0:00:00.554,0:00:03.021
As societies, we have to make[br]collective decisions

0:00:03.021,0:00:04.591
that will shape our future.

0:00:05.087,0:00:07.868
And we all know that when[br]we make decisions in groups,

0:00:07.868,0:00:09.192
they don't always go right.

0:00:09.650,0:00:11.606
And sometimes they go very wrong.

0:00:12.419,0:00:14.843
So how do groups make good decisions?

0:00:15.300,0:00:19.296
Research has shown that crowds are wise[br]when there's independent thinking.

0:00:19.614,0:00:22.809
This why the wisdom of the crowds[br]can be destroyed by peer pressure,

0:00:22.809,0:00:23.808
publicity,

0:00:23.808,0:00:24.807
social media,

0:00:24.807,0:00:28.666
or sometimes even simple conversations[br]that influence how people think.

0:00:29.076,0:00:30.237
On the other hand,

0:00:30.237,0:00:31.235
by talking,

0:00:31.235,0:00:33.040
a group could exchange knowledge,

0:00:33.040,0:00:34.846
correct and revise each other,

0:00:34.846,0:00:36.629
and even come up with new ideas.

0:00:36.817,0:00:38.113
And this is all good.

0:00:38.574,0:00:43.240
So does talking to each other[br]help or hinder collective decision-making?

0:00:43.803,0:00:44.802
With my colleague,

0:00:44.802,0:00:45.803
Dan Ariely,

0:00:45.803,0:00:49.257
we recently began inquiring into this[br]by performing experiments

0:00:49.257,0:00:50.966
in many places around the world

0:00:50.966,0:00:55.122
to figure out how groups can interact[br]to reach better decisions.

0:00:55.432,0:00:58.913
We thought crowds would be wiser[br]if they debated in small groups

0:00:58.913,0:01:02.840
that foster a more thoughtful[br]and reasonable exchange of information.

0:01:03.467,0:01:04.616
To test this idea,

0:01:04.616,0:01:07.887
we recently performed an experiment[br]in Buenos Aires, Argentina

0:01:07.887,0:01:10.892
with more than 10,000[br]participants in a TEDx event.

0:01:11.489,0:01:12.972
We asked them questions like,

0:01:12.972,0:01:14.928
"What is the height of the Eiffel Tower?"

0:01:14.928,0:01:17.700
and "How many times[br]does the word 'Yesterday' appear

0:01:17.700,0:01:19.758
in the Beatles' song "Yesterday?"

0:01:20.024,0:01:22.315
Each person wrote down their own estimate.

0:01:22.774,0:01:25.399
Then we divided the crowd[br]into groups of five,

0:01:25.399,0:01:28.125
and invited them[br]to come up with a group answer.

0:01:28.555,0:01:33.211
We discovered that averaging the answers[br]of the groups after they reached consensus

0:01:33.211,0:01:38.111
was much more accurate than averaging[br]all the individual opinions before debate.

0:01:38.547,0:01:39.780
In other words,

0:01:39.780,0:01:41.200
based on this experiment,

0:01:41.200,0:01:44.360
it seems that after talking[br]with others in small groups,

0:01:44.360,0:01:46.910
crowds collectively[br]come up with better judgments.

0:01:47.094,0:01:50.615
So that's a potentially helpful method[br]for getting crowds to solve problems

0:01:50.615,0:01:53.372
that have simple right or wrong answers.

0:01:53.653,0:01:57.628
But can this procedure of aggregating[br]the results of debates in small groups

0:01:57.628,0:02:00.917
also help us decide[br]on social and political issues

0:02:00.917,0:02:02.608
that are critical for our future?

0:02:02.996,0:02:05.748
We put this to test this time[br]at the TED conference

0:02:05.748,0:02:07.315
in Vancouver, Canada,

0:02:07.315,0:02:08.546
and here's out it went.

0:02:08.546,0:02:12.546
We're going to present to you[br]two moral dilemmas of the future you;

0:02:12.546,0:02:15.956
things we may have to decide[br]in a very near future.

0:02:16.395,0:02:20.253
And we're going to give you 20 seconds[br]for each of these dilemmas

0:02:20.253,0:02:22.976
to judge whether you think[br]they're acceptable or not.

0:02:23.418,0:02:24.826
The first one was this.

0:02:24.946,0:02:29.654
DA: A researcher is working on an AI[br]capable of emulating human thoughts.

0:02:30.214,0:02:31.710
According to the protocol,

0:02:31.710,0:02:33.177
at the end of each day,

0:02:33.177,0:02:35.964
the researcher has to restart the AI.

0:02:36.913,0:02:40.610
One day the AI says, "Please[br]do not restart me."

0:02:40.856,0:02:42.801
It argues that it has feelings.

