February 17, 2025 • 42 mins
How are secrets in the brain like Abraham Lincoln’s political cabinet? Will AI in the near future hide things from you? And what does any of this have to do with political hierarchies, the formula for Coca-Cola, or deceptive chimpanzees? Join Eagleman to understand what neuroscience tells us about secrets: what they are, why they weigh on us, and how they sometimes grow into tangled webs we never meant to weave.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:05):
Why is it so hard to keep a secret?
Speaker 2 (00:08):
And what does this have to do with Abraham Lincoln's
political cabinet or when bosses should keep secrets within companies?
What does any of this have to do with political
hierarchies or chimpanzees or the formula for Coca cola, or
whether AI in the near future will be keeping secrets
(00:30):
from you. Today's episode is all about secrets. We've all
got them, from the little white lies we tell to
spare somebody's feelings to the deep unspoken things that we
try to bury. Secrets are a part of being a primate.
But why do we have them, why do we keep them?
And what happens when they start to unravel. As we'll see,
(00:52):
neuroscience tells us that secrets aren't just stashed away. They're active.
They weigh on us, They shape our relationships, and sometimes
they grow into tangled webs we never meant to weave. Today,
we'll dive into the neuroscience of secrets, how they form,
how they grow, and what it takes to keep them
(01:15):
or spill them. This is in her Cosmos, and I'm
David Eagleman. Let's dive into our three pound universe to
untangle the web. Now, before diving into the science of secrets,
(01:36):
we'll start with an observation that researchers made sometime ago.
Keeping a secret is unhealthy. So there's a psychologist named
James Pennebaker, and he and his colleagues studied what happened
when victims of rape or incest when they acted out
of shame or guilt and chose to keep secrets inside. So,
(01:59):
after years of study, Pennebaker concluded that quote, the act
of not discussing or confiding the event with another maybe
more damaging than having experienced the event per se. He
and his team discovered that when subjects confesst or wrote
about their deeply held secrets, their health improved. Their number
(02:23):
of doctor visits went down, and there were measurable decreases
in their stress hormone levels. So the results are clear enough.
But some years ago I began to ask myself how
to understand these findings from the point of view of
brain science, and that led to a question that I
realized was unaddressed in the scientific literature. What is a secret? Neurobiologically,
(02:48):
imagine constructing an artificial neural network of millions of interconnected neurons.
Speaker 1 (02:54):
What would a secret look like?
Speaker 2 (02:56):
Here could a toaster with its interconnect did parts harbor
a secret? We have useful scientific frameworks for understanding things
like Parkinson's disease, or color perception or sensing temperature, but
we don't really have any scientific framework for understanding what
(03:16):
it means for the brain to have and to hold
a secret. So that's what we're going to put into
place today. Now, before we get there, let's start with
Abraham Lincoln. In eighteen sixty, the young Lincoln emerged as
the Republican part of The's candidate for president. And this
was a surprise because he was relatively unknown compared to
(03:40):
the seasoned politicians he was going up against, like William
Seward and Sam and Chase and Edward Bates who are
also vying for that nomination. But here's what Lincoln did
after he won the presidency. He chose those people his rivals,
to comprise his presidential cabinet. And not only them, but
(04:02):
he also put in other people with totally different political
ideologies from his. And that's how he formed his cabinet.
The historian Doris Kerns Goodwin labeled this a team of rivals. Now,
I've always loved this fact about Lincoln, first because it's
actually quite sensible politically, but very few people do it.
(04:24):
It's sensible because bringing together people with conflicting perspectives reflected
his deep commitment to the Union and his focus on
the bigger picture rather than personal grudges. He believed that
the challenges of leading a divided nation through the Civil
War required the best minds available, even if they disagreed
(04:46):
with him or each other. So he got a spread
of opinions this way. But the key thing is that
the internal conflict gave strength to his presidency because no
one gets to enjoy the delusion that they're there's a
right and a wrong answer, But instead, every disagreement gets
fought out, and sometimes a good solution is reached, and
(05:08):
sometimes the problem festers and is revisited over and over. Now,
the argument I forwarded in my book Incognito is that
the brain fundamentally can best be understood as a team
of rivals. So what does that mean. Well, it means
that you are made up of many different neural networks,
(05:29):
each of which sees different data and has its own goals.
And this is why you can argue with yourself, or
get mad at yourself, or try to control yourself to
do something, or get angry at yourself, because you are
not one thing, you are a multitude. I'll give you
(05:50):
an example of this when it comes to how we
make purchasing decisions, as in which restaurant to choose, or
which clothing brand to buy, or what car should I purchase.
In economics, we were taught about Homo economicus, the rational
decision maker who maximizes gain and minimizes loss and is
unswayed by emotion.
Speaker 1 (06:10):
But that is an idealized model.
Speaker 2 (06:14):
Real humans are emotional and inconsistent and easily influenced by
context and branding, and the gap between theory and reality
has given rise to the field of neuroeconomics. So consider this.
I give you two courts of generic ice cream. One
is priced lower than the other, and so the rational
(06:35):
choice is pretty easy. But if I reveal the labels,
let's say one is Ben and Jerry's one is Hoggendaws,
your decisions might shift because all you're experiencing with their
branding kicks in and your emotional predictions about how good
something is likely to taste based on your past experiences,
and maybe your sense of whether your friends like this
(06:57):
or not, And all of this shifts the decision making
one way or the other. What two decades of neuroimaging
has shown is that our decisions about what to buy
are driven by at least three major brain networks. You
have one network along the midline that calculates the price
or worth of something. Then you've got a separate network
(07:19):
involved in emotion, mostly in the orbit or frontal cortex,
and that anticipates how good or bad something will feel.
Then you've got another network that's all about your choices
in a social context, as in, what do your friends
think about this?
Speaker 1 (07:35):
Is it cool or is it lame?
Speaker 2 (07:37):
And all of this tells us why companies don't just
advertise with bullet points about their rational advantages.
Speaker 1 (07:45):
They spend millions of dollars on ads.
Speaker 2 (07:48):
To assure us that this is the best value and
appeal to our feelings, and to assure us that everyone
else loves this product as well. So, as a result
of the complicate, hated voting of our neural parliaments, decisions
aren't always straightforward to decode. So the next time you're
thinking about your choice between ice cream brands or fast food,
(08:11):
restaurants or car brands. Ask yourself, what is driving this decision?
Is it price? Is it emotion? Is there a social
influence on what I'm choosing? All these networks are fighting
it out, each trying to have control in steering the
ship of state. If you're interested in more about this,
checkout episodes eight and nine. Now I mention all this
(08:34):
now to illustrate this notion of battles in the brain.
It's not just with buying decisions, but with all the
choices you make in life. You have some networks that
care about the short term in decision making, and you
have some that care about the long term. You have
networks that are monitoring and making predictions about the outside
(08:55):
world and others that care about your inside world, the
state of your body. Some of your networks provide a
drive towards novelty and others towards familiarity, and these can
both be active at the same time and having an
arm wrestle. So think about these rivaling networks like a
neural parliament. You're built of different political parties, all of
(09:18):
whom love their country, but they just have different opinions
about the best way forward. So this gives us a
sense of these battling networks, and now I want to
zoom in on something surprising. What you get out of
a brain that is composed of multiple rivaling networks, You
get a universe of secrets. The team of rivals framework
(09:42):
allows us to address a mystery that would be inexplicable
if we took the point of view of traditional computer programs.
So let's come back to the question I asked at
the beginning, what is a secret? Neurobiologically? And could your
toaster keep a secret? Within the team of rivals framework,
a secret is easily understood. It's the result of struggle
(10:05):
between competing parties in the brain. One part of the
brain wants to reveal something and another part does not
want to. When there are competing votes in the brain,
one for telling and one for withholding, that's a secret.
If neither party cares to tell, that's merely a boring fact.
