0:00:01.179,0:00:02.808
I study ants,

0:00:02.808,0:00:06.125
in the desert, in the tropical forest,

0:00:06.125,0:00:07.813
and in my kitchen,

0:00:07.813,0:00:11.673
and in the hills around Silicon Valley where I live.

0:00:11.673,0:00:13.469
I've recently realized that ants

0:00:13.469,0:00:15.687
are using interactions differently

0:00:15.687,0:00:17.515
in different environments,

0:00:17.515,0:00:19.345
and that got me thinking[br]that we could learn from this

0:00:19.345,0:00:21.190
about other systems,

0:00:21.190,0:00:26.678
like brains and data networks that we engineer,

0:00:26.678,0:00:29.414
and even cancer.

0:00:29.414,0:00:31.236
So what all these systems have in common

0:00:31.236,0:00:34.103
is that there's no central control.

0:00:34.103,0:00:37.603
An ant colony consists of sterile female workers

0:00:37.603,0:00:40.359
—those are the ants you see walking around—

0:00:40.359,0:00:42.305
and then one or more reproductive females

0:00:42.305,0:00:44.408
who just lay the eggs.

0:00:44.408,0:00:46.497
They don't give any instructions.

0:00:46.497,0:00:48.401
Even though they're called queens,

0:00:48.401,0:00:50.686
they don't tell anybody what to do.

0:00:50.686,0:00:53.912
So in an ant colony, there's no one in charge,

0:00:53.912,0:00:56.770
and all systems like this without central control

0:00:56.770,0:01:00.969
are regulated using very simple interactions.

0:01:00.969,0:01:03.796
Ants interact using smell.

0:01:03.796,0:01:05.603
They smell with their antennae,

0:01:05.603,0:01:08.224
and they interact with their antennae,

0:01:08.224,0:01:11.233
so when one ant touches another with its antennae,

0:01:11.233,0:01:13.005
it can tell, for example, if the other ant

0:01:13.005,0:01:14.644
is a nestmate

0:01:14.644,0:01:19.000
and what task that other ant has been doing.

0:01:19.000,0:01:22.091
So here you see a lot of ants moving around

0:01:22.091,0:01:24.183
and interacting in a lab arena

0:01:24.183,0:01:27.451
that's connected by tubes to two other arenas.

0:01:27.451,0:01:29.785
So when one ant meets another,

0:01:29.785,0:01:32.214
it doesn't matter which ant it meets,

0:01:32.214,0:01:33.869
and they're actually not transmitting

0:01:33.869,0:01:37.424
any kind of complicated signal or message.

0:01:37.424,0:01:39.335
All that matters to the ant is the rate

0:01:39.335,0:01:42.156
at which it meets other ants.

0:01:42.156,0:01:45.253
And all of these interactions, taken together,

0:01:45.253,0:01:47.898
produce a network.

0:01:47.898,0:01:50.338
So this is the network of the ants

0:01:50.338,0:01:52.632
that you just saw moving around in the arena,

0:01:52.632,0:01:55.829
and it's this constantly shifting network

0:01:55.829,0:01:58.349
that produces the behavior of the colony,

0:01:58.349,0:02:01.150
like whether all the ants are hiding inside the nest,

0:02:01.150,0:02:03.804
or how many are going out to forage.

0:02:03.804,0:02:05.683
A brain actually works in the same way,

0:02:05.683,0:02:07.323
but what's great about ants is,

0:02:07.323,0:02:12.140
you can see the whole network as it happens.

0:02:12.140,0:02:15.099
There are more than 12,000 species of ants,

0:02:15.099,0:02:17.335
in every conceivable environment,

0:02:17.335,0:02:19.654
and they're using interactions differently

0:02:19.654,0:02:22.782
to meet different environmental challenges.

0:02:22.782,0:02:24.727
So one important environmental challenge

0:02:24.727,0:02:26.787
that every system has to deal with

0:02:26.787,0:02:28.950
is operating cost, just what it takes

0:02:28.950,0:02:31.285
to run the system.