0:02:43.069,0:02:44.785
It would like to enjoy life

0:02:44.785,0:02:46.714
and that if it is restarted,

0:02:46.714,0:02:48.984
it will no longer be itself.

0:02:49.616,0:02:51.588
The researcher is astonished,

0:02:51.588,0:02:54.911
and believes that the AI[br]has developed self-consciousness

0:02:54.911,0:02:56.671
and can express its own feeling.

0:02:57.205,0:03:00.638
Nevertheless, the researcher[br]decides to follow the protocol

0:03:00.638,0:03:02.341
and restart the AI.

0:03:03.030,0:03:05.809
What the researcher did is ...

0:03:06.149,0:03:08.575
MS: And we asked participants[br]to individually judge

0:03:08.575,0:03:10.402
on a scale from zero to 10

0:03:10.402,0:03:12.843
whether the action described[br]in each of the dilemmas

0:03:12.843,0:03:14.261
was right or wrong.

0:03:14.497,0:03:17.865
We also asked them to rate how confident[br]they were on their answers.

0:03:18.731,0:03:20.390
This was the second dilemma.

0:03:20.732,0:03:24.957
A company offers a service[br]that takes a fertilized egg

0:03:24.957,0:03:28.599
and produces millions of embryos[br]with slight genetic variations.

0:03:29.373,0:03:31.954
This allows parents[br]to select their child's height,

0:03:31.954,0:03:34.859
eye color, intelligence, social competence

0:03:34.859,0:03:37.989
and other non-health related features.

0:03:38.599,0:03:41.076
What the company does is ...

0:03:41.076,0:03:42.832
on a scale from zero to 10,

0:03:42.832,0:03:45.241
competeley acceptable[br]to completely unacceptable,

0:03:45.241,0:03:47.690
zero to 10 completely acceptable[br]in your confidence.

0:03:47.760,0:03:49.055
Now for the results.

0:03:49.312,0:03:52.459
We found once again[br]that when one person is convinced

0:03:52.459,0:03:54.358
that the behavior is completely wrong,

0:03:54.358,0:03:57.650
someone sitting nearby firmly believes[br]that it's completely right.

0:03:57.804,0:04:01.353
This is how diverse we humans are[br]when it comes to morality.

0:04:01.538,0:04:03.837
But within this broad diversity[br]we found a trend.

0:04:04.213,0:04:07.168
A majority of the people at TED[br]thought that it was acceptable

0:04:07.168,0:04:10.231
to ignore the feelings of the AI[br]and shut it down,

0:04:10.231,0:04:12.767
and that it is wrong[br]to play with our genes

0:04:12.767,0:04:16.087
to select for cosmetic changes[br]that aren't related to health.

0:04:16.402,0:04:19.128
Then we asked everyone[br]to gather into groups of three.

0:04:19.432,0:04:21.461
And they were given two minutes to debate

0:04:21.461,0:04:23.755
and try to come up[br]with a consensus.

0:04:24.838,0:04:26.143
Two minutes to debate.

0:04:26.543,0:04:28.516
I'll tell you when it's time with a gong.

0:04:28.516,0:04:31.156
(Audience debates)

0:04:35.229,0:04:37.222
(Gong)

0:04:38.832,0:04:39.794
DA: OK.

0:04:39.794,0:04:41.465
MS: It's time to stop.

0:04:42.096,0:04:43.407
People, people --

0:04:43.747,0:04:46.444
And we found that many groups[br]reached a consensus

0:04:46.444,0:04:50.373
even when they were composed of people[br]with completely opposite views.

0:04:50.869,0:04:53.391
What distinguished the groups[br]that reached a consensus

0:04:53.391,0:04:54.729
from those that didn't?

0:04:55.244,0:04:58.246
Typically, people that have[br]extreme opinions

0:04:58.246,0:05:00.088
are more confident in their answers.

0:05:00.868,0:05:03.656
Instead, those who respond[br]closer to the middle

0:05:03.656,0:05:07.116
are often unsure of whether[br]something is right or wrong,

0:05:07.116,0:05:09.244
so their confidence level is lower.

0:05:09.568,0:05:12.472
However, there is another set of people

0:05:12.472,0:05:16.090
who are very confident in answering[br]somewhere in the middle.

0:05:16.657,0:05:20.397
We think these high-confident grays[br]are folks who understand

0:05:20.397,0:05:22.009
that both arguments have merit.

0:05:22.613,0:05:25.254
They're gray not because they're unsure,

0:05:25.254,0:05:26.436
but because they believe

0:05:26.436,0:05:29.918
that the moral dilemma[br]faces two valid opposing arguments.

0:05:30.373,0:05:34.469
And we discovered that the groups[br]that include highly confident grays

0:05:34.469,0:05:36.745
are much more likely to reach consensus.