And if both parties want to tell, that's just a
(10:27):
good story. Without the framework of rivalry, we would have
no way to understand a secret. The reason a secret
is experienced consciously is because it results from a conflict.
It's not business as usual, and so quite often our
consciousness is called upon to deal with it. Now, the
(10:48):
main reason not to tell a secret is aversion to
the long term consequences. A friend might think ill of you,
or a relationship might be revocably damaged, or a community
might ostracize you. This concern about the outcome is evidenced
by the fact that people are more likely to tell
(11:10):
their secrets to total strangers. With somebody that you don't know,
the neural conflict can be dissipated with none of the costs.
And that's why strangers are occasionally so forthcoming on airplanes
or in bars telling you all the details of their
marital troubles. And we can see a modern twist on
(11:32):
this ancient need to confess to strangers in the form
of websites like postsecret dot com. This is a site
where people go to anonymously disclose their confessions. Here are
some examples.
Speaker 1 (11:47):
Quote.
Speaker 2 (11:47):
When my daughter was stillborn, I not only thought about
kidnapping a baby, I planned it out in my head.
I even found myself watching new mothers with their babies,
trying to pick the perfect Now we don't know who
posted this, but for that person something was released by
confessing this. Here's another one. I was adopted, and my
(12:09):
biggest fear is passing my biological father on the street.
Or here's another one. I'm almost certain that your son
has autism, but I have no idea how to tell you.
There are tens of thousands of secrets on this website
and why. It's because just telling your secret to the website,
where you believe it will be read by other people
(12:30):
strangers to you, is sufficient. And this reduction of tension
by releasing a secret. This, I think also explains a
staple in one of the world's largest religions, the confessional booth.
You get something off your chest and without the consequences
that might otherwise accrue.
Speaker 1 (12:49):
This same need to reveal.
Speaker 2 (12:51):
Ourselves might also shine light on the appeal of prayer,
especially with religions that have a very personal God who
lends an ear with undivided attention and non judgmental love.
So why is a stranger or a website or a
deity useful. It's because it's the act of telling the
secret that matters. Things don't have to get solved, the
(13:13):
cognitive tension just has to be reduced. We've all seen
this that people often will vent a secret for its
own sake, not as an invitation for advice. For example,
somebody tells you a confession and you make the mistake
of suggesting some obvious solution that only frustrates the other person,
because all they really wanted was to tell to reduce
(13:36):
the neural tension. The act of telling the secret was
the solution. Now an open question is why the receiver
of the secrets has to be human or human like
in the case of deities. If you tell your secret
to a wall or a lizard or a goat, that's
going to be a lot less satisfying. Now, one thing
(13:57):
that strikes me is interesting in this regard is the
right popularity of AI therapist bots.
Speaker 1 (14:03):
Where will lees sit in.
Speaker 2 (14:05):
The hierarchy of human to non human I think it'll
be pretty close to a person or a deity. I
suspect that in the future we'll be pretty happy to
let our secrets out to our AI therapist, But only
if you're convinced about the security of the company. Is
it just a matter of time before the FBI says
(14:26):
to the tech company, Hey, we can subpoena your therapy
history under special circumstances. So it's still to be seen
how much people will unload on their bots because the
key with unloading your secrets is to get the cognitive
benefit without the social consequences. Okay, so I've been and
(15:00):
referring to this cognitive load of keeping a secret. But
the question is why do we have difficulty? Why isn't
it easy to keep a secret? Here's why, because keeping
a secret requires ongoing mental effort to monitor your speech
and actions. You have to make sure that the hidden
(15:20):
information isn't accidentally revealed. Even if this is happening at
a subconscious level, it's a state of constant vigilance to
keep from blurting out the secret. So how do you
measure this? So some years ago, a scientist named Sean
Spence and his colleagues interviewed participants on what activities they
(15:40):
had done the previous day, and then they put them
in the brain scanner fMRI. These people were given questions
about what they did and they were given two buttons
yes or no. Now, on each round, they were given
either a red light or a green light to tell
them whether they should lie or whether they should tell
the truth on that trial. So the first thing the
(16:02):
researchers noted is that it always takes a little bit
longer to lie, about a fifth of a second longer,
and this is because you are taking the answer that
you already know the true answer, and you're squelching that.
So they demonstrated that this squelching of the truth responses
activates a part of the prefrontal cortex, a part that's
down load near the side. It's called the ventrilateral prefrontal cortex.
(16:27):
The thing to appreciate is that this region becomes active
whenever you need a suppressive behavior that you would normally
just go with, in this case telling the truth. And
other brain errors are involved too. In a study bi
Langeleman and colleagues the next year, participants went into a
scanner and they saw a playing card on the screen
and they were asked to remember it.
Speaker 1 (16:48):
Then they were shown.
Speaker 2 (16:49):
A series of cards and they were asked was that
your card or not?
Speaker 1 (16:53):
And they were sometimes instructed to lie. And here's the key.
Speaker 2 (16:58):
When they lied, researchers saw a lot of activity in
an area called the anterior singulate cortex, which is a
region that detects conflict between other brain areas. So in
both these studies, people were just saying yes or no
to lie or not lie. But in the next study
by Spence and colleagues, they wanted to see what happened
(17:18):
in the brain when people are forced to get more
imaginative when they go beyond yes or no answers to
make up a new story. So imagine I ask you
where were you on the evening of February twenty second,
and you can figure out the answer that.
Speaker 1 (17:34):
Let's say you were at a great party that night.
Speaker 2 (17:37):
But imagine that you lie to me and you say, oh,
I met my friend for a quiet movie on Netflix
that night. Now, in order to do this, in order
to lie, you have to think of the true response
and suppress that like we saw before, but also you
then have to make something up. So you're now doing
two things. You're suppressing the true answer, which cranks the
(18:00):
the ventralateral prefrontal cortex like we saw, and then you
need to cook up the lie, and that activates a
different region than dorsilateral prefrontal cortex. The dorsilateral area cranks
up when you generate something new, like a new story,
and especially if this is the first time that you're
making up that story. And we see other areas too,
(18:20):
like the anterior cingulate cortex, which reflects internal conflict. And
we see another part of the prefrontal cortex, the ventromedial region,
which cranks up when you're trying to regulate your emotions.
So covering up the truth is a high energy endeavor
for the brain. We never see areas where there's more
(18:41):
activity when you tell the truth.
Speaker 1 (18:43):
There's only more activity when you're lying.
Speaker 2 (18:47):
So the idea that it's easier to tell the truth
holds as a rule in life, presumably because it holds
as a rule in the brain. And if the cooked
up story doesn't go so per we end up painting
ourselves into a corner, which brings with it lots more
brain energy expenditure. And this observation is one that the
(19:10):
Scottish poet novelist Sir Walter Scott immortalized in his poem
Marmion in eighteen oh eight. This is where he penned
the immortal line, Oh what a tangled web we weave
when first we practice to deceive. He was pointing to
a common problem that arises when we make up a
(19:30):
lie to cover a secret, and then that leads to
another necessary lie, which leads to another, and now you
have to keep track of this whole web of lies
to try to keep your story consistent, which can lead,
as Sir Walter Scott may have intuited, to a high
cognitive load. And cognitive load matters because it equates to
emotional stress. We all know from experience that holding on
(19:55):
to a secret creates anxiety, and often these bad feelings
increase our urge to confess or share the secret to
relieve the burden. So secret keeping requires a lot of work.
But what I want to highlight is the double edged
nature of the secret because, perhaps surprisingly for better or worse,
(20:18):
the ability to keep a secret is one of the
main drivers of the shape of our species and society. Now,
we can talk all day about the downsides of keeping
secrets from one another, but in fact, the ability to
keep secrets seems to have evolved many millions of years
ago as a survival mechanism, as it turns out, concealing
(20:41):
information about food sources or about shelter or defensive strategies.
