1
00:00:00,479 --> 00:00:03,062
(bees buzzing)

2
00:00:14,750 --> 00:00:17,150
Many things are impressive about the honeybee.

3
00:00:24,270 --> 00:00:29,270
When you work this closely, you see their intelligence,

4
00:00:29,270 --> 00:00:31,160
you see their individuality,

5
00:00:31,160 --> 00:00:33,060
you see their collective behavior,

6
00:00:33,060 --> 00:00:34,610
you see the structures they've built,

7
00:00:34,610 --> 00:00:37,030
you see the organization of that society.

8
00:00:37,030 --> 00:00:39,230
You can't do anything but admire it, you can't.

9
00:00:39,230 --> 00:00:42,280
They are the most beautiful, phenomenal creatures.

10
00:00:42,280 --> 00:00:43,368
They really are.

11
00:00:43,368 --> 00:00:46,618
(soft classical music)

12
00:00:53,340 --> 00:00:55,480
This bee has learned that if it moves

13
00:00:55,480 --> 00:00:58,430
that yellow ball into the yellow circle,

14
00:00:58,430 --> 00:01:01,130
the well beneath the ball fills up with nectar

15
00:01:01,130 --> 00:01:03,180
and it gets a drink,

16
00:01:03,180 --> 00:01:07,500
and all of that intelligence, all of that smarts,

17
00:01:07,500 --> 00:01:09,580
come somehow from the bee brain,

18
00:01:09,580 --> 00:01:12,763
and I want to understand this bee level of intelligence.

19
00:01:18,310 --> 00:01:21,300
We have a jumbo jet, we have a bumble bee,

20
00:01:21,300 --> 00:01:23,920
we have an osprey, I could not say

21
00:01:23,920 --> 00:01:27,110
which one is a better flier than the other

22
00:01:27,110 --> 00:01:28,953
because they're different.

23
00:01:30,530 --> 00:01:33,540
Putting the envelope around what intelligence is

24
00:01:33,540 --> 00:01:35,450
is extremely difficult,

25
00:01:35,450 --> 00:01:37,820
and I think what will help us frame that envelope

26
00:01:37,820 --> 00:01:41,220
is if we can study the diversity.

27
00:01:41,220 --> 00:01:43,810
If we study intelligence, not just in humans,

28
00:01:43,810 --> 00:01:48,340
but in other living things, potentially even other machines,

29
00:01:48,340 --> 00:01:52,440
we can tidy up that definition of what intelligence is

30
00:01:52,440 --> 00:01:53,730
and where we draw the boundary on

31
00:01:53,730 --> 00:01:56,030
what's intelligent and what's not.

32
00:01:56,030 --> 00:02:00,820
My project is particularly focusing on honeybee intelligence

33
00:02:00,820 --> 00:02:03,670
because it gives us such an informative lens,

34
00:02:03,670 --> 00:02:06,200
sort of informative, comparative lens,

35
00:02:06,200 --> 00:02:09,603
on the intelligence of other animals, including humans.

36
00:02:11,500 --> 00:02:14,280
Things like complex learning, complex memory,

37
00:02:14,280 --> 00:02:17,450
complex navigation, complex assessment,

38
00:02:17,450 --> 00:02:20,910
we'll learn some evolved solutions for that,

39
00:02:20,910 --> 00:02:22,930
and we can then ask is the human brain

40
00:02:22,930 --> 00:02:24,330
doing this in a similar way.

41
00:02:28,400 --> 00:02:33,400
We have these tiny little animals with really minute brains.

42
00:02:34,060 --> 00:02:35,770
They have a million neurons.

43
00:02:35,770 --> 00:02:38,273
It's minute compared to a human brain.

44
00:02:38,273 --> 00:02:41,177
(soft classical music)

45
00:02:41,177 --> 00:02:43,850
The honeybee brain is very small,

46
00:02:43,850 --> 00:02:45,590
but it would be wrong to characterize it

47
00:02:45,590 --> 00:02:47,440
as a simple system.

48
00:02:47,440 --> 00:02:51,310
We do still have 1 million neurons in a bee brain,

49
00:02:51,310 --> 00:02:55,240
and they are organized in quite beautiful

50
00:02:55,240 --> 00:02:57,500
different structural regions that interact and

51
00:02:57,500 --> 00:03:00,940
intersect in very complex ways.

