WEBVTT

1
00:00:00.379 --> 00:00:02.630


2
00:00:02.630 --> 00:00:04.700
Artificial intelligence can improve

3
00:00:04.700 --> 00:00:07.200
how individuals and teams communicate,

4
00:00:07.200 --> 00:00:11.650
share ideas, and work together, but it isn't easy.

5
00:00:11.650 --> 00:00:13.460
Leaders have to wrestle with questions

6
00:00:13.460 --> 00:00:17.880
like who should have access to the data our AI is gathering?

7
00:00:17.880 --> 00:00:19.390
Who should be making decisions

8
00:00:19.390 --> 00:00:21.330
about what we're seeing in the data?

9
00:00:21.330 --> 00:00:25.070
And how should we make sure our AI is working properly?

10
00:00:25.070 --> 00:00:28.230
In the age of AI, the best performing organizations

11
00:00:28.230 --> 00:00:30.800
will take a whole new approach to leadership,

12
00:00:30.800 --> 00:00:33.660
and leaders will take on a whole new role.

13
00:00:33.660 --> 00:00:35.610 line:15% 
Leadership has always been about

14
00:00:37.010 --> 00:00:39.190 line:15% 
figuring out where you should go

15
00:00:39.190 --> 00:00:43.300 line:15% 
and then getting people on the same page to go there.

16
00:00:43.300 --> 00:00:45.070
But what's happened now 

17
00:00:45.070 --> 00:00:48.250
is you need to be more engaging to people

18
00:00:48.250 --> 00:00:51.370
to draw out ideas from the people you work with.

19
00:00:51.370 --> 00:00:54.180
Not just direct reports, but the people to decide,

20
00:00:54.180 --> 00:00:57.160
the people in other parts of the organization.

21
00:00:57.160 --> 00:01:01.120
So, it's really more like you're the head sense-maker.

22
00:01:01.120 --> 00:01:05.090
You need to have a discussion where, together,

23
00:01:05.090 --> 00:01:07.920
you come up with that new direction

24
00:01:07.920 --> 00:01:09.400
and everybody understands it

25
00:01:09.400 --> 00:01:10.930
because they were part of the discussion.

26
00:01:10.930 --> 00:01:12.820
So, instead of telling people what to do,

27
00:01:12.820 --> 00:01:16.250
you're acting as a curator to help

28
00:01:16.250 --> 00:01:18.470
sort of pick out good ideas,

29
00:01:18.470 --> 00:01:22.670
make sure that there aren't objections to them.

30
00:01:22.670 --> 00:01:27.670
You're really more like a farmer in the field

31
00:01:27.870 --> 00:01:30.870
than you are the military commander.

32
00:01:30.870 --> 00:01:32.740
What that does is it puts an emphasis

33
00:01:32.740 --> 00:01:35.690
on this coalition building skills,

34
00:01:35.690 --> 00:01:38.550
knowing who might be interested, who might have an idea,

35
00:01:38.550 --> 00:01:42.030
building those as personal relationships,

36
00:01:42.030 --> 00:01:47.030
not necessarily as just business relationships.

37
00:01:47.130 --> 00:01:48.350
because people won't tell you things

38
00:01:48.350 --> 00:01:49.910
if it's just business, right?

39
00:01:49.910 --> 00:01:53.990
You need to actually sort of be a trusted person

40
00:01:53.990 --> 00:01:56.480
who can keep some things quiet and some things not.

41
00:01:56.480 --> 00:01:58.363
You have to build this network.

42
00:02:00.220 --> 00:02:02.080
To gather data with AI,

43
00:02:02.080 --> 00:02:04.070
companies will have to give up old notions

44
00:02:04.070 --> 00:02:07.100
about who owns what data and what can be done with it,

45
00:02:07.100 --> 00:02:10.230
including data about how employees are interacting,

46
00:02:10.230 --> 00:02:12.920
and they'll have to determine where the data should reside

47
00:02:12.920 --> 00:02:15.000
as well as who should have access to it

48
00:02:15.000 --> 00:02:16.310
and who should make decisions

49
00:02:16.310 --> 00:02:18.400
about what they're seeing in the data.

50
00:02:18.400 --> 00:02:21.660
You don't own data just because somebody gave it to you.

51
00:02:21.660 --> 00:02:23.180
They have to give it to you

52
00:02:23.180 --> 00:02:26.010
and have the ability to pull it back.

53
00:02:26.010 --> 00:02:27.700
They have to know what's happening to it.

54
00:02:27.700 --> 00:02:29.010
They have to approve of that.

55
00:02:29.010 --> 00:02:31.423
People have rights over data about them.

56
00:02:32.750 --> 00:02:34.920
However, if it's aggregated enough,

57
00:02:34.920 --> 00:02:37.580
so it refers to no one person,

58
00:02:37.580 --> 00:02:41.040
then that's a very strong argument that the company owns it.