This is often what made the difference between life and death.
And it's not just humans who practice deception. Other primates
do this as well. Now I'm not talking about deception
like using camouflage. Instead I'm talking thinking about active, clever
(21:02):
deception that requires theory of mind, in other words, the
ability to step into the perspective of another to understand
what they know and don't know. It turns out that
only primates seem to do this, and in fact they're
constantly doing things to deceive one another. On this note,
here's a short clip from my interview with Max Bennett
(21:23):
from last year, who wrote a great book called A
Brief History of Intelligence. And here's Max describing an example
of deception in chimpanzees.
Speaker 3 (21:32):
So you can see non human apes do things like
they will hide transgressions from other people to try and
prevent themselves from getting in trouble. There's this famous study
that I love by Emil Menzel. I think it was
in the seventies where he put two chimpanzees in the
sort of one acre forest and he showed the location
of treats to one of the chimpanzees named Belle, and
(21:56):
she initially would share the treat with another chimpanze he
named Rock, but then Rock started just stealing the treat
from her. So what she started doing is when she
knew the location of the treat, she would wait for
a rock to look away, and then she would run
over and grab it. So then rock in response to this,
decided to pretend to look away so that when she
(22:17):
started running, then he would turn around and run. Then
in response to this, what she would do is she
would pretend to run in the wrong direction, lead him
to the wrong place, and then run back.
Speaker 2 (22:26):
So the tricky deceptive capacity we have to not spill
the beans enhances the chances of individual survival. Now, when
we zoom in on humans and study them very carefully,
what we find is that keeping secrets and more generally,
lying is a cognitive development in children. In other words,
(22:49):
And first young children.
Speaker 1 (22:51):
Don't know how to lie.
Speaker 2 (22:52):
And then they hit a milestone one day, let's say
between two and four years old, and they get a
great idea, maybe they can assert something that's not true
and maybe the other person won't know. So they carefully
float this trial balloon and sometimes it works. Suddenly they
have eaten the cooky but avoided punishment, so they start
(23:16):
to lie. As soon as they develop theory of mind,
they start to understand that their thoughts and minds are
separate from those of their parents. They also begin to
realize that people aren't perfect mind readers, so the child
can say something that isn't true, and miraculously the other
person doesn't always know. Obviously, sometimes the child doesn't get
(23:39):
away with it, and so they conclude that they need
to work even harder to cook up better fabrications that
might fool the other person. And for this they need
to get better and better at stepping into the other
person's shoes to understand what that person might fall for.
So children start to lie more frequently as they get
older and they get better at lying. They learn to
(24:01):
match their facial expressions and tone of voice to what
they're saying, and they shape their lies based on their
assumptions of what the other person may or may not know.
And lying sucks, especially when it's in your own kids.
But I'll just mention that some researchers view lying as
an important developmental stage that indicates early intelligence. So while
(24:24):
none of us appreciate being lied to by our kids
or having secrets kept from us by our colleagues, it's
one of the signatures of the extraordinarily intelligent primate brain.
And in fact, when you look across human history, the
thing that becomes clear is that secrets quickly expanded beyond
the individual level to the group level. So trust within
(24:48):
small groups usually depended on the ability to keep certain
knowledge hidden from outsiders, and by the way, keeping internal
group secrets has the added benefit of thought during a
sense of belonging in mutual protection and secrets have always
been a key part of military strategy. The immortal book
(25:09):
The Art of War by Sun Su paints a picture
of deception and secrecy as absolutely essential for successful warfare,
and more generally, secrets have always been a lever for
power within larger groups. As civilizations grew, secrets became pervasive
(25:29):
wherever people maintained and exercised power. You see this all
over ancient empires like Egypt and China, where they leveraged
secrecy to control knowledge and to reinforce hierarchies. And as
we all know from our history classes and from Game
of Thrones, the secrets swirling around among advisors in royal
(25:52):
courts often dictated the rise and fall of leaders. Now
people have been grappling with the tension between keeping secrets
in a society versus telling everything to the public, and
they've been debating these points for a long time. So,
for example, the Enlightenment was characterized by arguments going on
about the value of secrecy versus transparency. You have some
(26:16):
philosophers like Jeremy Bentham who hated the idea of secrecy
in governance, but secrecy has always persisted in areas like
diplomacy and statecraft, where its defenders deem it as essential
for stability. And you can see the fingerprint of this
very old debate in modern questions about surveillance. To what
(26:39):
extent should a government be able to keep track of
who's doing what with cameras or data tracking? Should anyone
be able to keep total privacy from the states such
that I can work on three D printing guns and
making nuclear bombs in the privacy of my home. These
debates will presumably go on forever because because there is
(27:00):
no single right answer for balancing the privacy of the
individual against the security of the society. So back to secrets,
it's not just governments that traffic in these. Religious traditions
almost always involve sacred secrets accessible only to initiates or clergy.
(27:21):
In ancient Greece, there were rituals performed every year called
the Eleucinian Mysteries, and to participate you had to swear
a vow of secrecy, and in return you were promised
spiritual enlightenment. And you then learned and safeguarded the secret rights.
And these were kept so secret that all these centuries
later we still don't know what happened in those rituals.
(27:44):
Or similarly, in the medieval era, knowledge of alchemy or
mystical texts was always hidden so that it was accessible
only to a select few. More generally, this is the
kind of glue that keeps people locked into whether that's
religions or organizations or fraternities. It's the tension from letting
(28:06):
them know that as they level up, they'll get access
to the next big secrets, so they'll be part of
the in group, and they'll link arms to keep those
secrets from everyone else. As I mentioned a moment ago,
sharing a secret builds strong bonds between individuals, It fosters intimacy,
and it creates alliances. I'll just mention one more thing
(28:46):
about secrets on a societal level. Beyond warfare and politics
and religions, subgroups in society have to keep secrets to
maintain competitive advantages. In technology. Take something like trade secrets. Historically,
you've got things like the exact technique for making Chinese silk,
(29:07):
or currently the exact formula for Coca cola, or the
software details in cybersecurity firms. In all of these cases,
knowledge is closely guarded to keep a competitive advantage, and
people have studied the issue about keeping secrets even inside
a company, like the upper management knows something but the
(29:29):
employees don't, so that sounds bad. The researchers at Stanford
just published a paper on the benefits that companies derive
from what they call the confidentiality effect. The research suggests
that keeping company secrets has the benefit of boosting feelings
of privilege and status among those who know the secret
(29:51):
and protecting those who do not. Now, I want to
tie this into something I talked about before in the
episode titled why do you Brains love conspiracy theories? The
brain is always seeking explanations to reduce cognitive dissonance and
provide a sense of certainty. And when we solve puzzles,
(30:12):
that's deeply rewarding. And that's why uncovering patterns, even ones
that aren't accurate, that's why that feels so rewarding. Now,
one can never say that all conspiracy theories are false
because certainly, at moments in history, people can try to
get away with something, and they are very incentivized to
(30:32):
cover their tracks. But most conspiracy theories fall apart under scrutiny.
Why we'll just think about conspiracy theories through the lens
of human behavior and probabilities. Imagine a grand conspiracy like
faking the moon landing or hiding evidence of alien life
(30:53):
in Roswell. These cover ups would require hundreds, if not thousands,
of people to remain silent across agencies, across generations, across families.
But real life doesn't work that way. People get drunk,
they talk in their sleep, they have moments of guilt,
they decide to confess on their deathbeds. Secrets are fragile,
(31:16):
and the longer they need to be kept, the more
likely they are to leak. Now, you can model the
odds of someone spilling the secret using a math technique
called survival analysis, and it goes like this. Say there's
a tiny daily chance that someone involved lets something slip.
Let's say it's really small chance, like one in one
(31:38):
thousand on any given day that they'd screw up, but
multiply that over hundreds of people in years of time
and the probability of the secret being kept approaches zero.