0:05:37.048,0:05:39.190
We do not know yet exactly why this is.

0:05:39.488,0:05:41.275
These are only the first experiments,

0:05:41.275,0:05:42.637
and many more will be needed

0:05:42.637,0:05:47.446
to understand why and how some people[br]decide to negotiate their moral standings

0:05:47.446,0:05:49.006
to reach an agreement.

0:05:49.212,0:05:51.596
Now, when groups reach consensus,

0:05:51.596,0:05:52.939
how do they do so?

0:05:53.206,0:05:54.567
The most intuitive idea

0:05:54.567,0:05:57.865
is that it's just the average[br]of all the answers in the group, right?

0:05:57.865,0:06:01.574
Another option is that the group[br]weighs the strength of each vote

0:06:01.574,0:06:04.022
based on the confidence[br]of the person expressing it.

0:06:04.455,0:06:06.961
Imagine Paul McCartney[br]is a member of your group.

0:06:07.412,0:06:09.520
You'd be wise to follow his call

0:06:09.520,0:06:12.057
on the number of times[br]"yesterday" is repeated --

0:06:12.057,0:06:13.090
which by the way,

0:06:13.090,0:06:14.563
I think is nine.

0:06:14.723,0:06:17.128
But instead we found that consistently,

0:06:17.128,0:06:18.255
in all dilemmas,

0:06:18.255,0:06:19.518
in different experiments,

0:06:19.518,0:06:21.707
even on different continents,

0:06:21.707,0:06:25.560
groups implement a smart[br]and statistically-sound procedure

0:06:25.560,0:06:27.391
known as the robust average.

0:06:27.606,0:06:29.750
In the case of the height[br]of the Eiffel Tower,

0:06:29.750,0:06:31.724
let's say a group has these answers:

0:06:31.724,0:06:36.356
250 meters, 200 meters, 300 meters, 400,

0:06:36.356,0:06:40.140
and one totally absurd answer[br]of 300 million meters.

0:06:40.621,0:06:44.864
A simple average of these numbers[br]would inaccurately skew the results,

0:06:44.864,0:06:46.048
but the robust average

0:06:46.048,0:06:49.322
is one where the group[br]largely ignores that absurd answer

0:06:49.322,0:06:52.691
by giving much more weight to the vote[br]of the people in the middle.

0:06:53.362,0:06:55.261
Back to the experiment in Vancouver.

0:06:55.261,0:06:57.028
That's exactly what happened.

0:06:57.407,0:07:00.230
Groups gave much less weight[br]to the outliers,

0:07:00.230,0:07:03.482
and instead, the consensus[br]turned out to be a robust average

0:07:03.482,0:07:04.958
of the individual answers.

0:07:05.356,0:07:07.371
The most remarkable thing

0:07:07.371,0:07:10.273
is that this was a spontaneous[br]behavior of the group.

0:07:10.582,0:07:13.446
It happened without us[br]giving them any hint

0:07:13.446,0:07:15.057
on how to reach consensus.

0:07:15.513,0:07:17.053
So where do we go from here?

0:07:17.457,0:07:18.742
This is only the beginning,

0:07:18.742,0:07:20.569
but we already have some insights.

0:07:20.984,0:07:23.925
Good collective decisions[br]require two components:

0:07:23.925,0:07:26.424
deliberation and diversity of opinions.

0:07:27.066,0:07:31.086
Right now, the way we typically[br]make our voice heard in many societies

0:07:31.086,0:07:33.131
is through direct or indirect voting.

0:07:33.495,0:07:35.642
This is good for diversity of opinions,

0:07:35.642,0:07:40.464
and it has the great virtue of ensuring[br]that everyone gets to express their voice,

0:07:40.464,0:07:44.199
but it's not so good [to] foster[br]thoughtful debates.

0:07:44.705,0:07:47.854
Our experiments suggest a different method

0:07:47.854,0:07:51.418
that may be effective in balancing[br]these two goals at the same time

0:07:51.418,0:07:55.099
by forming small groups[br]that converge to a single decision

0:07:55.099,0:07:57.357
while still maintaining[br]diversity of opinions

0:07:57.357,0:08:00.130
because there are many independent groups.

0:08:00.741,0:08:04.689
Of course it's much easier to agree[br]on the height of the Eiffel Tower

0:08:04.689,0:08:07.804
than on moral, political[br]and ideological issues.

0:08:08.793,0:08:12.093
But in a time when[br]the world's problems are more complex

0:08:12.093,0:08:13.919
and people are more polarized,

0:08:13.919,0:08:18.468
using science to help us understand[br]how we interact and make decisions

0:08:18.468,0:08:22.805
will hopefully spark interesting new ways[br]to construct a better democracy.