59
00:02:41.040 --> 00:02:43.640
So, what we do when we help companies with this

60
00:02:43.640 --> 00:02:45.500
is we produce aggregate pictures.

61
00:02:45.500 --> 00:02:48.943
How much does this silo talk to that silo?

62
00:02:49.900 --> 00:02:51.310
It's not about any particular person.

63
00:02:51.310 --> 00:02:52.743
You can't name and shame with that data.

64
00:02:52.743 --> 00:02:54.540
It's just not possible .

65
00:02:54.540 --> 00:02:57.060
The company doesn't actually own the data.

66
00:02:57.060 --> 00:03:00.320
One of the mistakes, or misunderstandings that people do

67
00:03:00.320 --> 00:03:03.370
is that AI requires centralizing data.

68
00:03:03.370 --> 00:03:06.550
If you centralize data, you've made a huge error.

69
00:03:06.550 --> 00:03:08.130
When you put data in one spot,

70
00:03:08.130 --> 00:03:10.670
you've told the bad guys where to go.

71
00:03:10.670 --> 00:03:12.110
You've taken power away

72
00:03:12.110 --> 00:03:14.670
from the people who know the situation best.

73
00:03:14.670 --> 00:03:17.860
You've actually become very much the opposite

74
00:03:17.860 --> 00:03:20.050
of a data-driven organization,

75
00:03:20.050 --> 00:03:22.110
because all the people that are interacting

76
00:03:22.110 --> 00:03:23.850
with the customers in the situations

77
00:03:23.850 --> 00:03:26.823
are no longer empowered to do anything about it.

78
00:03:27.992 --> 00:03:30.350

79
00:03:30.350 --> 00:03:31.580
So, what do you do?

80
00:03:31.580 --> 00:03:34.130
Well, it's expensive to redo all your systems.

81
00:03:34.130 --> 00:03:36.010
Turns out you don't have to.

82
00:03:36.010 --> 00:03:38.293
The very best things don't do that.

83
00:03:38.293 --> 00:03:39.740
What they do is they set up

84
00:03:39.740 --> 00:03:41.910
what's called a federated data system.

85
00:03:41.910 --> 00:03:43.400
You have a common language

86
00:03:43.400 --> 00:03:45.110
between all the different systems.

87
00:03:45.110 --> 00:03:46.850
It's like a communication bus,

88
00:03:46.850 --> 00:03:48.560
a little bit like the internet,

89
00:03:48.560 --> 00:03:51.280
but it's an internet of your data.

90
00:03:51.280 --> 00:03:54.340
And you can make it encrypted and controlled.

91
00:03:54.340 --> 00:03:56.940
You don't have to redo all your systems.

92
00:03:56.940 --> 00:03:59.300
It's safer because the people that collect the data

93
00:03:59.300 --> 00:04:00.530
are still holding the data.

94
00:04:00.530 --> 00:04:02.430
They sort of know the nuances.

95
00:04:02.430 --> 00:04:05.140
They can answer the regulators, et cetera.

96
00:04:05.140 --> 00:04:08.860
You get to try these natural experiments.

97
00:04:08.860 --> 00:04:12.940
That's almost the definition of an innovative company.

98
00:04:12.940 --> 00:04:15.080
You have to be trying different things.

99
00:04:15.080 --> 00:04:17.380
It'll work one place, it won't work the other.

100
00:04:17.380 --> 00:04:18.930
The fact that you're not rolling it out

101
00:04:18.930 --> 00:04:23.330
across the whole company means that you can't lose big time.

102
00:04:23.330 --> 00:04:26.750
You lose this or that piece, but you also win,

103
00:04:26.750 --> 00:04:29.923
because if you see it working, you can spread it.

104
00:04:29.923 --> 00:04:32.610


105
00:04:32.610 --> 00:04:36.530
As useful as AI is, it also has limitations.

106
00:04:36.530 --> 00:04:39.170
Smart leaders will understand those limitations

107
00:04:39.170 --> 00:04:41.470
and keep a sharp eye on their AI

108
00:04:41.470 --> 00:04:44.015
to make sure it's not going off the rails.

109
00:04:44.015 --> 00:04:46.313
AI today is quite limited.

110
00:04:47.346 --> 00:04:52.346
It takes a huge amount of data from a static situation

111
00:04:52.760 --> 00:04:55.160
to be able to train these things.

112
00:04:55.160 --> 00:04:57.010
So, static means, well, you know,

113
00:04:57.010 --> 00:04:59.570
the way we speak English doesn't change very quickly,

114
00:04:59.570 --> 00:05:02.350
so you can record a million people speaking

115
00:05:02.350 --> 00:05:06.000
and build a model for it that's really bang up.

116
00:05:06.000 --> 00:05:09.530
But things like how we use Twitter change all the time.

117
00:05:09.530 --> 00:05:11.590
And so, when you build a model, you find

118
00:05:11.590 --> 00:05:13.010
oh, I can get good results,

119
00:05:13.010 --> 00:05:15.320
but then a month later you find it's not so good.

