 00:01:03.085
A connection intervention.

30
00:01:03.345 --> 00:01:05.525
Of course, that's very difficult to do in real life.

31
00:01:06.145 --> 00:01:07.165
Uh, you can't go to someone

32
00:01:07.185 --> 00:01:08.845
and say, stop talking to your friends

33
00:01:08.865 --> 00:01:10.565
or divorce your partner, for instance,

34
00:01:10.945 --> 00:01:12.085
online, it's a little easier.

35
00:01:12.465 --> 00:01:14.565
You can engineer patterns

36
00:01:14.565 --> 00:01:16.085
of connection that affect outcomes.

37
00:01:16.905 --> 00:01:19.685
The second category is contagion interventions

38
00:01:20.055 --> 00:01:21.965
where you accept the social, the pattern

39
00:01:21.985 --> 00:01:23.605
of social connections between people.

40
00:01:24.265 --> 00:01:25.965
But now you th shrewdly

41
00:01:25.965 --> 00:01:28.525
and thoughtfully select particular individuals

42
00:01:28.665 --> 00:01:31.885
to whom you introduce, information or behavior change,

43
00:01:32.385 --> 00:01:34.965
and then watch the spreading of that behavior

44
00:01:35.355 --> 00:01:38.245
through the network such that if you change my behavior,

45
00:01:38.345 --> 00:01:40.045
you teach me to breastfeed my kids,

46
00:01:40.425 --> 00:01:42.765
and then that affects my friends and my friend's friends

47
00:01:42.785 --> 00:01:44.445
and my friend's, friends, friends, probability

48
00:01:44.505 --> 00:01:47.165
or you get me to quit smoking and hundreds

49
00:01:47.165 --> 00:01:49.525
or maybe a thousand other people wind up quitting smoking

50
00:01:49.755 --> 00:01:51.765
because you persuaded me to quit

51
00:01:52.545 --> 00:01:54.565
and you persuade me to quit smoking.

52
00:01:54.865 --> 00:01:57.725
And perhaps dozens or a hundred other people quit smoking

53
00:01:57.875 --> 00:01:59.685
because you persuaded me to quit.

54
00:01:59.985 --> 00:02:02.485
That's called a a contagion intervention.

55
00:02:02.545 --> 00:02:03.645
And the last category

56
00:02:03.645 --> 00:02:05.205
of intervention is something called the position

57
00:02:05.205 --> 00:02:07.365
intervention, which is probably not something we need

58
00:02:07.365 --> 00:02:09.125
to talk about, but it has to do with

59
00:02:09.175 --> 00:02:12.485
where you put people in a network such that

60
00:02:12.505 --> 00:02:14.365
by arranging them in a particular way,

61
00:02:14.865 --> 00:02:17.445
it can cause the group to, do well.

62
00:02:17.785 --> 00:02:21.525
For example, if you have a network of desks in a classroom

63
00:02:21.905 --> 00:02:24.525
and you have a mix of students, some are good, some are not

64
00:02:24.525 --> 00:02:27.885
so good students, where do you put the kids in that network

65
00:02:27.905 --> 00:02:31.685
of chairs such that learning in the classroom is optimized?

66
00:02:32.225 --> 00:02:34.725
So we've been doing lots of experiments

67
00:02:34.725 --> 00:02:37.845
and interventions in my laboratory over the last 20 years

68
00:02:38.555 --> 00:02:41.285
exploring, for example, social contagion in,

69
00:02:41.345 --> 00:02:43.245
in real complex networks.

70
00:02:43.745 --> 00:02:45.965
And we had this paper that I think you're interested in

71
00:02:45.965 --> 00:02:50.005
that was published in May of 2024 that,

72
00:02:50.155 --> 00:02:53.165
illustrated the contagion idea in a very powerful way.

73
00:02:53.385 --> 00:02:56.045
So what we were interested in doing in this study was

74
00:02:56.045 --> 00:02:57.885
understanding whether it was possible

75
00:02:57.885 --> 00:03:01.365
to exploit a deep understanding of social network structure

76
00:03:01.385 --> 00:03:02.845
and function so as

77
00:03:02.845 --> 00:03:05.205
to intervene in networks to make the world better.

78
00:03:05.585 --> 00:03:07.645
For example, development economics interventions

79
00:03:07.645 --> 00:03:09.765
or public health interventions,

80
00:03:09.785 --> 00:03:11.525
or behavior change interventions.

81
00:03:12.225 --> 00:03:14.845
Can we, understand the structure

82
00:03:14.845 --> 00:03:16.165
and function of social networks

83
00:03:16.305 --> 00:03:19.565
and then do something to facilitate the flow

84
00:03:19.585 --> 00:03:22.085
of desirable knowledge and desirable practices?

85
00:03:22.625 --> 00:03:25.085
So in this particular study, we

86
00:03:25.345 --> 00:03:26.765
mapped the social networks

87
00:03:26.765 --> 00:03:31.285
of 176 villages in the western highlands of Honduras.

88
00:03:31.285 --> 00:03:33.205
These were isolated villages deep in the jungle,

89
00:03:33.505 --> 00:03:34.925
30,000 people.