52
00:03:00,940 --> 00:03:03,010
People normally think they're very clever as groups

53
00:03:03,010 --> 00:03:05,590
but simply rather stupid individually,

54
00:03:05,590 --> 00:03:08,260
and nothing could be further from the truth.

55
00:03:08,260 --> 00:03:10,110
Honeybees have been documented to find

56
00:03:10,110 --> 00:03:12,170
their way home from 12 kilometers away.

57
00:03:12,170 --> 00:03:13,940
In a routine foraging flight,

58
00:03:13,940 --> 00:03:16,280
bees will fly 5 or 6 kilometers,

59
00:03:16,280 --> 00:03:19,560
which doesn't sound much, but when you scale that by

60
00:03:19,560 --> 00:03:20,930
the size of an individual bee,

61
00:03:20,930 --> 00:03:23,000
that's a really huge distance.

62
00:03:23,000 --> 00:03:25,490
Our own machine learning and AR algorithms

63
00:03:25,490 --> 00:03:28,140
for navigation aren't that sophisticated or reliable.

64
00:03:29,130 --> 00:03:31,890
But what stands out as a unique feature of the honeybee

65
00:03:31,890 --> 00:03:34,500
would have to be its symbolic dance language.

66
00:03:34,500 --> 00:03:37,520
When they dance, the vigor with which they shake their butt

67
00:03:37,520 --> 00:03:41,200
and how many times they dance is the quality of

68
00:03:41,200 --> 00:03:42,700
the sugar reward they have found.

69
00:03:42,700 --> 00:03:47,070
They are transforming information about

70
00:03:47,070 --> 00:03:50,230
distance and direction to things in the real world,

71
00:03:50,230 --> 00:03:52,170
to these remote food sources,

72
00:03:52,170 --> 00:03:55,150
into a single vector that they can then

73
00:03:55,150 --> 00:03:57,590
signal through a dance,

74
00:03:57,590 --> 00:03:59,790
so the dance is a readout of this

75
00:03:59,790 --> 00:04:02,700
subjective evaluation of how good

76
00:04:02,700 --> 00:04:04,170
that reward was for the bee.

77
00:04:04,170 --> 00:04:06,530
It's the tail wag for a bee.

78
00:04:06,530 --> 00:04:09,090
For me, the bee was in this unique position

79
00:04:09,090 --> 00:04:13,160
where its behavior was complex enough to be interesting,

80
00:04:13,160 --> 00:04:16,280
but its newer biology in its brain was simple enough

81
00:04:16,280 --> 00:04:17,430
that we could study it.

82
00:04:20,920 --> 00:04:23,473
The honeybees really are spectacular learners.

83
00:04:24,409 --> 00:04:27,710
They learn very fast and very robustly.

84
00:04:27,710 --> 00:04:29,340
As an example, if we give a honeybee

85
00:04:29,340 --> 00:04:30,520
something simple to learn,

86
00:04:30,520 --> 00:04:33,260
like this odor is associated with nectar,

87
00:04:33,260 --> 00:04:35,120
this odor's where you find nectar,

88
00:04:35,120 --> 00:04:37,290
it will learn that on one trial.

89
00:04:37,290 --> 00:04:38,700
If you give it three trials,

90
00:04:38,700 --> 00:04:41,830
it will learn that for the rest of its lifetime,

91
00:04:41,830 --> 00:04:44,180
so that's very fast acquisition of

92
00:04:44,180 --> 00:04:46,480
relationships between information.

93
00:04:46,480 --> 00:04:48,610
They can even learn things that we would consider

94
00:04:48,610 --> 00:04:50,130
to be abstract concepts,

95
00:04:50,130 --> 00:04:52,320
things that we would call learning of sameness,

96
00:04:52,320 --> 00:04:53,350
learning of difference.

97
00:04:53,350 --> 00:04:54,840
Honeybees able to do that.

98
00:04:54,840 --> 00:04:56,310
That hasn't been shown in

99
00:04:57,430 --> 00:05:00,170
any other invertebrate that I know of.

100
00:05:00,170 --> 00:05:04,870
A statement, I don't know, is an example of metacognition.