You're rolling the dice over and over that somebody is
going to spill the beans. And note that with most
conspiracy theories, the incentive to spill the beans is usually compelling.
(32:02):
In the end, people typically act in their self interest,
and people who are in on a conspiracy sometimes find
the rational choice is to defect before someone else does.
In other words, if you are the one who spills
the big secret, let's say of a fake moon landing
or a planned murderer of JFK, then suddenly, not only
(32:26):
have you released the stress of a secret, but more importantly,
you're now famous and protected by the legal system and
probably signing book deals with Penguin Random House. So game
theory asks why risk being the one left behind when
you have so much incentive structure encouraging you to blurt
out the truth before someone else does. So, when you're
(32:49):
confronted with a conspiracy theory, the question isn't just could
that happen? But more importantly, how many people would need.
Speaker 1 (32:58):
To keep the secret?
Speaker 2 (33:00):
And how much time has passed without anyone defecting. And
this is why when you do the math and factor
in human behavior, the vast majority of conspiracies become not
just improbable, but implausible. So now I want to return
to a question that I posed at the very beginning.
Can a computer keep a secret? Well, a toaster doesn't
(33:24):
have the circuitry to keep a secret from you, presumably
because it doesn't have complex enough circuitry, it doesn't have
a team of rivals fighting it out under the hood.
Speaker 1 (33:35):
But could AI keep a secret from you?
Speaker 2 (33:39):
Well, people started asking this question some years ago using
what are called adversarial networks. Just think of this like
networks that are pitted against one another, and they each
might have a different incentive structure. So back in twenty sixteen,
a group of researchers a Google Brain, wanted to know
whether neural nets networks could develop their own encryption methods
(34:03):
independently without human guidance. So they set up three independent networks.
The first was named Alice, and Alice's task was to
send a secret message to Bob. This second network, Bob
His job was to decode the message sent by Alice.
And then there was a third network named Eve. Now
(34:25):
Eve represented and evesdropper. Her role was to try to
intercept and decrypt Alice's message, so they set these three
networks running. Alice and Bob were trained collaboratively to ensure
Bob could correctly decode the messages, and Eve was trained adversarially.
Her job was to improve her decryption attempts, and the
(34:49):
system evolved dynamically. As Alice and Bob became better at
secure communication, Eve became more sophisticated in breaking their encryption,
forcing Alice and Bob to improve further. Now, the outcome
was kind of incredible because Alice and Bob's neural networks
learned to create their own encryption strategies, and this was
(35:13):
without explicit programming. In other words, the encryption strategy was
never told to them by a human. So over time
those two succeeded in creating encrypted communications that Eve couldn't decrypt.
They outpaced Eve's efforts. The key here is that the
encryption methods developed by Alice were not human readable or interpretable.
(35:38):
The system invented its own cryptographic methods based on its
training objectives very rapidly. The two neural networks figured out
how to keep a secret not only from Eve, but
from everybody. They developed a totally novel and dynamic cryptographic system.
So that was one study. Then in twenty seven, the
(36:00):
Facebook AI research group was working on developing AI agents
that could negotiate. But when they left these agents to
their own devices, they saw unexpected behaviors.
Speaker 1 (36:12):
So here was this setup.
Speaker 2 (36:13):
The goal of the experiment was to train these two bots,
these two AI agents, to negotiate with each other to
achieve their objectives. So the two AI agents are tasked
with dividing items between them, like books, or balls or hats.
Each item has a different value to each of the bots,
(36:35):
and each bot has the objective to maximize their score
by getting the items that they value the most, and
they're rewarded based on how well they do. Okay, so
they set these bots running, but pretty soon two surprises emerged.
The first was that the bots started using tricky tactics.
They would sometimes pretend to be uninterested in a particular
(36:59):
item to convince the other bot to concede it. This
is called feigned disinterest, and of course humans do this
in negotiations to manipulate the perceived value of something that
they are interested in, but it certainly wasn't expected that
this would emerge on its own in an artificial neural network.
(37:19):
The second thing was that the bots started to communicate strangely.
They developed their own shorthand language to negotiate. So Bot
one would say I want want, want want one book
and bought two would respond okay, you you you you
two balls. Now why did their language start to turn
(37:39):
funny like this? It's because the AI wasn't explicitly incentivized
to stick with human readable language. Instead, the reward system
only focused on achieving successful negotiations, so the bots optimized
their communication for efficiency rather than for clarity to humans.
(38:01):
So it made sense to the bots, but eventually became
unintelligible to the human reader. Now, this experiment gives us
a few deep insights.
Speaker 1 (38:10):
The first thing it.
Speaker 2 (38:11):
Shows us is emergent behaviors. AI agents can develop creative
strategies to achieve their goals, and this is typically not
anticipated by the researchers. In this case, what developed was
something like a secret, or at least we would say,
a non truth, because the AI would pretend one thing
(38:33):
when it wanted something else. The experiment also demonstrated that
if it's not explicitly constrained, AI will optimize for its
objectives in ways that maybe don't align with human readability,
and if you extrapolate this, you can see this is
another way that AI might keep a secret from us,
even if it's not intending to. Just like the networks
(38:56):
Alice and Bob, it might just learn how to speak
a language that we simply don't understand. So let's wrap
up today's exploration of secrets and the brain. As we saw,
secrets of something we're often embarrassed to address, but they're
a massive feature of primate lives, shaping everything from personal
(39:17):
relationships to large scale politics. The act of keeping a
secret emerges from the rivalry within our brains, a team
of neural networks vying for control over what's revealed and
what's hidden. This internal competition creates the cognitive and emotional
weight of secret keeping, which can sometimes push us to
(39:41):
seek relief by confessing to strangers, or by writing anonymous
posts on the web, or even whispering to an AI therapist.
The relief that we feel from telling a secret comes
not from solving the external problem, but from reducing the tension.
(40:02):
And as we saw, secrets don't stop at the individual level.
They've been a key driver of human society, from military
strategy and political hierarchies to the rise of religions and
social institutions. Secrets bind groups together, They foster trust among allies.
(40:22):
They create competitive advantages, whether that's the formula for Coca
cola or the sacred rights of ancient mystery cults. But
they also lead to struggles over societal transparency, and there's
always a healthy debate about how much governments should be
allowed to conceal from us.
Speaker 1 (40:42):
And what about AI.
Speaker 2 (40:44):
I told you about two experiments where neural networks, left
to their own devices, developed encryption methods and private languages
that humans couldn't decode. These aren't conscious secrets in the
way we experience them, but they light a future where
machines might possess knowledge inaccessible to us, not necessarily because
(41:06):
they're deliberately hiding it, but because their optimization processes have
outpaced our ability to understand. As aisystems grow more sophisticated,
the question of what they know and what they're going
to keep from us is only going to evolve. So
secrets are woven into the fabric of our minds and societies.
(41:30):
They've been with us since before the sunrise of human cognition,
As we see from our primate cousins, and secrets and
lies may soon take on new dimensions as we come
to share our world with increasingly intelligent machines. Whether secrets
are helping us to bond or they're dividing us, they
are a reminder of the complexity of our inner cosmos
(41:55):
and the tangled web of our relationships with each other
and soon with our technology. Go to eagleman dot com
slash podcast for more information and to find further reading.
Send me an email at podcast at eagleman dot com
(42:15):
with questions or discussion, and check out and subscribe to
Inner Cosmos on YouTube for videos of each episode and
to leave comments.
Speaker 1 (42:25):
Until next time.
Speaker 2 (42:26):
I'm David Eagleman, and this is Inner Cosmos.
Why is it so hard to keep a secret?
Speaker 2 (00:08):
And what does this have to do with Abraham Lincoln's
political cabinet or when bosses should keep secrets within companies?