90
00:03:35.235 --> 00:03:36.645
This was not easy to do.

91
00:03:37.025 --> 00:03:39.525
And we mapped the network repeatedly across time,

92
00:03:40.025 --> 00:03:41.605
but at baseline we mapped the network

93
00:03:41.705 --> 00:03:45.285
and then we used mathematical algorithms to select

94
00:03:46.325 --> 00:03:48.565
structurally influential individuals within the network.

95
00:03:48.745 --> 00:03:52.125
So it's not people, we didn't choose people by virtue

96
00:03:52.185 --> 00:03:54.245
of their attributes, who was rich

97
00:03:54.245 --> 00:03:56.485
or poor, who was a man or a woman, for example.

98
00:03:56.945 --> 00:03:58.365
We chose them by virtue of

99
00:03:58.365 --> 00:04:00.685
where they were located within the social fabric.

100
00:04:01.105 --> 00:04:02.645
For example, did they have many friends

101
00:04:02.665 --> 00:04:04.045
and did their friends have many friends?

102
00:04:04.045 --> 00:04:05.525
Were they very central in the network?

103
00:04:06.465 --> 00:04:08.885
Uh, and then we gave them a two year maternal

104
00:04:08.905 --> 00:04:10.125
and child health intervention.

105
00:04:10.905 --> 00:04:15.485
We came back two years later and we checked not only whether

106
00:04:15.545 --> 00:04:17.885
or not the people to whom we gave the intervention,

107
00:04:18.545 --> 00:04:20.925
not only did they adopt the intervention,

108
00:04:21.065 --> 00:04:23.205
but did everyone else in the village also adopt?

109
00:04:23.585 --> 00:04:27.525
So the idea is can we map a network, pick 10%

110
00:04:27.525 --> 00:04:29.645
of the people within the network, for example,

111
00:04:30.185 --> 00:04:31.925
who are structurally influential,

112
00:04:32.265 --> 00:04:33.925
and then come back a year later and,

113
00:04:33.945 --> 00:04:37.565
and measure not only did that 10% adopt the breastfeeding

114
00:04:37.705 --> 00:04:41.645
or, or cord care or vaccination or diarrhea intervention,

115
00:04:42.265 --> 00:04:44.245
but did everyone else adopt the intervention?

116
00:04:44.935 --> 00:04:47.045
Could we foster cascades?

117
00:04:47.045 --> 00:04:49.245
Could we create artificial tipping points

118
00:04:49.945 --> 00:04:53.965
by thoughtfully picking that 10% relies on something known

119
00:04:53.965 --> 00:04:55.285
as the friendship paradox.

120
00:04:55.285 --> 00:04:58.525
This is a deep and fundamental mathematical fact about human

121
00:04:58.525 --> 00:05:00.405
social networks, which is

122
00:05:00.405 --> 00:05:03.965
that your friends have more friends than you do actually,

123
00:05:03.965 --> 00:05:06.405
your sexual partners have more sexual partners than you do

124
00:05:06.435 --> 00:05:09.925
also, and most depressing for me as a scientist is

125
00:05:09.925 --> 00:05:12.805
that your co-authors have more co-authors than you do.

126
00:05:13.465 --> 00:05:16.525
So this is a fundamental mathematical fact about networks.

127
00:05:16.545 --> 00:05:19.725
But the beauty of this is that you can go into a village,

128
00:05:20.275 --> 00:05:23.125
pick 10% of the people at random, have them tell you

129
00:05:23.125 --> 00:05:25.885
who their friends are, pick one of those friends at random.

130
00:05:26.145 --> 00:05:29.485
So now you have a new 10% sample, which was the friends

131
00:05:29.505 --> 00:05:30.845
of the random 10%,

132
00:05:31.305 --> 00:05:33.805
and you give the intervention to that 10%.

133
00:05:34.265 --> 00:05:36.085
And by doing that little trick,

134
00:05:36.425 --> 00:05:40.125
you significantly enhance the spillover effects the cascades

135
00:05:40.125 --> 00:05:41.805
because that latter 10%, the friends

136
00:05:41.805 --> 00:05:44.885
of the random are much more structurally influential

137
00:05:45.265 --> 00:05:48.965
and can induce these cascades within these villages.

138
00:05:48.985 --> 00:05:50.445
And that's the experiment we did.

139
00:05:50.445 --> 00:05:54.045
It took us eight years of, of my life, we had a lot

140
00:05:54.045 --> 00:05:56.805
of support from the Gates Foundation, from the

141
00:05:57.015 --> 00:05:59.805
Nomis Foundation, from the Robert Wood Johnson Foundation

142
00:05:59.805 --> 00:06:01.045
and from the federal government,

143
00:06:01.545 --> 00:06:04.485
and also support from the Ministry of Health in Honduras

144
00:06:04.485 --> 00:06:06.165
and the Inter-American Development Bank.