101
00:05:04,870 --> 00:05:07,380
You're assessing a circumstance,

102
00:05:07,380 --> 00:05:09,770
and you're coming to the conclusion that

103
00:05:09,770 --> 00:05:12,400
you don't have enough information to address that,

104
00:05:12,400 --> 00:05:13,443
or to answer that.

105
00:05:22,060 --> 00:05:24,510
If we'd look comparatively across the literature,

106
00:05:25,970 --> 00:05:29,320
in many tests, even these tests of very simple learning,

107
00:05:29,320 --> 00:05:31,400
or even tests of very complex learning,

108
00:05:31,400 --> 00:05:34,170
we see the bees learning faster than rats.

109
00:05:34,170 --> 00:05:37,630
I don't have an answer for you as to why that is yet.

110
00:05:37,630 --> 00:05:39,200
It fascinates me.

111
00:05:39,200 --> 00:05:41,950
We have an organism that where our assumption is,

112
00:05:41,950 --> 00:05:45,900
this is smarter, and yet in a whole battery of tests,

113
00:05:45,900 --> 00:05:49,310
at learning, tests of memory, tests of spatial cognition,

114
00:05:49,310 --> 00:05:51,680
the bees are outperforming the rats.

115
00:05:51,680 --> 00:05:54,100
If the bee is solving a task that

116
00:05:54,100 --> 00:05:57,030
we think demonstrates metacognition,

117
00:05:57,030 --> 00:06:01,620
how can an animal with just one million neurons do that?

118
00:06:01,620 --> 00:06:04,110
It forces us to rethink our assumptions.

119
00:06:04,110 --> 00:06:07,400
What is the minimal computational architecture

120
00:06:07,400 --> 00:06:08,963
that could do this.

121
00:06:13,840 --> 00:06:17,250
A computational model is, it's building

122
00:06:17,250 --> 00:06:20,190
a circuit diagram of the brain in a virtual world,

123
00:06:20,190 --> 00:06:22,640
and we can then make it a dynamic system

124
00:06:22,640 --> 00:06:25,414
that we can feed input to.

125
00:06:25,414 --> 00:06:28,747
(soft electronic music)

126
00:06:30,610 --> 00:06:32,490
It will process the input in the way

127
00:06:32,490 --> 00:06:35,940
that we think that the honeybee brain is processing it,

128
00:06:35,940 --> 00:06:37,750
and it will give us an output.

129
00:06:37,750 --> 00:06:39,780
We can analyze that output in terms of,

130
00:06:39,780 --> 00:06:42,960
well, is this system doing what the bee's doing?

131
00:06:42,960 --> 00:06:46,920
If it is, maybe our model is close to reality.

132
00:06:46,920 --> 00:06:50,300
We can do exactly the same with bits of mammalian brain,

133
00:06:50,300 --> 00:06:52,310
and that means that we can actually compare

134
00:06:52,310 --> 00:06:55,650
what are superficially very, very different-looking systems.

135
00:06:55,650 --> 00:06:57,790
We've done something that no one else has done,

136
00:06:57,790 --> 00:07:00,460
in that we've taken an abstract concept,

137
00:07:00,460 --> 00:07:04,750
and we have given you a neuron-by-neuron connected circuit.

138
00:07:04,750 --> 00:07:07,120
If we can model the bee brain,

139
00:07:07,120 --> 00:07:10,400
we can take insights from those models

140
00:07:10,400 --> 00:07:13,653
and translate them directly into technological applications.

141
00:07:24,180 --> 00:07:27,270
We're building drones that can fly in

142
00:07:27,270 --> 00:07:28,370
a comparable way to a bee,

143
00:07:28,370 --> 00:07:30,200
but not exactly the same as a bee.

144
00:07:30,200 --> 00:07:32,110
You know, with only a million neurons

145
00:07:32,110 --> 00:07:34,770
in the bee brain, they were already well in advance

146
00:07:34,770 --> 00:07:36,710
of our own abilities in artificial intelligence

147
00:07:36,710 --> 00:07:39,840
and robotics, so really what we'd like to do is

148
00:07:39,840 --> 00:07:43,090
try and make silicon versions of bee brains,

149
00:07:43,090 --> 00:07:45,300
or at least of the aspects of the bee brains

150
00:07:45,300 --> 00:07:48,720
that generate behavior we find useful for our own robots,

151
00:07:48,720 --> 00:07:50,750
so especially around navigation.