What does any of this have to do with political
hierarchies or chimpanzees or the formula for Coca cola, or
whether AI in the near future will be keeping secrets
(00:30):
from you. Today's episode is all about secrets. We've all
got them, from the little white lies we tell to
spare somebody's feelings to the deep unspoken things that we
try to bury. Secrets are a part of being a primate.
But why do we have them, why do we keep them?
And what happens when they start to unravel. As we'll see,
(00:52):
neuroscience tells us that secrets aren't just stashed away. They're active.
They weigh on us, They shape our relationships, and sometimes
they grow into tangled webs we never meant to weave. Today,
we'll dive into the neuroscience of secrets, how they form,
how they grow, and what it takes to keep them
(01:15):
or spill them. This is in her Cosmos, and I'm
David Eagleman. Let's dive into our three pound universe to
untangle the web. Now, before diving into the science of secrets,
(01:36):
we'll start with an observation that researchers made sometime ago.
Keeping a secret is unhealthy. So there's a psychologist named
James Pennebaker, and he and his colleagues studied what happened
when victims of rape or incest when they acted out
of shame or guilt and chose to keep secrets inside. So,
(01:59):
after years of study, Pennebaker concluded that quote, the act
of not discussing or confiding the event with another maybe
more damaging than having experienced the event per se. He
and his team discovered that when subjects confesst or wrote
about their deeply held secrets, their health improved. Their number
(02:23):
of doctor visits went down, and there were measurable decreases
in their stress hormone levels. So the results are clear enough.
But some years ago I began to ask myself how
to understand these findings from the point of view of
brain science, and that led to a question that I
realized was unaddressed in the scientific literature. What is a secret? Neurobiologically,
(02:48):
imagine constructing an artificial neural network of millions of interconnected neurons.
Speaker 1 (02:54):
What would a secret look like?
Speaker 2 (02:56):
Here could a toaster with its interconnect did parts harbor
a secret? We have useful scientific frameworks for understanding things
like Parkinson's disease, or color perception or sensing temperature, but
we don't really have any scientific framework for understanding what
(03:16):
it means for the brain to have and to hold
a secret. So that's what we're going to put into
place today. Now, before we get there, let's start with
Abraham Lincoln. In eighteen sixty, the young Lincoln emerged as
the Republican part of The's candidate for president. And this
was a surprise because he was relatively unknown compared to
(03:40):
the seasoned politicians he was going up against, like William
Seward and Sam and Chase and Edward Bates who are
also vying for that nomination. But here's what Lincoln did
after he won the presidency. He chose those people his rivals,
to comprise his presidential cabinet. And not only them, but
(04:02):
he also put in other people with totally different political
ideologies from his. And that's how he formed his cabinet.
The historian Doris Kerns Goodwin labeled this a team of rivals. Now,
I've always loved this fact about Lincoln, first because it's
actually quite sensible politically, but very few people do it.
(04:24):
It's sensible because bringing together people with conflicting perspectives reflected
his deep commitment to the Union and his focus on
the bigger picture rather than personal grudges. He believed that
the challenges of leading a divided nation through the Civil
War required the best minds available, even if they disagreed
(04:46):
with him or each other. So he got a spread
of opinions this way. But the key thing is that
the internal conflict gave strength to his presidency because no
one gets to enjoy the delusion that they're there's a
right and a wrong answer, But instead, every disagreement gets
fought out, and sometimes a good solution is reached, and
(05:08):
sometimes the problem festers and is revisited over and over. Now,
the argument I forwarded in my book Incognito is that
the brain fundamentally can best be understood as a team
of rivals. So what does that mean. Well, it means
that you are made up of many different neural networks,
(05:29):
each of which sees different data and has its own goals.
And this is why you can argue with yourself, or
get mad at yourself, or try to control yourself to
do something, or get angry at yourself, because you are
not one thing, you are a multitude. I'll give you
(05:50):
an example of this when it comes to how we
make purchasing decisions, as in which restaurant to choose, or
which clothing brand to buy, or what car should I purchase.
In economics, we were taught about Homo economicus, the rational
decision maker who maximizes gain and minimizes loss and is
unswayed by emotion.
Speaker 1 (06:10):
But that is an idealized model.
Speaker 2 (06:14):
Real humans are emotional and inconsistent and easily influenced by
context and branding, and the gap between theory and reality
has given rise to the field of neuroeconomics. So consider this.
I give you two courts of generic ice cream. One
is priced lower than the other, and so the rational
(06:35):
choice is pretty easy. But if I reveal the labels,
let's say one is Ben and Jerry's one is Hoggendaws,
your decisions might shift because all you're experiencing with their
branding kicks in and your emotional predictions about how good
something is likely to taste based on your past experiences,
and maybe your sense of whether your friends like this
(06:57):
or not, And all of this shifts the decision making
one way or the other. What two decades of neuroimaging
has shown is that our decisions about what to buy
are driven by at least three major brain networks. You
have one network along the midline that calculates the price
or worth of something. Then you've got a separate network
(07:19):
involved in emotion, mostly in the orbit or frontal cortex,
and that anticipates how good or bad something will feel.
Then you've got another network that's all about your choices
in a social context, as in, what do your friends
think about this?
Speaker 1 (07:35):
Is it cool or is it lame?
Speaker 2 (07:37):
And all of this tells us why companies don't just
advertise with bullet points about their rational advantages.
Speaker 1 (07:45):
They spend millions of dollars on ads.
Speaker 2 (07:48):
To assure us that this is the best value and
appeal to our feelings, and to assure us that everyone
else loves this product as well. So, as a result
of the complicate, hated voting of our neural parliaments, decisions
aren't always straightforward to decode. So the next time you're
thinking about your choice between ice cream brands or fast food,
(08:11):
restaurants or car brands. Ask yourself, what is driving this decision?
Is it price? Is it emotion? Is there a social
influence on what I'm choosing? All these networks are fighting
it out, each trying to have control in steering the
ship of state. If you're interested in more about this,
checkout episodes eight and nine. Now I mention all this
(08:34):
now to illustrate this notion of battles in the brain.
It's not just with buying decisions, but with all the
choices you make in life. You have some networks that
care about the short term in decision making, and you
have some that care about the long term. You have
networks that are monitoring and making predictions about the outside
(08:55):
world and others that care about your inside world, the
state of your body. Some of your networks provide a
drive towards novelty and others towards familiarity, and these can
both be active at the same time and having an
arm wrestle. So think about these rivaling networks like a
neural parliament. You're built of different political parties, all of
(09:18):
whom love their country, but they just have different opinions
about the best way forward. So this gives us a
sense of these battling networks, and now I want to
zoom in on something surprising. What you get out of
a brain that is composed of multiple rivaling networks, You
get a universe of secrets. The team of rivals framework
(09:42):
allows us to address a mystery that would be inexplicable
if we took the point of view of traditional computer programs.
So let's come back to the question I asked at
the beginning, what is a secret? Neurobiologically? And could your
toaster keep a secret? Within the team of rivals framework,
a secret is easily understood. It's the result of struggle
(10:05):
between competing parties in the brain. One part of the
brain wants to reveal something and another part does not
want to. When there are competing votes in the brain,
one for telling and one for withholding, that's a secret.
If neither party cares to tell, that's merely a boring fact.
And if both parties want to tell, that's just a
(10:27):
good story. Without the framework of rivalry, we would have
no way to understand a secret. The reason a secret
is experienced consciously is because it results from a conflict.
It's not business as usual, and so quite often our
consciousness is called upon to deal with it. Now, the
(10:48):
main reason not to tell a secret is aversion to
the long term consequences. A friend might think ill of you,
or a relationship might be revocably damaged, or a community
might ostracize you. This concern about the outcome is evidenced
by the fact that people are more likely to tell
(11:10):
their secrets to total strangers. With somebody that you don't know,
the neural conflict can be dissipated with none of the costs.