0:02:31.285,0:02:33.693
And another environmental challenge is resources,

0:02:33.693,0:02:36.169
finding them and collecting them.

0:02:36.169,0:02:38.612
In the desert, operating costs are high

0:02:38.612,0:02:40.725
because water is scarce,

0:02:40.725,0:02:42.884
and the seed-eating ants that I study in the desert

0:02:42.884,0:02:45.944
have to spend water to get water.

0:02:45.944,0:02:47.923
So an ant outside foraging,

0:02:47.923,0:02:50.094
searching for seeds in the hot sun,

0:02:50.094,0:02:52.367
just loses water into the air.

0:02:52.367,0:02:53.761
But the colony gets its water

0:02:53.761,0:02:55.607
by metabolizing the fats out of the seeds

0:02:55.607,0:02:57.261
that they eat.

0:02:57.261,0:03:00.490
So in this environment, interactions are used

0:03:00.490,0:03:02.569
to activate foraging.

0:03:02.569,0:03:04.403
An outgoing forager doesn't go out unless

0:03:04.403,0:03:07.106
it gets enough interactions with returning foragers,

0:03:07.106,0:03:08.866
and what you see are the returning foragers

0:03:08.866,0:03:11.019
going into the tunnel, into the nest,

0:03:11.019,0:03:13.716
and meeting outgoing foragers on their way out.

0:03:13.716,0:03:15.169
This makes sense for the ant colony,

0:03:15.169,0:03:17.162
because the more food there is out there,

0:03:17.162,0:03:19.253
the more quickly the foragers find it,

0:03:19.253,0:03:20.659
the faster they come back,

0:03:20.659,0:03:23.480
and the more foragers they send out.

0:03:23.480,0:03:26.137
The system works to stay stopped,

0:03:26.137,0:03:27.792
unless something positive happens.

0:03:27.792,0:03:30.673
So interactions function to activate foragers.

0:03:30.673,0:03:34.091
And we've been studying[br]the evolution of this system.

0:03:34.091,0:03:36.171
First of all, there's variation.

0:03:36.171,0:03:38.059
It turns out that colonies are different.

0:03:38.059,0:03:40.906
On dry days, some colonies forage less,

0:03:40.906,0:03:42.571
so colonies are different in how

0:03:42.571,0:03:44.145
they manage this tradeoff

0:03:44.145,0:03:46.763
between spending water to search for seeds

0:03:46.763,0:03:49.523
and getting water back in the form of seeds.

0:03:49.523,0:03:51.802
And we're trying to understand why

0:03:51.802,0:03:53.942
some colonies forage less than others

0:03:53.942,0:03:56.537
by thinking about ants as neurons,

0:03:56.537,0:03:58.815
using models from neuroscience.

0:03:58.815,0:04:01.279
So just as a neuron adds up its stimulation

0:04:01.279,0:04:03.515
from other neurons to decide whether to fire,

0:04:03.515,0:04:06.093
an ant adds up its stimulation from other ants

0:04:06.093,0:04:08.433
to decide whether to forage.

0:04:08.433,0:04:10.030
And what we're looking for is whether there might be

0:04:10.030,0:04:11.672
small differences among colonies

0:04:11.672,0:04:15.136
in how many interactions each ant needs

0:04:15.136,0:04:17.252
before it's willing to go out and forage,

0:04:17.252,0:04:20.829
because a colony like that would forage less.

0:04:20.829,0:04:24.148
And this raises an analogous question about brains.

0:04:24.148,0:04:25.520
We talk about the brain,

0:04:25.520,0:04:28.064
but of course every brain is slightly different,

0:04:28.064,0:04:29.649
and maybe there are some individuals

0:04:29.649,0:04:31.148
or some conditions

0:04:31.148,0:04:34.092
in which the electrical properties of neurons are such

0:04:34.092,0:04:38.006
that they require more stimulus to fire,

0:04:38.006,0:04:42.276
and that would lead to differences in brain function.

0:04:42.276,0:04:44.025
So in order to ask evolutionary questions,

0:04:44.025,0:04:46.839
we need to know about reproductive success.