145
00:06:07.445 --> 00:06:11.065
And we did this enormous field trial to show that,

146
00:06:11.295 --> 00:06:14.985
it's possible to manipulate social contagion within real on

147
00:06:14.985 --> 00:06:16.345
the ground, face-to-face social

148
00:06:16.385 --> 00:06:18.225
networks to make the world better.

149
00:06:18.555 --> 00:06:20.665
Let's say you have a village of a thousand people

150
00:06:21.165 --> 00:06:22.745
and you pick 10% of them.

151
00:06:23.365 --> 00:06:25.665
So a hundred people, and you give them a maternal

152
00:06:25.665 --> 00:06:26.825
and child health intervention

153
00:06:27.685 --> 00:06:31.225
and let's say 50 of them adopt the practice,

154
00:06:31.225 --> 00:06:33.825
you're encouraging them to adopt, for example, using zinc

155
00:06:34.045 --> 00:06:36.745
to treat diarrheal illness in their minor children.

156
00:06:37.765 --> 00:06:41.745
And those 50 who adopt in turn influence some fraction

157
00:06:41.745 --> 00:06:44.425
of the un of the 900 untreated people.

158
00:06:44.965 --> 00:06:47.865
So maybe they influence 50 of those 900 people.

159
00:06:48.285 --> 00:06:50.545
So a total of 100 people in this village

160
00:06:50.545 --> 00:06:53.105
of a thousand have adopted the new practice

161
00:06:53.405 --> 00:06:54.865
of properly using zinc

162
00:06:54.865 --> 00:06:56.985
to treat diarrhea illness in minor children.

163
00:06:58.735 --> 00:06:59.915
Now, in a different village,

164
00:07:00.415 --> 00:07:01.915
you pick a hundred people again,

165
00:07:02.295 --> 00:07:05.435
but now out of the thousand, this is a different village

166
00:07:05.815 --> 00:07:07.115
and they nominate their friends.

167
00:07:07.455 --> 00:07:10.195
And you pick a hundred friends of the a hundred randoms,

168
00:07:10.195 --> 00:07:11.835
which is a different set of human beings.

169
00:07:12.375 --> 00:07:14.915
And of those people, of those a hundred people,

170
00:07:15.045 --> 00:07:17.715
let's say once again, 50 of those people

171
00:07:18.275 --> 00:07:20.355
actually would be slightly more for various reasons.

172
00:07:20.495 --> 00:07:23.515
But let's say 50 of those people again respond

173
00:07:23.515 --> 00:07:26.075
to the public health intervention of, in this case zinc

174
00:07:26.095 --> 00:07:27.635
for treating diarrheal illness.

175
00:07:28.455 --> 00:07:31.475
But because they are more centrally located in the network,

176
00:07:31.475 --> 00:07:33.035
they're more structurally influential,

177
00:07:33.345 --> 00:07:35.635
they might influence 200 out

178
00:07:35.635 --> 00:07:37.515
of the 900 other people to adopt.

179
00:07:37.935 --> 00:07:40.955
So in this latter village, we have 250 people adopting,

180
00:07:40.955 --> 00:07:42.035
whereas in the first village,

181
00:07:42.095 --> 00:07:43.955
we had a hundred people adopting overall.

182
00:07:44.695 --> 00:07:46.115
So this is a way to greatly

183
00:07:46.635 --> 00:07:47.915
increase the return on investment

184
00:07:48.015 --> 00:07:49.755
and the impact of these public health

185
00:07:49.755 --> 00:07:51.515
and development economics interventions

186
00:07:51.515 --> 00:07:53.795
because it's the same counterfactually,

187
00:07:53.795 --> 00:07:55.755
it's the same village, the same number of people,

188
00:07:56.095 --> 00:07:58.035
the same treatment, two

189
00:07:58.035 --> 00:08:01.155
and a half times the effect, sometimes five times the effect

190
00:08:01.695 --> 00:08:04.635
simply because of how you pick the initial people

191
00:08:04.695 --> 00:08:05.795
to seed in the network.

192
00:08:06.335 --> 00:08:09.715
So this, this exploitation of an understanding

193
00:08:09.775 --> 00:08:12.595
of social contagion within social networks offers,

194
00:08:12.775 --> 00:08:14.435
offers us tremendous new tools

195
00:08:14.855 --> 00:08:16.835
to intervene in networks to make the world better.

196
00:08:16.895 --> 00:08:19.075
And by the way, this isn't just in, in,

197
00:08:19.075 --> 00:08:20.275
developing world villages.

198
00:08:20.295 --> 00:08:22.515
The same ideas apply to factories.

199
00:08:22.515 --> 00:08:25.195
If you're trying to get people to put on hard hats

200
00:08:25.195 --> 00:08:28.235
or reduce absenteeism in a factory to schools,

201
00:08:28.335 --> 00:08:29.675
if you're trying to reduce bullying

202
00:08:29.895 --> 00:08:33.035
or reduce drug use to hospitals, if you're trying

203
00:08:33.035 --> 00:08:34.275
to change the behavior of doctors

204
00:08:34.375 --> 00:08:36.515
or nurses to our whole community,

205
00:08:36.895 --> 00:08:40.315
you can use this understanding of social contagion

206
00:08:40.315 --> 00:08:43.555
and complex networks here shown using experiments

207
00:08:43.575 --> 00:08:45.035
so we can very rigorous,

208
00:08:45.095 --> 00:08:46.715
we can be sure we know what's going on.