And that's why strangers are occasionally so forthcoming on airplanes
or in bars telling you all the details of their
marital troubles. And we can see a modern twist on
(11:32):
this ancient need to confess to strangers in the form
of websites like postsecret dot com. This is a site
where people go to anonymously disclose their confessions. Here are
some examples.
Speaker 1 (11:47):
Quote.
Speaker 2 (11:47):
When my daughter was stillborn, I not only thought about
kidnapping a baby, I planned it out in my head.
I even found myself watching new mothers with their babies,
trying to pick the perfect Now we don't know who
posted this, but for that person something was released by
confessing this. Here's another one. I was adopted, and my
(12:09):
biggest fear is passing my biological father on the street.
Or here's another one. I'm almost certain that your son
has autism, but I have no idea how to tell you.
There are tens of thousands of secrets on this website
and why. It's because just telling your secret to the website,
where you believe it will be read by other people
(12:30):
strangers to you, is sufficient. And this reduction of tension
by releasing a secret. This, I think also explains a
staple in one of the world's largest religions, the confessional booth.
You get something off your chest and without the consequences
that might otherwise accrue.
Speaker 1 (12:49):
This same need to reveal.
Speaker 2 (12:51):
Ourselves might also shine light on the appeal of prayer,
especially with religions that have a very personal God who
lends an ear with undivided attention and non judgmental love.
So why is a stranger or a website or a
deity useful. It's because it's the act of telling the
secret that matters. Things don't have to get solved, the
(13:13):
cognitive tension just has to be reduced. We've all seen
this that people often will vent a secret for its
own sake, not as an invitation for advice. For example,
somebody tells you a confession and you make the mistake
of suggesting some obvious solution that only frustrates the other person,
because all they really wanted was to tell to reduce
(13:36):
the neural tension. The act of telling the secret was
the solution. Now an open question is why the receiver
of the secrets has to be human or human like
in the case of deities. If you tell your secret
to a wall or a lizard or a goat, that's
going to be a lot less satisfying. Now, one thing
(13:57):
that strikes me is interesting in this regard is the
right popularity of AI therapist bots.
Speaker 1 (14:03):
Where will lees sit in.
Speaker 2 (14:05):
The hierarchy of human to non human I think it'll
be pretty close to a person or a deity. I
suspect that in the future we'll be pretty happy to
let our secrets out to our AI therapist, But only
if you're convinced about the security of the company. Is
it just a matter of time before the FBI says
(14:26):
to the tech company, Hey, we can subpoena your therapy
history under special circumstances. So it's still to be seen
how much people will unload on their bots because the
key with unloading your secrets is to get the cognitive
benefit without the social consequences. Okay, so I've been and
(15:00):
referring to this cognitive load of keeping a secret. But
the question is why do we have difficulty? Why isn't
it easy to keep a secret? Here's why, because keeping
a secret requires ongoing mental effort to monitor your speech
and actions. You have to make sure that the hidden
(15:20):
information isn't accidentally revealed. Even if this is happening at
a subconscious level, it's a state of constant vigilance to
keep from blurting out the secret. So how do you
measure this? So some years ago, a scientist named Sean
Spence and his colleagues interviewed participants on what activities they
(15:40):
had done the previous day, and then they put them
in the brain scanner fMRI. These people were given questions
about what they did and they were given two buttons
yes or no. Now, on each round, they were given
either a red light or a green light to tell
them whether they should lie or whether they should tell
the truth on that trial. So the first thing the
(16:02):
researchers noted is that it always takes a little bit
longer to lie, about a fifth of a second longer,
and this is because you are taking the answer that
you already know the true answer, and you're squelching that.
So they demonstrated that this squelching of the truth responses
activates a part of the prefrontal cortex, a part that's
down load near the side. It's called the ventrilateral prefrontal cortex.
(16:27):
The thing to appreciate is that this region becomes active
whenever you need a suppressive behavior that you would normally
just go with, in this case telling the truth. And
other brain errors are involved too. In a study bi
Langeleman and colleagues the next year, participants went into a
scanner and they saw a playing card on the screen
and they were asked to remember it.
Speaker 1 (16:48):
Then they were shown.
Speaker 2 (16:49):
A series of cards and they were asked was that
your card or not?
Speaker 1 (16:53):
And they were sometimes instructed to lie. And here's the key.
Speaker 2 (16:58):
When they lied, researchers saw a lot of activity in
an area called the anterior singulate cortex, which is a
region that detects conflict between other brain areas. So in
both these studies, people were just saying yes or no
to lie or not lie. But in the next study
by Spence and colleagues, they wanted to see what happened
(17:18):
in the brain when people are forced to get more
imaginative when they go beyond yes or no answers to
make up a new story. So imagine I ask you
where were you on the evening of February twenty second,
and you can figure out the answer that.
Speaker 1 (17:34):
Let's say you were at a great party that night.
Speaker 2 (17:37):
But imagine that you lie to me and you say, oh,
I met my friend for a quiet movie on Netflix
that night. Now, in order to do this, in order
to lie, you have to think of the true response
and suppress that like we saw before, but also you
then have to make something up. So you're now doing
two things. You're suppressing the true answer, which cranks the
(18:00):
the ventralateral prefrontal cortex like we saw, and then you
need to cook up the lie, and that activates a
different region than dorsilateral prefrontal cortex. The dorsilateral area cranks
up when you generate something new, like a new story,
and especially if this is the first time that you're
making up that story. And we see other areas too,
(18:20):
like the anterior cingulate cortex, which reflects internal conflict. And
we see another part of the prefrontal cortex, the ventromedial region,
which cranks up when you're trying to regulate your emotions.
So covering up the truth is a high energy endeavor
for the brain. We never see areas where there's more
(18:41):
activity when you tell the truth.
Speaker 1 (18:43):
There's only more activity when you're lying.
Speaker 2 (18:47):
So the idea that it's easier to tell the truth
holds as a rule in life, presumably because it holds
as a rule in the brain. And if the cooked
up story doesn't go so per we end up painting
ourselves into a corner, which brings with it lots more
brain energy expenditure. And this observation is one that the
(19:10):
Scottish poet novelist Sir Walter Scott immortalized in his poem
Marmion in eighteen oh eight. This is where he penned
the immortal line, Oh what a tangled web we weave
when first we practice to deceive. He was pointing to
a common problem that arises when we make up a
(19:30):
lie to cover a secret, and then that leads to
another necessary lie, which leads to another, and now you
have to keep track of this whole web of lies
to try to keep your story consistent, which can lead,
as Sir Walter Scott may have intuited, to a high
cognitive load. And cognitive load matters because it equates to
emotional stress. We all know from experience that holding on
(19:55):
to a secret creates anxiety, and often these bad feelings
increase our urge to confess or share the secret to
relieve the burden. So secret keeping requires a lot of work.
But what I want to highlight is the double edged
nature of the secret because, perhaps surprisingly for better or worse,
(20:18):
the ability to keep a secret is one of the
main drivers of the shape of our species and society. Now,
we can talk all day about the downsides of keeping
secrets from one another, but in fact, the ability to
keep secrets seems to have evolved many millions of years
ago as a survival mechanism, as it turns out, concealing
(20:41):
information about food sources or about shelter or defensive strategies.
This is often what made the difference between life and death.
And it's not just humans who practice deception. Other primates
do this as well. Now I'm not talking about deception
like using camouflage. Instead I'm talking thinking about active, clever
(21:02):
deception that requires theory of mind, in other words, the
ability to step into the perspective of another to understand
what they know and don't know. It turns out that
only primates seem to do this, and in fact they're
constantly doing things to deceive one another. On this note,
here's a short clip from my interview with Max Bennett
(21:23):
from last year, who wrote a great book called A
Brief History of Intelligence. And here's Max describing an example
of deception in chimpanzees.