152
00:07:50,750 --> 00:07:52,603
I thought if we could just reverse engineer the bee brain,

153
00:07:52,603 --> 00:07:53,980
that could actually try

154
00:07:53,980 --> 00:07:56,163
and really advance the state-of-the-art.

155
00:07:58,050 --> 00:08:00,200
Bees have evolved for millions of years

156
00:08:00,200 --> 00:08:04,180
to be fantastic autonomous behavioral control systems.

157
00:08:04,180 --> 00:08:07,370
They're really robust, they're really reliable,

158
00:08:07,370 --> 00:08:11,450
they're amazing navigators across very large distances.

159
00:08:11,450 --> 00:08:15,750
All of these are current challenges in autonomous robotics,

160
00:08:15,750 --> 00:08:16,960
and yet the bee's doing it

161
00:08:16,960 --> 00:08:19,623
with incredible computational efficiency.

162
00:08:22,230 --> 00:08:23,880
In particular,

163
00:08:23,880 --> 00:08:25,170
we want to

164
00:08:26,010 --> 00:08:27,750
be able to reproduce, for example,

165
00:08:27,750 --> 00:08:30,480
the collision avoidance or navigation dependencies

166
00:08:30,480 --> 00:08:32,550
would be in robot form.

167
00:08:32,550 --> 00:08:33,920
Let's imagine autonomous

168
00:08:33,920 --> 00:08:36,480
drones that we could use in exploration.

169
00:08:36,480 --> 00:08:38,773
or agriculture, or in mining.

170
00:08:40,980 --> 00:08:43,190
At least eight people have been killed

171
00:08:43,190 --> 00:08:47,540
after a magnitude 6.1 earthquake struck the Philippines.

172
00:08:47,540 --> 00:08:50,090
For example, trying to deploy drones

173
00:08:50,090 --> 00:08:52,450
to search for survivors of an earthquake

174
00:08:52,450 --> 00:08:53,700
or something like that.

175
00:08:53,700 --> 00:08:54,980
Time is gonna be of the essence.

176
00:08:54,980 --> 00:08:58,250
You want to automate as much of the process as possible,

177
00:08:58,250 --> 00:09:01,210
have fully autonomous flight and navigation for the robots,

178
00:09:01,210 --> 00:09:04,310
then we could have some real benefits

179
00:09:04,310 --> 00:09:06,410
in that kind of technology.

180
00:09:06,410 --> 00:09:10,670
That would be the Holy Grail for so much robotics.

181
00:09:10,670 --> 00:09:13,880
Bees have solved that with this minute brain.

182
00:09:13,880 --> 00:09:15,810
We're finding that actually

183
00:09:15,810 --> 00:09:18,000
bee navigation may be a lot more map-like

184
00:09:18,000 --> 00:09:20,470
than people have previously assumed.

185
00:09:20,470 --> 00:09:21,930
I mean, the idea of a mental map is

186
00:09:21,930 --> 00:09:24,290
that you have kind of representation of the

187
00:09:24,290 --> 00:09:26,490
relationship between points in space.

188
00:09:26,490 --> 00:09:28,400
That seems like a much higher level kind

189
00:09:28,400 --> 00:09:29,880
of cognitive ability

190
00:09:29,880 --> 00:09:31,590
than people have typically assumed bees

191
00:09:31,590 --> 00:09:34,333
and other insects are able to employ.

192
00:09:35,410 --> 00:09:38,040
We've been looking at an algorithm

193
00:09:38,040 --> 00:09:40,270
inspired by how the honeybee brain works,

194
00:09:40,270 --> 00:09:41,740
what's called an optic flow estimator,

195
00:09:41,740 --> 00:09:43,560
which basically tells you how fast things

196
00:09:43,560 --> 00:09:46,120
are moving across the visual field, and you can use that.