209
00:08:47.055 --> 00:08:48.315
You can use this understanding

210
00:08:48.375 --> 00:08:50.075
to intervene in the world in a variety

211
00:08:50.075 --> 00:08:52.835
of policy domains in a different line of work.

212
00:08:53.295 --> 00:08:56.205
We are taking seriously the observation

213
00:08:56.275 --> 00:08:59.325
that artificial intelligence agents are increasingly being

214
00:08:59.325 --> 00:09:00.565
added into our midst.

215
00:09:00.945 --> 00:09:02.925
So we are adding, for example, bots

216
00:09:02.925 --> 00:09:04.365
to online social networks

217
00:09:04.705 --> 00:09:07.885
or physical agents into our, our everyday networks,

218
00:09:07.885 --> 00:09:09.485
autonomous vehicles on highways

219
00:09:09.485 --> 00:09:13.445
or automatic checkout clerks in supermarkets and so on.

220
00:09:13.825 --> 00:09:16.885
And what's interesting about these AI endowed agents is, is

221
00:09:16.885 --> 00:09:18.885
that they interact with us socially on a level,

222
00:09:19.095 --> 00:09:21.205
level playing field as if they were social,

223
00:09:21.545 --> 00:09:23.445
as if they were human beings in some sense.

224
00:09:24.585 --> 00:09:27.945
So let me give you a toy example to help fix ideas,

225
00:09:28.005 --> 00:09:29.625
and then I can tell you a little bit about some

226
00:09:29.625 --> 00:09:32.105
of the experiments we've been doing in my lab over the last

227
00:09:32.125 --> 00:09:35.145
10 years to understand these hybrid systems

228
00:09:35.205 --> 00:09:36.465
of humans and machines.

229
00:09:37.205 --> 00:09:39.465
So the manufacturer of a digital assistant,

230
00:09:39.535 --> 00:09:40.785
like an Alexa device,

231
00:09:40.885 --> 00:09:43.985
is very interested in the human machine interaction,

232
00:09:44.555 --> 00:09:47.025
optimizing this product so that when you use the product,

233
00:09:47.255 --> 00:09:48.465
it's satisfying to you.

234
00:09:48.845 --> 00:09:51.545
For example, you would never buy an Alexa if it required

235
00:09:51.935 --> 00:09:53.425
obscene levels of politeness.

236
00:09:53.445 --> 00:09:56.585
If you had to say, excuse me, Alexa, I'm very sorry

237
00:09:56.585 --> 00:09:58.465
to interrupt you if you don't mind.

238
00:09:58.755 --> 00:10:00.505
Would you please tell me the weather tomorrow?

239
00:10:01.095 --> 00:10:02.145
This is ridiculous.

240
00:10:02.165 --> 00:10:04.225
You would you expect to be able to say to the machine,

241
00:10:04.275 --> 00:10:06.825
Alexa, weather, and then the machine tells you the weather,

242
00:10:06.835 --> 00:10:10.505
which is fine until you bring that machine into your house

243
00:10:10.925 --> 00:10:14.625
and your children use it and your children learn to be rude,

244
00:10:15.125 --> 00:10:16.425
and then they go to the playground

245
00:10:16.425 --> 00:10:17.945
and they are rude to other children.

246
00:10:19.155 --> 00:10:21.975
So what my laboratory is interested in is human,

247
00:10:22.025 --> 00:10:24.695
human interactions in the presence of machines.

248
00:10:25.115 --> 00:10:28.175
How does the addition of these AI endowed systems

249
00:10:29.165 --> 00:10:32.775
into the social fabric affect how we treat each other?

250
00:10:33.235 --> 00:10:35.455
And we've done a variety of experiments that show

251
00:10:35.765 --> 00:10:37.775
that these agents can be programmed

252
00:10:38.075 --> 00:10:40.135
to affect how we treat each other.

253
00:10:40.515 --> 00:10:43.735
For example, we did one experiment in which we had humanoid

254
00:10:43.915 --> 00:10:46.055
robots, these cute little robots that,

255
00:10:46.055 --> 00:10:47.095
you know, looked like people.

256
00:10:47.635 --> 00:10:49.935
And we did an experiment in which we brought three people

257
00:10:49.995 --> 00:10:51.295
who were strangers to each other

258
00:10:51.715 --> 00:10:53.255
and added the robot to the mix.

259
00:10:53.255 --> 00:10:55.375
And we told the four of them they needed

260
00:10:55.375 --> 00:10:57.255
to play a game on a tablet computer

261
00:10:57.425 --> 00:10:59.215
where they were laying railroad track

262
00:10:59.215 --> 00:11:00.375
from point A to point B.

263
00:11:00.555 --> 00:11:02.855
So the people and the robot were all playing together,

264
00:11:03.315 --> 00:11:04.775
laying this railroad track.