Speaker 3 (21:32):
So you can see non human apes do things like
they will hide transgressions from other people to try and
prevent themselves from getting in trouble. There's this famous study
that I love by Emil Menzel. I think it was
in the seventies where he put two chimpanzees in the
sort of one acre forest and he showed the location
of treats to one of the chimpanzees named Belle, and
(21:56):
she initially would share the treat with another chimpanze he
named Rock, but then Rock started just stealing the treat
from her. So what she started doing is when she
knew the location of the treat, she would wait for
a rock to look away, and then she would run
over and grab it. So then rock in response to this,
decided to pretend to look away so that when she
(22:17):
started running, then he would turn around and run. Then
in response to this, what she would do is she
would pretend to run in the wrong direction, lead him
to the wrong place, and then run back.
Speaker 2 (22:26):
So the tricky deceptive capacity we have to not spill
the beans enhances the chances of individual survival. Now, when
we zoom in on humans and study them very carefully,
what we find is that keeping secrets and more generally,
lying is a cognitive development in children. In other words,
(22:49):
And first young children.
Speaker 1 (22:51):
Don't know how to lie.
Speaker 2 (22:52):
And then they hit a milestone one day, let's say
between two and four years old, and they get a
great idea, maybe they can assert something that's not true
and maybe the other person won't know. So they carefully
float this trial balloon and sometimes it works. Suddenly they
have eaten the cooky but avoided punishment, so they start
(23:16):
to lie. As soon as they develop theory of mind,
they start to understand that their thoughts and minds are
separate from those of their parents. They also begin to
realize that people aren't perfect mind readers, so the child
can say something that isn't true, and miraculously the other
person doesn't always know. Obviously, sometimes the child doesn't get
(23:39):
away with it, and so they conclude that they need
to work even harder to cook up better fabrications that
might fool the other person. And for this they need
to get better and better at stepping into the other
person's shoes to understand what that person might fall for.
So children start to lie more frequently as they get
older and they get better at lying. They learn to
(24:01):
match their facial expressions and tone of voice to what
they're saying, and they shape their lies based on their
assumptions of what the other person may or may not know.
And lying sucks, especially when it's in your own kids.
But I'll just mention that some researchers view lying as
an important developmental stage that indicates early intelligence. So while
(24:24):
none of us appreciate being lied to by our kids
or having secrets kept from us by our colleagues, it's
one of the signatures of the extraordinarily intelligent primate brain.
And in fact, when you look across human history, the
thing that becomes clear is that secrets quickly expanded beyond
the individual level to the group level. So trust within
(24:48):
small groups usually depended on the ability to keep certain
knowledge hidden from outsiders, and by the way, keeping internal
group secrets has the added benefit of thought during a
sense of belonging in mutual protection and secrets have always
been a key part of military strategy. The immortal book
(25:09):
The Art of War by Sun Su paints a picture
of deception and secrecy as absolutely essential for successful warfare,
and more generally, secrets have always been a lever for
power within larger groups. As civilizations grew, secrets became pervasive
(25:29):
wherever people maintained and exercised power. You see this all
over ancient empires like Egypt and China, where they leveraged
secrecy to control knowledge and to reinforce hierarchies. And as
we all know from our history classes and from Game
of Thrones, the secrets swirling around among advisors in royal
(25:52):
courts often dictated the rise and fall of leaders. Now
people have been grappling with the tension between keeping secrets
in a society versus telling everything to the public, and
they've been debating these points for a long time. So,
for example, the Enlightenment was characterized by arguments going on
about the value of secrecy versus transparency. You have some
(26:16):
philosophers like Jeremy Bentham who hated the idea of secrecy
in governance, but secrecy has always persisted in areas like
diplomacy and statecraft, where its defenders deem it as essential
for stability. And you can see the fingerprint of this
very old debate in modern questions about surveillance. To what
(26:39):
extent should a government be able to keep track of
who's doing what with cameras or data tracking? Should anyone
be able to keep total privacy from the states such
that I can work on three D printing guns and
making nuclear bombs in the privacy of my home. These
debates will presumably go on forever because because there is
(27:00):
no single right answer for balancing the privacy of the
individual against the security of the society. So back to secrets,
it's not just governments that traffic in these. Religious traditions
almost always involve sacred secrets accessible only to initiates or clergy.
(27:21):
In ancient Greece, there were rituals performed every year called
the Eleucinian Mysteries, and to participate you had to swear
a vow of secrecy, and in return you were promised
spiritual enlightenment. And you then learned and safeguarded the secret rights.
And these were kept so secret that all these centuries
later we still don't know what happened in those rituals.
(27:44):
Or similarly, in the medieval era, knowledge of alchemy or
mystical texts was always hidden so that it was accessible
only to a select few. More generally, this is the
kind of glue that keeps people locked into whether that's
religions or organizations or fraternities. It's the tension from letting
(28:06):
them know that as they level up, they'll get access
to the next big secrets, so they'll be part of
the in group, and they'll link arms to keep those
secrets from everyone else. As I mentioned a moment ago,
sharing a secret builds strong bonds between individuals, It fosters intimacy,
and it creates alliances. I'll just mention one more thing
(28:46):
about secrets on a societal level. Beyond warfare and politics
and religions, subgroups in society have to keep secrets to
maintain competitive advantages. In technology. Take something like trade secrets. Historically,
you've got things like the exact technique for making Chinese silk,
(29:07):
or currently the exact formula for Coca cola, or the
software details in cybersecurity firms. In all of these cases,
knowledge is closely guarded to keep a competitive advantage, and
people have studied the issue about keeping secrets even inside
a company, like the upper management knows something but the
(29:29):
employees don't, so that sounds bad. The researchers at Stanford
just published a paper on the benefits that companies derive
from what they call the confidentiality effect. The research suggests
that keeping company secrets has the benefit of boosting feelings
of privilege and status among those who know the secret
(29:51):
and protecting those who do not. Now, I want to
tie this into something I talked about before in the
episode titled why do you Brains love conspiracy theories? The
brain is always seeking explanations to reduce cognitive dissonance and
provide a sense of certainty. And when we solve puzzles,
(30:12):
that's deeply rewarding. And that's why uncovering patterns, even ones
that aren't accurate, that's why that feels so rewarding. Now,
one can never say that all conspiracy theories are false
because certainly, at moments in history, people can try to
get away with something, and they are very incentivized to
(30:32):
cover their tracks. But most conspiracy theories fall apart under scrutiny.
Why we'll just think about conspiracy theories through the lens
of human behavior and probabilities. Imagine a grand conspiracy like
faking the moon landing or hiding evidence of alien life
(30:53):
in Roswell. These cover ups would require hundreds, if not thousands,
of people to remain silent across agencies, across generations, across families.
But real life doesn't work that way. People get drunk,
they talk in their sleep, they have moments of guilt,
they decide to confess on their deathbeds. Secrets are fragile,
(31:16):
and the longer they need to be kept, the more
likely they are to leak. Now, you can model the
odds of someone spilling the secret using a math technique
called survival analysis, and it goes like this. Say there's
a tiny daily chance that someone involved lets something slip.
Let's say it's really small chance, like one in one
(31:38):
thousand on any given day that they'd screw up, but
multiply that over hundreds of people in years of time
and the probability of the secret being kept approaches zero.
You're rolling the dice over and over that somebody is
going to spill the beans. And note that with most
conspiracy theories, the incentive to spill the beans is usually compelling.
(32:02):
In the end, people typically act in their self interest,
and people who are in on a conspiracy sometimes find
the rational choice is to defect before someone else does.
