Welcome to the lab!
I'm Dr. Rhanor Gillette,
Emeritus Professor of Molecular and Integrative
Physiology at the University of Illinois
I'll be showing you around the equipment and
teaching you how everything works here.
Our research focuses on the brain of an animal called
about the size of a grapefruit. It lives in the Pacific
Ocean, where it spends its time crawling along the ocean
floor hunting tasty morsels. While it will try to eat
anything that moves, its most common prey is other smaller
sea slugs like Flabellina and Hermissenda.
Its brain is pretty small and simple — altogether, it's only
several thousand neurons. And yet with this tiny brain, the
Pleurobranchaea performs some relatively
nontrivial decision-making, and is even capable of
learning from experience! If we can crack exactly how it does this,
we might learn a lot about how our own brains work —
the human brain has over ten billion neurons, but
the principles of operation aren't necessarily all that different.
We've charted enough of this creature's brain to build a
software simulation of some of its decision-making circuitry.
That's what you're looking at right now.
In this computer model, the Pleurobranchaea
forages amongst three different kinds of prey.
are nutritious and lack natural defenses, so they're
Cyberslug's favorite food.
are just as nutritious, but have toxic spines.
Cyberslug will only approach them if it's incredibly
is a Batesian mimic. It exudes a chemical odor signature
(i.e. smells) like Flabellina,
but doesn't have those toxic spines.
With only a handful of neurons controlling its actions,
the Pleurobranchaea is able to strategize
about what to eat and where to go. By making the right
choices, it's able to make optimal use of its food
resources, acquiring enough nutrition to survive while
generally avoiding harm as much as possible.
Cyberslug lets you watch this decision-making process in
real time. The equipment around you gives you an inside
view into the slug's decision-making processes, opening
a window into the neural pathways that make this
animal's behavior possible.
In our lab, we think this is pretty interesting in its own right!
But we also believe that a software simulation of the slug's
simple yet powerful neurocircuitry could yield applications
in robotics and intelligent agent design.
Well I've certainly done enough talking for now! I'll let you
play around for yourself.
I'll stick around nearby. If you have any questions about
any of the equipment, simply hover over any control
and I'll pop up and tell you all about it.
Once you don't need me anymore, flip the "Info" switch below you
(the one with the
icon), and I'll get out of your hair.
Have a good science!
This dial adjusts the audio levels of the Cyberslug 2017 lab.
It has four notches.
This software presents studies conducted by
in the School of Molecular and Cellular Biology
at the University of Illinois at Urbana-Champaign.
My research team includes Graduate Research Assistant
Ekaterina D. Gribkova
Cyberslug was developed by Mikhail Voloshin,
a computational neuroscience graduate student of mine from
back in 2000. He's since gone on to work at
Microsoft, Google, a few dot-coms, and
a couple of hedge funds.
He's even written a novel!
You should go check it out!
Flip this switch if you don't want to keep seeing me
pop up here anymore. Dont worry, though — I won't be far!
You can always flip it back on if you need me again.
Our research has shown that, like a very simple robot,
the real-life Pleurobranchaea runs in
discrete modes of operation. Two of those modes are
regulated by an internal "appetitive state switch".
To put it simply, you can think of these modes as
"Hungry" and "Not Hungry".
When the slug is sated, it actively avoids all food sources.
Indeed, when this "Hungry" light is off, try
picking up the slug and dragging it near a tasty green
, which has no natural defenses and represents
nothing but pure nutrients.
Ryan Gosling refusing to eat his cereal
, the slug will turn away from the delicious morsel.
Strategically, there are several good reasons for this.
One is that it's generally better to eat as few prey
animals as you need, in order to preserve the prey population
for the future — predators that are too aggressive
risk exterminating their entire food supply. Another is
for safety reasons — if the slug smells something
yummy nearby, then it's possible that other predators smell
it too, and if the slug is sated then it really doesn't
need to risk a potentially lethal fight over a completely
superfluous pack of calories.
Joe Pesci in a Snicker's commercial
the slug acts different when it's hungry. When its stomach
is growling (yes, it has a stomach; no, it doesn't really growl),
it goes straight for the best meal it can find.
"Best" is a
combination of primarily three factors: how far away the prey is,
how nutritious the prey is, and how much of a fight the prey
is expected to put up (i.e. what kind of risks may be involved).
Usually, the slug will choose the safest meal it can reach, so
it'll tend to prefer a distant
over a nearby
But as it gets progressively hungrier, it starts to care less and
less about the risk — it needs those calories now!
Let it get hungry enough, and it'll devour Hermi and Flab
alike, spines be damned.
We know that
humans act more or less the same way
. That's why we research these animals —
because, for all our language and technology and culture and art
deep down we're really not all that different.
This readout shows how many organisms of each species
the slug has eaten.
You'll probably notice that, over time,
the slug eats more
than any other prey type, even if Hermis are relative
rare and sparse in the environment. This is a rather
big deal! It means that the slug is correctly avoiding
seeing no point in risking its safety on
Flabellina's natural defenses except in extreme
caloric emergencies — which, if the slug is
doing its job of foraging correctly, should be rare.
This row of buttons allows you to start and stop the simulation.
These dials allow you to set the populations of
Each species can have between 0 and 15 specimens.
Specimens don't really die when they get eaten. They get
taken to a happier place. That happier place is somewhere
else in the environment far away from the Cyberslug.
That is, their positions get randomly regenerated, effectively
reincarnating each eaten animal. Therefore, the number of
specimens of each species stays the same throughout the
run of the simulation.
Your changes will take effect after you hit the "Stop" button.
You'll see your new prey populations the next time you
run the simulation.