265
00:11:05.235 --> 00:11:08.535
And then we contrived to have this game not work sometimes,

266
00:11:08.565 --> 00:11:11.975
like it was impossible to pick from the pieces of track

267
00:11:12.515 --> 00:11:15.055
in a way that allowed you to connect point A to point B

268
00:11:15.055 --> 00:11:18.095
that just wasn't the right pattern of tracks, the straits

269
00:11:18.095 --> 00:11:19.095
and the curves and so on.

270
00:11:19.915 --> 00:11:21.735
And then what we manipulated is whether

271
00:11:21.755 --> 00:11:24.655
or not the robot told dad jokes,

272
00:11:25.165 --> 00:11:28.215
like sometimes the robot made silly, stupid jokes,

273
00:11:28.365 --> 00:11:29.615
that is what we call dad jokes.

274
00:11:30.115 --> 00:11:31.735
And other times it was just very neutral.

275
00:11:31.735 --> 00:11:33.815
It would say, you know, we have played five rounds

276
00:11:33.815 --> 00:11:35.815
of the game and have not completed the track.

277
00:11:36.315 --> 00:11:37.855
And other times it would say, you know,

278
00:11:37.855 --> 00:11:39.135
people think robots are perfect,

279
00:11:39.195 --> 00:11:41.695
but actually we are imperfect or something silly.

280
00:11:42.615 --> 00:11:45.255
And then what we studied was not how the humans interacted

281
00:11:45.255 --> 00:11:46.455
with a robot, but

282
00:11:46.455 --> 00:11:50.095
how did the human human conversation change depending on

283
00:11:50.095 --> 00:11:52.855
whether we program the robot to make vulnerable statements

284
00:11:52.855 --> 00:11:54.100
or, or tell silly jokes.

285
00:11:54.385 --> 00:11:57.685
And we found that modifying the nature of speech

286
00:11:57.685 --> 00:12:01.365
of the machine changed the frequency of conversation

287
00:12:01.455 --> 00:12:03.645
among the humans, the quality of conversation

288
00:12:03.745 --> 00:12:06.245
and the satisfaction the humans had with that conversation.

289
00:12:06.905 --> 00:12:10.405
So simple manipulations in how these AI agents

290
00:12:11.145 --> 00:12:13.805
act can affect how we treat each other.

291
00:12:14.465 --> 00:12:16.765
And in a series of other experiments we've shown,

292
00:12:16.785 --> 00:12:19.365
for instance, that we can add human,

293
00:12:19.365 --> 00:12:23.205
we can add these AI agents endowed with what we call dumb ai

294
00:12:23.895 --> 00:12:26.945
because in my lab we're not interested in inventing super

295
00:12:27.115 --> 00:12:29.745
smart AI to replace human cognition.

296
00:12:30.595 --> 00:12:33.985
We're inventing dumb AI to supplement human interaction.

297
00:12:34.565 --> 00:12:36.145
So what we're doing is, is we're,

298
00:12:36.145 --> 00:12:37.425
because the humans are smart,

299
00:12:37.445 --> 00:12:39.225
the AI agents can afford to be stupid.

300
00:12:39.565 --> 00:12:41.265
All they need to do is make it easier for us

301
00:12:41.265 --> 00:12:43.305
to work together to solve problems.

302
00:12:43.605 --> 00:12:44.905
And we've shown that this can be done

303
00:12:44.905 --> 00:12:49.225
to help humans coordinate, cooperate, innovate, and so on.

304
00:12:49.285 --> 00:12:52.345
In a series of experiments, my most fundamental desire is

305
00:12:52.345 --> 00:12:54.345
to just deeply understand human beings.

306
00:12:54.645 --> 00:12:57.985
And you know, I think we are magnificent creatures

307
00:12:58.485 --> 00:13:01.865
and we have, you know, as objects in the natural world,

308
00:13:01.965 --> 00:13:03.265
we are deeply interesting.

309
00:13:04.045 --> 00:13:07.345
For example, the mere fact that we have friends, which,

310
00:13:07.355 --> 00:13:09.705
which many listeners are probably taking for granted.

311
00:13:09.945 --> 00:13:12.945
I mean, we, we reproduce sexually like many other species,

312
00:13:13.205 --> 00:13:15.705
the vast number of species reproduce sexually.

313
00:13:15.965 --> 00:13:17.785
So we're not so interesting in that regard

314
00:13:18.165 --> 00:13:19.305
as far as I'm concerned.

315
00:13:19.605 --> 00:13:20.825
But we do something that's very

316
00:13:20.825 --> 00:13:22.225
unusual in the animal kingdom.

317
00:13:22.285 --> 00:13:26.785
We form volitional long-term non reproductive unions

318
00:13:26.785 --> 00:13:30.905
with unrelated con specifics, namely we have friends, okay?

319
00:13:31.205 --> 00:13:34.385
We, we form social ties to people to who,

320
00:13:34.385 --> 00:13:35.425
who are not our kin.

321
00:13:35.725 --> 00:13:36.865
And we help those people.

322
00:13:37.375 --> 00:13:39.145
This is exceeding, this phenotype is

323
00:13:39.145 --> 00:13:40.585
exceedingly rare in the animal kingdom.