In other words, if you are the one who spills
the big secret, let's say of a fake moon landing
or a planned murderer of JFK, then suddenly, not only
(32:26):
have you released the stress of a secret, but more importantly,
you're now famous and protected by the legal system and
probably signing book deals with Penguin Random House. So game
theory asks why risk being the one left behind when
you have so much incentive structure encouraging you to blurt
out the truth before someone else does. So, when you're
(32:49):
confronted with a conspiracy theory, the question isn't just could
that happen? But more importantly, how many people would need.
Speaker 1 (32:58):
To keep the secret?
Speaker 2 (33:00):
And how much time has passed without anyone defecting. And
this is why when you do the math and factor
in human behavior, the vast majority of conspiracies become not
just improbable, but implausible. So now I want to return
to a question that I posed at the very beginning.
Can a computer keep a secret? Well, a toaster doesn't
(33:24):
have the circuitry to keep a secret from you, presumably
because it doesn't have complex enough circuitry, it doesn't have
a team of rivals fighting it out under the hood.
Speaker 1 (33:35):
But could AI keep a secret from you?
Speaker 2 (33:39):
Well, people started asking this question some years ago using
what are called adversarial networks. Just think of this like
networks that are pitted against one another, and they each
might have a different incentive structure. So back in twenty sixteen,
a group of researchers a Google Brain, wanted to know
whether neural nets networks could develop their own encryption methods
(34:03):
independently without human guidance. So they set up three independent networks.
The first was named Alice, and Alice's task was to
send a secret message to Bob. This second network, Bob
His job was to decode the message sent by Alice.
And then there was a third network named Eve. Now
(34:25):
Eve represented and evesdropper. Her role was to try to
intercept and decrypt Alice's message, so they set these three
networks running. Alice and Bob were trained collaboratively to ensure
Bob could correctly decode the messages, and Eve was trained adversarially.
Her job was to improve her decryption attempts, and the
(34:49):
system evolved dynamically. As Alice and Bob became better at
secure communication, Eve became more sophisticated in breaking their encryption,
forcing Alice and Bob to improve further. Now, the outcome
was kind of incredible because Alice and Bob's neural networks
learned to create their own encryption strategies, and this was
(35:13):
without explicit programming. In other words, the encryption strategy was
never told to them by a human. So over time
those two succeeded in creating encrypted communications that Eve couldn't decrypt.
They outpaced Eve's efforts. The key here is that the
encryption methods developed by Alice were not human readable or interpretable.
(35:38):
The system invented its own cryptographic methods based on its
training objectives very rapidly. The two neural networks figured out
how to keep a secret not only from Eve, but
from everybody. They developed a totally novel and dynamic cryptographic system.
So that was one study. Then in twenty seven, the
(36:00):
Facebook AI research group was working on developing AI agents
that could negotiate. But when they left these agents to
their own devices, they saw unexpected behaviors.
Speaker 1 (36:12):
So here was this setup.
Speaker 2 (36:13):
The goal of the experiment was to train these two bots,
these two AI agents, to negotiate with each other to
achieve their objectives. So the two AI agents are tasked
with dividing items between them, like books, or balls or hats.
Each item has a different value to each of the bots,
(36:35):
and each bot has the objective to maximize their score
by getting the items that they value the most, and
they're rewarded based on how well they do. Okay, so
they set these bots running, but pretty soon two surprises emerged.
The first was that the bots started using tricky tactics.
They would sometimes pretend to be uninterested in a particular
(36:59):
item to convince the other bot to concede it. This
is called feigned disinterest, and of course humans do this
in negotiations to manipulate the perceived value of something that
they are interested in, but it certainly wasn't expected that
this would emerge on its own in an artificial neural network.
(37:19):
The second thing was that the bots started to communicate strangely.
They developed their own shorthand language to negotiate. So Bot
one would say I want want, want want one book
and bought two would respond okay, you you you you
two balls. Now why did their language start to turn
(37:39):
funny like this? It's because the AI wasn't explicitly incentivized
to stick with human readable language. Instead, the reward system
only focused on achieving successful negotiations, so the bots optimized
their communication for efficiency rather than for clarity to humans.
(38:01):
So it made sense to the bots, but eventually became
unintelligible to the human reader. Now, this experiment gives us
a few deep insights.
Speaker 1 (38:10):
The first thing it.
Speaker 2 (38:11):
Shows us is emergent behaviors. AI agents can develop creative
strategies to achieve their goals, and this is typically not
anticipated by the researchers. In this case, what developed was
something like a secret, or at least we would say,
a non truth, because the AI would pretend one thing
(38:33):
when it wanted something else. The experiment also demonstrated that
if it's not explicitly constrained, AI will optimize for its
objectives in ways that maybe don't align with human readability,
and if you extrapolate this, you can see this is
another way that AI might keep a secret from us,
even if it's not intending to. Just like the networks
(38:56):
Alice and Bob, it might just learn how to speak
a language that we simply don't understand. So let's wrap
up today's exploration of secrets and the brain. As we saw,
secrets of something we're often embarrassed to address, but they're
a massive feature of primate lives, shaping everything from personal
(39:17):
relationships to large scale politics. The act of keeping a
secret emerges from the rivalry within our brains, a team
of neural networks vying for control over what's revealed and
what's hidden. This internal competition creates the cognitive and emotional
weight of secret keeping, which can sometimes push us to
(39:41):
seek relief by confessing to strangers, or by writing anonymous
posts on the web, or even whispering to an AI therapist.
The relief that we feel from telling a secret comes
not from solving the external problem, but from reducing the tension.
(40:02):
And as we saw, secrets don't stop at the individual level.
They've been a key driver of human society, from military
strategy and political hierarchies to the rise of religions and
social institutions. Secrets bind groups together, They foster trust among allies.
(40:22):
They create competitive advantages, whether that's the formula for Coca
cola or the sacred rights of ancient mystery cults. But
they also lead to struggles over societal transparency, and there's
always a healthy debate about how much governments should be
allowed to conceal from us.
Speaker 1 (40:42):
And what about AI.
Speaker 2 (40:44):
I told you about two experiments where neural networks, left
to their own devices, developed encryption methods and private languages
that humans couldn't decode. These aren't conscious secrets in the
way we experience them, but they light a future where
machines might possess knowledge inaccessible to us, not necessarily because
(41:06):
they're deliberately hiding it, but because their optimization processes have
outpaced our ability to understand. As aisystems grow more sophisticated,
the question of what they know and what they're going
to keep from us is only going to evolve. So
secrets are woven into the fabric of our minds and societies.
(41:30):
They've been with us since before the sunrise of human cognition,
As we see from our primate cousins, and secrets and
lies may soon take on new dimensions as we come
to share our world with increasingly intelligent machines. Whether secrets
are helping us to bond or they're dividing us, they
are a reminder of the complexity of our inner cosmos
(41:55):
and the tangled web of our relationships with each other
and soon with our technology. Go to eagleman dot com
slash podcast for more information and to find further reading.
Send me an email at podcast at eagleman dot com
(42:15):
with questions or discussion, and check out and subscribe to
Inner Cosmos on YouTube for videos of each episode and
to leave comments.
Speaker 1 (42:25):
Until next time.
Speaker 2 (42:26):
I'm David Eagleman, and this is Inner Cosmos.
Host
David Eagleman
Popular Podcasts
On Purpose with Jay Shetty
I’m Jay Shetty host of On Purpose the worlds #1 Mental Health podcast and I’m so grateful you found us. I started this podcast 5 years ago to invite you into conversations and workshops that are designed to help make you happier, healthier and more healed. I believe that when you (yes you) feel seen, heard and understood you’re able to deal with relationship struggles, work challenges and life’s ups and downs with more ease and grace. I interview experts, celebrities, thought leaders and athletes so that we can grow our mindset, build better habits and uncover a side of them we’ve never seen before. New episodes every Monday and Friday. Your support means the world to me and I don’t take it for granted — click the follow button and leave a review to help us spread the love with On Purpose. I can’t wait for you to listen to your first or 500th episode!
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com