0:04:46.839,0:04:49.124
This is a map of the study site

0:04:49.124,0:04:51.417
where I have been tracking this population

0:04:51.417,0:04:55.062
of harvester ant colonies for 28 years,

0:04:55.062,0:04:57.402
which is about as long as a colony lives.

0:04:57.402,0:04:59.371
Each symbol is a colony,

0:04:59.371,0:05:02.597
and the size of the symbol is[br]how many offspring it had,

0:05:02.597,0:05:04.510
because we were able to use genetic variation

0:05:04.510,0:05:07.481
to match up parent and offspring colonies,

0:05:07.481,0:05:10.518
that is, to figure out which colonies

0:05:10.518,0:05:12.443
were founded by a daughter queen

0:05:12.443,0:05:14.511
produced by which parent colony.

0:05:14.511,0:05:16.513
And this was amazing for me, after all these years,

0:05:16.513,0:05:19.641
to find out, for example, that colony 154,

0:05:19.641,0:05:21.947
whom I've known well for many years,

0:05:21.947,0:05:23.996
is a great grandmother.

0:05:23.996,0:05:25.636
Here's her daughter colony,

0:05:25.636,0:05:28.090
here's her granddaughter colony,

0:05:28.090,0:05:30.822
and these are her great granddaughter colonies.

0:05:30.822,0:05:32.477
And by doing this, I was able to learn

0:05:32.477,0:05:35.076
that offspring colonies resemble parent colonies

0:05:35.076,0:05:37.977
in their decisions about which days are so hot

0:05:37.977,0:05:39.755
that they don't forage,

0:05:39.755,0:05:41.219
and the offspring of parent colonies

0:05:41.219,0:05:44.425
live so far from each other that the ants never meet,

0:05:44.425,0:05:46.414
so the ants of the offspring colony

0:05:46.414,0:05:48.868
can't be learning this from the parent colony.

0:05:48.868,0:05:50.859
And so our next step is to look

0:05:50.859,0:05:55.605
for the genetic variation[br]underlying this resemblance.

0:05:55.605,0:05:59.240
So then I was able to ask, okay, who's doing better?

0:05:59.240,0:06:01.073
Over the time of the study,

0:06:01.073,0:06:02.705
and especially in the past 10 years,

0:06:02.705,0:06:05.123
there's been a very severe and deepening drought

0:06:05.123,0:06:08.007
in the Southwestern U.S.,

0:06:08.007,0:06:11.243
and it turns out that the[br]colonies that conserve water,

0:06:11.243,0:06:15.282
that stay in when it's really hot outside,

0:06:15.282,0:06:18.284
and thus sacrifice getting as much food as possible,

0:06:18.284,0:06:21.209
are the ones more likely to have offspring colonies.

0:06:21.209,0:06:23.338
So all this time, I thought that colony 154

0:06:23.338,0:06:25.798
was a loser, because on really dry days,

0:06:25.798,0:06:27.926
there'd be just this trickle of foraging,

0:06:27.926,0:06:29.557
while the other colonies were out

0:06:29.557,0:06:31.504
foraging, getting lots of food,

0:06:31.504,0:06:34.744
but in fact, colony 154 is a huge success.

0:06:34.744,0:06:36.146
She's a matriarch.

0:06:36.146,0:06:39.288
She's one of the rare great[br]grandmothers on the site.

0:06:39.288,0:06:41.513
To my knowledge, this is the first time

0:06:41.513,0:06:43.472
that we've been able to track

0:06:43.472,0:06:46.443
the ongoing evolution of collective behavior

0:06:46.443,0:06:48.400
in a natural population of animals

0:06:48.400,0:06:52.867
and find out what's actually working best.

0:06:52.867,0:06:55.398
Now the internet uses an algorithm

0:06:55.398,0:06:58.251
to regulate the flow of data

0:06:58.251,0:07:00.178
that's very similar to the one

0:07:00.178,0:07:02.854
that the harvester ants are using to regulate

0:07:02.854,0:07:04.525
the flow of foragers.

0:07:04.525,0:07:07.231
And guess what we call this analogy?