324
00:13:40.805 --> 00:13:43.785
We do it. Certain other primates do it, elephants do it.

325
00:13:44.105 --> 00:13:45.785
Certain Satan species do it.

326
00:13:45.785 --> 00:13:47.865
And this is very different, for example, than packs

327
00:13:47.865 --> 00:13:50.105
of wolves who are all cousins, for example.

328
00:13:50.485 --> 00:13:52.945
You know, we, we formed these relationships with people

329
00:13:52.965 --> 00:13:54.505
who are not our, our relatives.

330
00:13:55.205 --> 00:13:57.665
And this is astonishing actually, as far as I'm concerned

331
00:13:57.665 --> 00:13:59.425
that we have friends and

332
00:13:59.425 --> 00:14:01.585
that we independently evolved this phenotype.

333
00:14:01.765 --> 00:14:03.585
You know, the last common ancestor we had

334
00:14:03.585 --> 00:14:05.345
with elephants was 90 million years ago,

335
00:14:05.765 --> 00:14:09.585
and they evolved this capacity, and we did, for example, and

336
00:14:09.585 --> 00:14:10.625
because of this, we,

337
00:14:10.905 --> 00:14:13.425
assemble ourselves into these networks, these ornate

338
00:14:14.135 --> 00:14:15.305
magnificent things.

339
00:14:15.395 --> 00:14:17.585
These, these, these insubstantial you.

340
00:14:17.685 --> 00:14:19.025
You can't, you can't see

341
00:14:19.025 --> 00:14:20.065
them when you walk through the world.

342
00:14:20.065 --> 00:14:23.505
And yet they exist. They're real, these, these phenomena.

343
00:14:23.925 --> 00:14:26.305
And and it's just very captivating to me.

344
00:14:26.765 --> 00:14:29.665
So my fundamental concern is just to understand this is

345
00:14:29.665 --> 00:14:31.585
to understand social networks,

346
00:14:31.765 --> 00:14:35.425
but we also would like to, as I said,

347
00:14:35.565 --> 00:14:37.305
use this understanding to make the world better.

348
00:14:37.325 --> 00:14:40.825
And we've shown in multiple experiments over the last 20

349
00:14:40.825 --> 00:14:43.785
years that you can exploit an understanding

350
00:14:43.785 --> 00:14:44.825
of social network structure

351
00:14:44.825 --> 00:14:46.465
and function to intervene in the world,

352
00:14:46.565 --> 00:14:50.705
to effectuate desirable changes in human behavior,

353
00:14:50.845 --> 00:14:53.645
in the flow of information, in the ability of people

354
00:14:53.705 --> 00:14:54.765
to cooperate

355
00:14:54.765 --> 00:14:58.045
and treat each other well, in, in, in the ability

356
00:14:58.105 --> 00:14:59.885
to change public health behaviors

357
00:14:59.905 --> 00:15:01.605
or economic behaviors, for example.

358
00:15:02.225 --> 00:15:06.205
So a across a broad range of social behaviors

359
00:15:06.205 --> 00:15:09.485
and social phenomena, understanding networks gives us new

360
00:15:09.545 --> 00:15:11.485
levers, to intervene.

361
00:15:11.625 --> 00:15:14.165
And I find that kind of a secondary benefit,

362
00:15:14.365 --> 00:15:16.165
a pleasant secondary benefit of the work we're doing.

Interconnected Fates: Understanding Social Networks and Interventions

00:00:00 - Human beings are connected, and so their fates are connected. things that happen to me in my life depend on what's happening, not only to the people to whom I'm directly connected, such as my friends and coworkers and relatives and so on, but even the people to whom I'm indirectly connected, my friend's, friends or my friend's, friends friends. And we've all just gone through a period with a COVID-19, pandemic where this idea should be very clear to everyone. You weren't infected with a germ. Your friends weren't infected with a germ, but this unknown person to you, your friend's, friend's, friend, was infected.

00:00:34 - of the social network and reach you. So people are connected and therefore their fates are connected. And this key insight gives us opportunities to develop new ways to intervene in social systems to make the world better. And there are three basic ways these interventions can be done. The first way is to manipulate connection, is to rewire the network broker introductions or sever ties between people. That's a so-called topological manipulation.

00:01:03 - Of course, that's very difficult to do in real life. you can't go to someone and say, stop talking to your friends or divorce your partner, for instance, online, it's a little easier. You can engineer patterns of connection that affect outcomes. The second category is contagion interventions where you accept the social, the pattern of social connections between people. But now you th shrewdly and thoughtfully select particular individuals to whom you introduce, information or behavior change,

Interventions in Social Networks: Contagion and Positioning

00:01:35 - through the network such that if you change my behavior, you teach me to breastfeed my kids, and then that affects my friends and my friend's friends and my friend's, friends, friends, probability or you get me to quit smoking and hundreds or maybe a thousand other people wind up quitting smoking because you persuaded me to quit and you persuade me to quit smoking. And perhaps dozens or a hundred other people quit smoking because you persuaded me to quit. That's called a a contagion intervention.