120
00:05:15.320 --> 00:05:17.840
And then three months later it's nearly hopeless.

121
00:05:17.840 --> 00:05:21.100
You can use it like a specialized hammer,

122
00:05:21.100 --> 00:05:24.840
or a specialized tool, but you can't replace things.

123
00:05:24.840 --> 00:05:27.560
There's still a person that has to sort of figure out

124
00:05:27.560 --> 00:05:29.900
whether it's doing what it ought to do.

125
00:05:29.900 --> 00:05:34.190
Like all business processes, it's not set once and forget.

126
00:05:34.190 --> 00:05:36.810
You're a fool if you don't watch your AI

127
00:05:36.810 --> 00:05:39.160
to make sure it's doing what it's supposed to,

128
00:05:39.160 --> 00:05:43.450
because human behavior changes and the AI doesn't,

129
00:05:43.450 --> 00:05:45.510
and it can run off the rails pretty easily.

130
00:05:45.510 --> 00:05:47.280
Or you can train it wrong.

131
00:05:47.280 --> 00:05:50.320
The bottom line there is you have to be able

132
00:05:50.320 --> 00:05:53.920
to continuously audit what your system is doing.

133
00:05:53.920 --> 00:05:56.240
The AI that's been most successful are things

134
00:05:56.240 --> 00:05:59.080
that help people do what they're already doing,

135
00:05:59.080 --> 00:06:01.330
not things that replace people.

136
00:06:01.330 --> 00:06:05.100
And that's actually important, because today's AI tools,

137
00:06:05.100 --> 00:06:07.420
while they're good at things,

138
00:06:07.420 --> 00:06:10.080
can run off the rails really easily.

139
00:06:10.080 --> 00:06:13.560
They can be doing stuff that's counterproductive.

140
00:06:13.560 --> 00:06:17.270
They can be unfair, or biased, or illegal,

141
00:06:17.270 --> 00:06:20.260
and unless you're watching, you won't know it.

142
00:06:20.260 --> 00:06:22.410
So, what you have to do is think of them

143
00:06:22.410 --> 00:06:23.760
as tools for humans,

144
00:06:23.760 --> 00:06:28.040
but the humans are continuously monitoring their behavior.

145
00:06:28.040 --> 00:06:29.840
Are they doing what you want to do

146
00:06:29.840 --> 00:06:32.468
or do we need to retrain them?

147
00:06:32.468 --> 00:06:36.270


148
00:06:36.270 --> 00:06:37.250
So, when it comes to leading

149
00:06:37.250 --> 00:06:40.370
in the age of AI, what are the key takeaways?

150
00:06:40.370 --> 00:06:43.140
Understand the new role that leaders have to play

151
00:06:43.140 --> 00:06:46.540
as sense-makers of data, not dictators of rules.

152
00:06:46.540 --> 00:06:50.070
Get clear about who owns data about your employees,

153
00:06:50.070 --> 00:06:51.940
and how you're allowed to use it.

154
00:06:51.940 --> 00:06:54.560
Make sure that people closest to the action

155
00:06:54.560 --> 00:06:58.060
are also closest to the data from that action.

156
00:06:58.060 --> 00:06:59.730
And watch your AI closely

157
00:06:59.730 --> 00:07:02.010
to ensure it's working as it should,

158
00:07:02.010 --> 00:07:05.113
and be ready to retrain it if it starts misbehaving.

159
00:07:07.758 --> 00:07:10.130


160
00:07:10.130 --> 00:07:11.830
These imperatives are all vital

161
00:07:11.830 --> 00:07:13.890
for leading in the age of AI,

162
00:07:13.890 --> 00:07:15.640
but they become even more crucial

163
00:07:15.640 --> 00:07:18.170
when organizations use AI to interpret

164
00:07:18.170 --> 00:07:21.110
and simulate human beings' emotions.

165
00:07:21.110 --> 00:07:23.690
Imagine an AI showing empathy to someone

166
00:07:23.690 --> 00:07:26.340
who it perceives as having a bad day,

167
00:07:26.340 --> 00:07:27.780
or cheering on a team member

168
00:07:27.780 --> 00:07:30.970
who logged a delay on a tough project.

169
00:07:30.970 --> 00:07:32.830
This is a branch of computer science

170
00:07:32.830 --> 00:07:35.960
called affective computing, and it's presenting

171
00:07:35.960 --> 00:07:39.520
valuable new opportunities for organizations.

172
00:07:39.520 --> 00:07:43.440
In our next video, Rosalind Picard, founder and director

173
00:07:43.440 --> 00:07:45.670
of the Affective Computing Research Group

174
00:07:45.670 --> 00:07:48.310
at the MIT Media Lab, discusses

175
00:07:48.310 --> 00:07:51.100
the latest developments in affective computing,

176
00:07:51.100 --> 00:07:54.625
including how it's being used in organizations.

177
00:07:54.625 --> 00:07:57.125