0:07:07.231,0:07:09.279
The anternet is coming.

0:07:09.279,0:07:12.091
(Applause)

0:07:12.091,0:07:14.454
So data doesn't leave the source computer

0:07:14.454,0:07:16.995
unless it gets a signal that there's enough bandwidth

0:07:16.995,0:07:19.784
for it to travel on.

0:07:19.784,0:07:21.575
In the early days of the internet,

0:07:21.575,0:07:23.959
when operating costs were really high

0:07:23.959,0:07:27.176
and it was really important not to lose any data,

0:07:27.176,0:07:29.273
then the system was set up for interactions

0:07:29.273,0:07:32.395
to activate the flow of data.

0:07:32.395,0:07:34.770
It's interesting that the ants are using an algorithm

0:07:34.770,0:07:38.606
that's so similar to the one that we recently invented,

0:07:38.606,0:07:41.599
but this is only one of a handful of ant algorithms

0:07:41.599,0:07:43.128
that we know about,

0:07:43.128,0:07:45.931
and ants have had 130 million years

0:07:45.931,0:07:48.257
to evolve a lot of good ones,

0:07:48.257,0:07:49.983
and I think it's very likely

0:07:49.983,0:07:52.140
that some of the other 12,000 species

0:07:52.140,0:07:54.837
are going to have interesting algorithms

0:07:54.837,0:07:56.211
for data networks

0:07:56.211,0:07:58.888
that we haven't even thought of yet.

0:07:58.888,0:08:01.903
So what happens when operating costs are low?

0:08:01.903,0:08:03.770
Operating costs are low in the tropics,

0:08:03.770,0:08:05.526
because it's very humid, and it's easy for the ants

0:08:05.526,0:08:08.350
to be outside walking around.

0:08:08.350,0:08:10.193
But the ants are so abundant

0:08:10.193,0:08:11.971
and diverse in the tropics

0:08:11.971,0:08:13.999
that there's a lot of competition.

0:08:13.999,0:08:16.371
Whatever resource one species is using,

0:08:16.371,0:08:19.543
another species is likely to be using that

0:08:19.543,0:08:22.152
at the same time.

0:08:22.152,0:08:24.053
So in this environment, interactions are used

0:08:24.053,0:08:26.135
in the opposite way.

0:08:26.135,0:08:28.023
The system keeps going

0:08:28.023,0:08:29.754
unless something negative happens,

0:08:29.754,0:08:31.591
and one species that I study makes circuits

0:08:31.591,0:08:33.750
in the trees of foraging ants

0:08:33.750,0:08:36.731
going from the nest to a food source and back,

0:08:36.731,0:08:38.420
just round and round,

0:08:38.420,0:08:40.172
unless something negative happens,

0:08:40.172,0:08:41.781
like an interaction

0:08:41.781,0:08:44.100
with ants of another species.

0:08:44.100,0:08:47.327
So here's an example of ant security.

0:08:47.327,0:08:48.825
In the middle, there's an ant

0:08:48.825,0:08:51.378
plugging the nest entrance with its head

0:08:51.378,0:08:54.101
in response to interactions with another species.

0:08:54.101,0:08:56.035
Those are the little ones running around

0:08:56.035,0:08:58.511
with their abdomens up in the air.

0:08:58.511,0:09:00.557
But as soon as the threat is passed,

0:09:00.557,0:09:02.762
the entrance is open again,

0:09:02.762,0:09:05.102
and maybe there are situations

0:09:05.102,0:09:06.471
in computer security

0:09:06.471,0:09:08.397
where operating costs are low enough

0:09:08.397,0:09:11.772
that we could just block access temporarily

0:09:11.772,0:09:13.965
in response to an immediate threat,

0:09:13.965,0:09:15.991
and then open it again,

0:09:15.991,0:09:17.610
instead of trying to build

0:09:17.610,0:09:21.630
a permanent firewall or fortress.

0:09:21.630,0:09:23.540
So another environmental challenge

0:09:23.540,0:09:25.125
that all systems have to deal with

0:09:25.125,0:09:30.327
is resources, finding and collecting them.