00:02:03 - of intervention is something called the position intervention, which is probably not something we need to talk about, but it has to do with where you put people in a network such that by arranging them in a particular way, it can cause the group to, do well. For example, if you have a network of desks in a classroom and you have a mix of students, some are good, some are not so good students, where do you put the kids in that network of chairs such that learning in the classroom is optimized? So we've been doing lots of experiments

00:02:38 - exploring, for example, social contagion in, in real complex networks. And we had this paper that I think you're interested in that was published in May of 2024 that, illustrated the contagion idea in a very powerful way. So what we were interested in doing in this study was understanding whether it was possible to exploit a deep understanding of social network structure and function so as to intervene in networks to make the world better.

Mapping Networks for Health Interventions

00:03:07 - or public health interventions, or behavior change interventions. Can we, understand the structure and function of social networks and then do something to facilitate the flow of desirable knowledge and desirable practices? So in this particular study, we mapped the social networks of 176 villages in the western highlands of Honduras. These were isolated villages deep in the jungle, 30,000 people. This was not easy to do.

00:03:40 - but at baseline we mapped the network and then we used mathematical algorithms to select structurally influential individuals within the network. So it's not people, we didn't choose people by virtue of their attributes, who was rich or poor, who was a man or a woman, for example. We chose them by virtue of where they were located within the social fabric. For example, did they have many friends and did their friends have many friends? Were they very central in the network? and then we gave them a two year maternal

00:04:10 - We came back two years later and we checked not only whether or not the people to whom we gave the intervention, not only did they adopt the intervention, but did everyone else in the village also adopt? So the idea is can we map a network, pick 10% of the people within the network, for example, who are structurally influential, and then come back a year later and, and measure not only did that 10% adopt the breastfeeding or, or cord care or vaccination or diarrhea intervention,

Leveraging the Friendship Paradox for Social Interventions

00:04:44 - Could we foster cascades? Could we create artificial tipping points by thoughtfully picking that 10% relies on something known as the friendship paradox. This is a deep and fundamental mathematical fact about human social networks, which is that your friends have more friends than you do actually, your sexual partners have more sexual partners than you do also, and most depressing for me as a scientist is that your co-authors have more co-authors than you do.

00:05:16 - But the beauty of this is that you can go into a village, pick 10% of the people at random, have them tell you who their friends are, pick one of those friends at random. So now you have a new 10% sample, which was the friends of the random 10%, and you give the intervention to that 10%. And by doing that little trick, you significantly enhance the spillover effects the cascades because that latter 10%, the friends of the random are much more structurally influential

00:05:48 - And that's the experiment we did. It took us eight years of, of my life, we had a lot of support from the Gates Foundation, from the Nomis Foundation, from the Robert Wood Johnson Foundation and from the federal government, and also support from the Ministry of Health in Honduras and the Inter-American Development Bank. And we did this enormous field trial to show that, it's possible to manipulate social contagion within real on the ground, face-to-face social

Enhancing Public Health Impact Through Strategic Social Network Interventions

00:06:18 - Let's say you have a village of a thousand people and you pick 10% of them. So a hundred people, and you give them a maternal and child health intervention and let's say 50 of them adopt the practice, you're encouraging them to adopt, for example, using zinc to treat diarrheal illness in their minor children. And those 50 who adopt in turn influence some fraction of the un of the 900 untreated people. So maybe they influence 50 of those 900 people.

00:06:50 - of a thousand have adopted the new practice of properly using zinc to treat diarrhea illness in minor children. Now, in a different village, you pick a hundred people again, but now out of the thousand, this is a different village and they nominate their friends. And you pick a hundred friends of the a hundred randoms, which is a different set of human beings. And of those people, of those a hundred people, let's say once again, 50 of those people actually would be slightly more for various reasons.

00:07:23 - to the public health intervention of, in this case zinc for treating diarrheal illness. But because they are more centrally located in the network, they're more structurally influential, they might influence 200 out of the 900 other people to adopt. So in this latter village, we have 250 people adopting, whereas in the first village, we had a hundred people adopting overall. So this is a way to greatly increase the return on investment and the impact of these public health and development economics interventions

Leveraging Social Contagion for Effective Interventions

00:07:53 - it's the same village, the same number of people, the same treatment, two and a half times the effect, sometimes five times the effect simply because of how you pick the initial people to seed in the network. So this, this exploitation of an understanding of social contagion within social networks offers, offers us tremendous new tools to intervene in networks to make the world better. And by the way, this isn't just in, in, developing world villages. The same ideas apply to factories.

00:08:25 - or reduce absenteeism in a factory to schools, if you're trying to reduce bullying or reduce drug use to hospitals, if you're trying to change the behavior of doctors or nurses to our whole community, you can use this understanding of social contagion and complex networks here shown using experiments so we can very rigorous, we can be sure we know what's going on. You can use this understanding to intervene in the world in a variety of policy domains in a different line of work.