197
00:09:46,120 --> 00:09:48,160
You know, as you will have seen from looking out

198
00:09:48,160 --> 00:09:50,140
of the window on a train, for example,

199
00:09:50,140 --> 00:09:52,450
when things are close to you, they move much faster,

200
00:09:52,450 --> 00:09:53,860
apparently, across your visual field,

201
00:09:53,860 --> 00:09:56,170
and you can use that as depth information,

202
00:09:56,170 --> 00:09:57,627
of depth cue, or information that

203
00:09:57,627 --> 00:10:00,060
you're about to crash into something,

204
00:10:00,060 --> 00:10:01,340
but you could also use it for a variety

205
00:10:01,340 --> 00:10:02,300
of other applications,

206
00:10:02,300 --> 00:10:05,180
like using it to estimate how far you've traveled,

207
00:10:05,180 --> 00:10:06,570
or how fast you're traveling.

208
00:10:06,570 --> 00:10:09,430
and again, these are tremendously useful for navigation

209
00:10:09,430 --> 00:10:11,530
and for flight control, flight regulation.

210
00:10:16,586 --> 00:10:17,520
Whether we like it or not,

211
00:10:17,520 --> 00:10:19,510
we're in this robotic revolution.

212
00:10:19,510 --> 00:10:23,110
It's happening, it will only accelerate even further.

213
00:10:23,110 --> 00:10:26,740
What interests me is the capacity for safe robotics.

214
00:10:26,740 --> 00:10:29,370
If we're gonna have a system that is trustworthy,

215
00:10:29,370 --> 00:10:31,750
we need to understand how that system works

216
00:10:31,750 --> 00:10:34,130
very, very, very deeply.

217
00:10:34,130 --> 00:10:36,300
If we're starting our robotic systems

218
00:10:36,300 --> 00:10:38,710
in the basis of a deeply understood system

219
00:10:38,710 --> 00:10:41,970
like the bee brain, to me, we have a system that

220
00:10:41,970 --> 00:10:44,920
is more intrinsically understood,

221
00:10:44,920 --> 00:10:45,830
and I think, therefore,

222
00:10:45,830 --> 00:10:49,300
potentially safer and more trustworthy

223
00:10:49,300 --> 00:10:51,829
than some of the current approaches in robotics.

224
00:10:51,829 --> 00:10:54,412
(bees buzzing)

225
00:10:58,460 --> 00:11:00,240
I suspect I'm not alone in saying this,

226
00:11:00,240 --> 00:11:01,697
but I think that in the arc of understanding of

227
00:11:01,697 --> 00:11:04,563
the bee brain, we're at the most exciting point.

228
00:11:05,660 --> 00:11:08,650
We're really getting to the point where we can put,

229
00:11:08,650 --> 00:11:11,140
not just the bee brain, but insect brains together

230
00:11:11,140 --> 00:11:13,423
as an information flow system.

231
00:11:14,450 --> 00:11:16,840
Just being able to translate what we've learned

232
00:11:16,840 --> 00:11:19,280
from the bee as a hypothesis to

233
00:11:19,280 --> 00:11:22,900
help us analyze a human brain and mammalian brains,

234
00:11:22,900 --> 00:11:25,350
that's the value of the work I'm doing with bees.

235
00:11:28,900 --> 00:11:30,590
We all have an attachment to cats and dogs

236
00:11:30,590 --> 00:11:33,940
because they're so naturally empathic.

237
00:11:33,940 --> 00:11:36,630
When you look at a bee's face, it gives nothing away.

238
00:11:36,630 --> 00:11:37,970
It gives you nothing.

239
00:11:37,970 --> 00:11:39,723
It's face is a blank mask.

240
00:11:41,210 --> 00:11:44,380
I have as warm a relationship with bees

241
00:11:44,380 --> 00:11:47,970
because I developed so much respect for them.

242
00:11:47,970 --> 00:11:50,230
When I work with bees, usually I'm working

243
00:11:50,230 --> 00:11:51,510
with just one individual bee,

244
00:11:51,510 --> 00:11:54,870
who I've paint marked or number marked so I know who she is.

245
00:11:54,870 --> 00:11:57,080
In the course of that day,

246
00:11:57,080 --> 00:12:00,130
you get this really privileged insight into

247
00:12:00,130 --> 00:12:04,230
the kind of intelligence that this animal has,

248
00:12:04,230 --> 00:12:07,220
and you realize how astonishing it is,

249
00:12:07,220 --> 00:12:09,960
and what a cognitive, and elegant,

250
00:12:09,960 --> 00:12:12,683
and beautiful entity this animal is.

251
00:12:14,281 --> 00:12:17,614
(soft classical music)