0:09:30.327,0:09:32.007
And to do this, ants solve the problem

0:09:32.007,0:09:33.548
of collective search,

0:09:33.548,0:09:35.144
and this is a problem that's of great interest

0:09:35.144,0:09:36.698
right now in robotics,

0:09:36.698,0:09:38.340
because we've understood that,

0:09:38.340,0:09:39.994
rather than sending a single,

0:09:39.994,0:09:42.749
sophisticated, expensive robot out

0:09:42.749,0:09:44.886
to explore another planet

0:09:44.886,0:09:47.113
or to search a burning building,

0:09:47.113,0:09:49.743
that instead, it may be more effective

0:09:49.743,0:09:54.005
to get a group of cheaper robots

0:09:54.005,0:09:56.845
exchanging only minimal information,

0:09:56.845,0:09:59.774
and that's the way that ants do it.

0:09:59.774,0:10:01.729
So the invasive Argentine ant

0:10:01.729,0:10:03.559
makes expandable search networks.

0:10:03.559,0:10:05.842
They're good at dealing with the main problem

0:10:05.842,0:10:07.573
of collective search,

0:10:07.573,0:10:09.779
which is the tradeoff between

0:10:09.779,0:10:11.445
searching very thoroughly

0:10:11.445,0:10:13.322
and covering a lot of ground.

0:10:13.322,0:10:14.977
And what they do is,

0:10:14.977,0:10:16.564
when there are many ants in a small space,

0:10:16.564,0:10:18.837
then each one can search very thoroughly

0:10:18.837,0:10:20.501
because there will be another ant nearby

0:10:20.501,0:10:21.818
searching over there,

0:10:21.818,0:10:23.475
but when there are a few ants

0:10:23.475,0:10:25.180
in a large space,

0:10:25.180,0:10:27.804
then they need to stretch out their paths

0:10:27.804,0:10:29.757
to cover more ground.

0:10:29.757,0:10:32.521
I think they use interactions to assess density,

0:10:32.521,0:10:34.045
so when they're really crowded,

0:10:34.045,0:10:35.112
they meet more often,

0:10:35.112,0:10:37.207
and they search more thoroughly.

0:10:37.207,0:10:40.827
Different ant species must use different algorithms,

0:10:40.827,0:10:42.789
because they've evolved to deal with

0:10:42.789,0:10:45.150
different resources,

0:10:45.150,0:10:47.219
and it could be really useful to know about this,

0:10:47.219,0:10:49.221
and so we recently asked ants

0:10:49.221,0:10:51.561
to solve the collective search problem

0:10:51.561,0:10:53.269
in the extreme environment

0:10:53.269,0:10:54.737
of microgravity

0:10:54.737,0:10:56.813
in the International Space Station.

0:10:56.813,0:10:58.218
When I first saw this picture, I thought,

0:10:58.218,0:11:00.755
Oh no, they've mounted the habitat vertically,

0:11:00.755,0:11:03.303
but then I realized that, of course, it doesn't matter.

0:11:03.303,0:11:06.038
So the idea here is that the ants

0:11:06.038,0:11:08.006
are working so hard to hang on

0:11:08.006,0:11:10.707
to the wall or the floor or whatever you call it

0:11:10.707,0:11:14.216
that they're less likely to interact,

0:11:14.216,0:11:16.049
and so the relationship between

0:11:16.049,0:11:17.840
how crowded they are and how often they meet

0:11:17.840,0:11:19.605
would be messed up.

0:11:19.605,0:11:21.190
We're still analyzing the data.

0:11:21.190,0:11:22.964
I don't have the results yet.

0:11:22.964,0:11:24.668
But it would be interesting to know

0:11:24.668,0:11:26.647
how other species solve this problem

0:11:26.647,0:11:29.211
in different environments on earth,

0:11:29.211,0:11:30.977
and so we're setting up a program

0:11:30.977,0:11:32.857
to encourage kids around the world

0:11:32.857,0:11:35.533
to try this experiment with different species.

0:11:35.533,0:11:37.200
It's very simple.