00:08:56 - that artificial intelligence agents are increasingly being added into our midst. So we are adding, for example, bots to online social networks or physical agents into our, our everyday networks, autonomous vehicles on highways or automatic checkout clerks in supermarkets and so on. And what's interesting about these AI endowed agents is, is that they interact with us socially on a level, level playing field as if they were social, as if they were human beings in some sense.

Exploring Human-Machine Interaction in AI Systems

00:09:28 - and then I can tell you a little bit about some of the experiments we've been doing in my lab over the last 10 years to understand these hybrid systems of humans and machines. So the manufacturer of a digital assistant, like an Alexa device, is very interested in the human machine interaction, optimizing this product so that when you use the product, it's satisfying to you. For example, you would never buy an Alexa if it required obscene levels of politeness. If you had to say, excuse me, Alexa, I'm very sorry

00:09:58 - Would you please tell me the weather tomorrow? This is ridiculous. You would you expect to be able to say to the machine, Alexa, weather, and then the machine tells you the weather, which is fine until you bring that machine into your house and your children use it and your children learn to be rude, and then they go to the playground and they are rude to other children. So what my laboratory is interested in is human, human interactions in the presence of machines. How does the addition of these AI endowed systems

00:10:33 - And we've done a variety of experiments that show that these agents can be programmed to affect how we treat each other. For example, we did one experiment in which we had humanoid robots, these cute little robots that, you know, looked like people. And we did an experiment in which we brought three people who were strangers to each other and added the robot to the mix. And we told the four of them they needed to play a game on a tablet computer where they were laying railroad track

Impact of Robot Humor on Human Interaction

00:11:00 - So the people and the robot were all playing together, laying this railroad track. And then we contrived to have this game not work sometimes, like it was impossible to pick from the pieces of track in a way that allowed you to connect point A to point B that just wasn't the right pattern of tracks, the straits and the curves and so on. And then what we manipulated is whether or not the robot told dad jokes, like sometimes the robot made silly, stupid jokes, that is what we call dad jokes.

00:11:31 - It would say, you know, we have played five rounds of the game and have not completed the track. And other times it would say, you know, people think robots are perfect, but actually we are imperfect or something silly. And then what we studied was not how the humans interacted with a robot, but how did the human human conversation change depending on whether we program the robot to make vulnerable statements or, or tell silly jokes. And we found that modifying the nature of speech of the machine changed the frequency of conversation

00:12:03 - and the satisfaction the humans had with that conversation. So simple manipulations in how these AI agents act can affect how we treat each other. And in a series of other experiments we've shown, for instance, that we can add human, we can add these AI agents endowed with what we call dumb ai because in my lab we're not interested in inventing super smart AI to replace human cognition. We're inventing dumb AI to supplement human interaction.

Understanding Human Connections: AI's Role in Social Cooperation

00:12:36 - because the humans are smart, the AI agents can afford to be stupid. All they need to do is make it easier for us to work together to solve problems. And we've shown that this can be done to help humans coordinate, cooperate, innovate, and so on. In a series of experiments, my most fundamental desire is to just deeply understand human beings. And you know, I think we are magnificent creatures and we have, you know, as objects in the natural world, we are deeply interesting. For example, the mere fact that we have friends, which,

00:13:09 - I mean, we, we reproduce sexually like many other species, the vast number of species reproduce sexually. So we're not so interesting in that regard as far as I'm concerned. But we do something that's very unusual in the animal kingdom. We form volitional long-term non reproductive unions with unrelated con specifics, namely we have friends, okay? We, we form social ties to people to who, who are not our kin. And we help those people.

00:13:39 - exceedingly rare in the animal kingdom. We do it. Certain other primates do it, elephants do it. Certain Satan species do it. And this is very different, for example, than packs of wolves who are all cousins, for example. You know, we, we formed these relationships with people who are not our, our relatives. And this is astonishing actually, as far as I'm concerned that we have friends and that we independently evolved this phenotype. You know, the last common ancestor we had with elephants was 90 million years ago, and they evolved this capacity, and we did, for example, and

Understanding and Leveraging Social Networks for Positive Change

00:14:10 - assemble ourselves into these networks, these ornate magnificent things. These, these, these insubstantial you. You can't, you can't see them when you walk through the world. And yet they exist. They're real, these, these phenomena. And and it's just very captivating to me. So my fundamental concern is just to understand this is to understand social networks, but we also would like to, as I said, use this understanding to make the world better. And we've shown in multiple experiments over the last 20

00:14:43 - of social network structure and function to intervene in the world, to effectuate desirable changes in human behavior, in the flow of information, in the ability of people to cooperate and treat each other well, in, in, in the ability to change public health behaviors or economic behaviors, for example. So a across a broad range of social behaviors and social phenomena, understanding networks gives us new levers, to intervene.

00:15:14 - a pleasant secondary benefit of the work we're doing.

Faculti AI can make mistakes

Research

Policy

Analysis

Data

Originals

Explainers

Podcasts

Faculti

New on Faculti

Idea Explorer

Definition

News Context

Critical Questions and Answers

Research Summaries

Induction of Social Contagion

Media Literacy Toolkit

Critical Questions to Consider ▼

Bias and Perspective Awareness ▼

Further Reading ▼