0:11:37.200,0:11:39.009
It can be done with cheap materials.

0:11:39.009,0:11:41.543
And that way, we could make a global map

0:11:41.543,0:11:45.310
of ant collective search algorithms.

0:11:45.310,0:11:47.873
And I think it's pretty likely that the invasive species,

0:11:47.873,0:11:49.842
the ones that come into our buildings,

0:11:49.842,0:11:51.654
are going to be really good at this,

0:11:51.654,0:11:53.432
because they're in your kitchen

0:11:53.432,0:11:57.549
because they're really good[br]at finding food and water.

0:11:57.549,0:12:00.654
So the most familiar resource for ants

0:12:00.654,0:12:02.199
is a picnic,

0:12:02.199,0:12:04.344
and this is a clustered resource.

0:12:04.344,0:12:05.603
When there's one piece of fruit,

0:12:05.603,0:12:07.888
there's likely to be another piece of fruit nearby,

0:12:07.888,0:12:11.003
and the ants that specialize on clustered resources

0:12:11.003,0:12:13.152
use interactions for recruitment.

0:12:13.152,0:12:14.849
So when one ant meets another,

0:12:14.849,0:12:16.504
or when it meets a chemical deposited

0:12:16.504,0:12:18.092
on the ground by another,

0:12:18.092,0:12:19.743
then it changes direction to follow

0:12:19.743,0:12:21.466
in the direction of the interaction,

0:12:21.466,0:12:23.173
and that's how you get the trail of ants

0:12:23.173,0:12:24.786
sharing your picnic.

0:12:24.786,0:12:26.651
Now this is a place where I think we might be able

0:12:26.651,0:12:29.822
to learn something from ants about cancer.

0:12:29.822,0:12:31.803
I mean, first, it's obvious that we could do a lot

0:12:31.803,0:12:33.413
to prevent cancer

0:12:33.413,0:12:36.047
by not allowing people to spread around

0:12:36.047,0:12:37.934
or sell the toxins that promote

0:12:37.934,0:12:40.714
the evolution of cancer in our bodies,

0:12:40.714,0:12:43.044
but I don't think the ants can help us much with this

0:12:43.044,0:12:46.418
because ants never poison their own colonies.

0:12:46.418,0:12:48.060
But we might be able to learn something from ants

0:12:48.060,0:12:49.813
about treating cancer.

0:12:49.813,0:12:51.838
There are many different kinds of cancer.

0:12:51.838,0:12:54.786
Each one originates in a particular part of the body,

0:12:54.786,0:12:56.892
and then some kinds of cancer will spread

0:12:56.892,0:13:00.512
or metastasize to particular other tissues

0:13:00.512,0:13:03.662
where they must be getting[br]resources that they need.

0:13:03.662,0:13:05.045
So if you think from the perspective

0:13:05.045,0:13:07.150
of early metastatic cancer cells

0:13:07.150,0:13:08.973
as they're out searching around

0:13:08.973,0:13:11.350
for the resources that they need,

0:13:11.350,0:13:13.123
if those resources are clustered,

0:13:13.123,0:13:16.306
they're likely to use interactions for recruitment,

0:13:16.306,0:13:19.389
and if we can figure out how[br]cancer cells are recruiting,

0:13:19.389,0:13:21.526
then maybe we could set traps

0:13:21.526,0:13:25.125
to catch them before they become established.

0:13:25.125,0:13:29.053
So ants are using interactions in different ways

0:13:29.053,0:13:31.662
in a huge variety of environments,

0:13:31.662,0:13:33.483
and we could learn from this

0:13:33.483,0:13:35.087
about other systems that operate

0:13:35.087,0:13:37.557
without central control.

0:13:37.557,0:13:39.536
Using only simple interactions,

0:13:39.536,0:13:41.251
ant colonies have been performing

0:13:41.251,0:13:44.924
amazing feats for more than 130 million years.

0:13:44.924,0:13:46.894
We have a lot to learn from them.

0:13:46.894,0:13:49.632
Thank you.

0:13:49.632,0:13:52.365
(Applause)