This dial lets you see where the slug has traveled in the last
few hundred turns. The higher you turn the dial, the more of the
slug's foraging history you can see.
All living things exude a telltale chemical cocktail into their
environment. We humans recognize this as a "scent". And though
humans are primarily visual and auditory creatures, for most
organisms on Earth the olfactory sense is their foremost way
of sensing the world around them.
Such is the case with the
Pleurobranchaea. It is, for all practical purposes,
deaf, dumb, and blind — it gets around purely by sense
of smell and touch. And though a sea slug is unlikely to
dethrone Elton John as the reigning pinball champion of Sussex
anytime soon, it can at least use its skills to keep itself
from starving among the benthic fauna of the Pacific ocean floor.
The slug is able to home in on its prey by following
the source of their respective odors.
The way it does this is remarkably simple.
See those two antenna-like
protrusions on the slug's front? Those are stalks of chemoreceptors
— essentially, they're like our nostrils, only inside-out.
The slug compares an odor signature from the chemoreceptors on its left,
to the odor signature from the chemoreceptors on its right.
Whichever side the odor signature is strongest on, that's the direction
that the slug turns.
This is partly why we chose this animal for research, and why it's
well-suited for software simulation. This compare-left-right-and-turn
algorithm is so straightforward, so easy to build with even
the most rudimentary components, that it's often
given to middle-school children as a kid-friendly robotics project.
It should be no surprise, then, that we can find it in nature.
Though the actual odor signatures of real-world animals consist of
up to thousands of different peptide chains, this Cyberslug equipment
simplifies them for easy visualization.
The defenseless and nutritious
has an odor gradient depicted by a green diffusion.
A hungry slug should move in the direction where the
green is denser.
The spiny, well-defended
has an odor gradient depicted by a red diffusion.
A slug should move away from a direction in which the red
is denser. It can make exceptions, however, in situations
when it's very low on nutrients and at risk of starving to death.
Then, dealing with the Flab's defenses can be worth the cost.
The Batesian mimic
doesn't have any defenses, but it tries to protect itself
by exuding an odor that matches Flab's, i.e. a red
diffusion, thereby repelling all but the hungriest
This is actually a form of parasitism —
the Faux gains the protection of the Flab's spines without paying
the metabolic cost of having to grow any such structures itself.
And, like most parasitic relationships, it's harmful to
the Flabellina. Every now and then, a hungry slug
will eat a Faux and realize it's harmless, which will
cause it to have less fear of things that smell "red" —
which will result in several Flabs getting eaten while
the slug, getting stung, re-learns a sense of caution.
Lastly, betaine is worth mentioning only because
the Pleurobranchaea's betaine receptors are
directly connected to its "attack" reflex.
is a simple organic compound exuded by a wide variety
of invertebrate sea life. Some of this life is
prey, while some could be predators — larger
sea slugs, for example, looking to make the
Pleurobranchaea their dinner. In either event,
when the smell of betaine gets particularly strong,
it can only mean one thing: There's something nearby,
and it needs to be bitten immediately!
Try picking up the Cyberslug and dragging it into a patch
of betaine — that is, close to any prey. If it's not
hungry (see the tutorial about the "Hungry" light in the
lower right corner), it'll try to turn away. But if you force
it to get close against its will, you'll see the Cyberslug
retract its oral frond and extend its proboscis, viciously
biting and rending with all its might! Which, granted,
isn't much — the slug is a hydrostatic creature with
no solid parts, and its bite can't even damage human skin.
But it still feels kind of gross, so I wouldn't recommend it.
This gauge shows the slug's current nutrition levels,
which is a rough approximation of the animal's net
available caloric reserves.
In the real world, particularly in complex animals like
humans, this can't really be broken down into a single
number. Our "nutrient levels" include things
like blood glucose, adipose tissue, protein-vs-carbs
ratio, vitamins, and so on. There are entire fields of
biology and medicine revolving around how humans
Fortunately, the Pleurobranchaea's metabolism
isn't nearly as complicated. It turns out that just
having a single meter labeled "Nutrition", representing
a general abstraction of glucose and amino acid reserves,
makes for a pretty adequate approximation.
Naturally, the Nutrition gauge depletes as time passes. In theory,
a slug with a completely depleted Nutrition gauge should slow down,
weaken, and die of starvation. In practice, we don't implement
that because we don't really learn a lot from it. Also, it's
This gauge displays the slug's level of satiation
— how "full" it feels, or how strongly its
"I'm good for now and don't need to go foraging
at the moment" receptors are firing.
Satiation is essentially redundant with Nutrition
— while the Nutrition gauge moves linearly,
the Satiation gauge moves along a sigmoid function.
The only reason to represent it separately is
to emphasize that Satiation is a utility function
on Nutrition — that is, while Nutrition is
more or less of a straight line, Satiation has
distinct positive and negative states. This is
especially relevant when computing whether
or not the slug is in "Hungry" mode —
which you can get more information about by
hovering over the "Hungry" light in the lower
The slug's "Incentive Salience" gauge shows how strongly
the slug is compelled to turn toward or away from stimuli.
The "Incentive Salience" calculation takes into account both
the input from the slug's chemoreceptors, and also its
current Satiation state.
A very hungry slug may be strongly incentivized to turn
toward anything edible.
A very full slug, i.e. one that's just eaten,
will be strongly incentivized to turn away from new
sources of food. (If you want an explanation as to why
it would have evolved that behavior, hover over the "Hungry" light
in the lower right corner for a tutorial about the
slug's "appetitive state".) One that's only moderately hungry may
feel some incentive to turn toward a Hermi, but away from a Flab.
The "somatic map" is merely an indicator of which
direction the slug feels compelled to turn in.
A very negative value indicates that the slug is feeling
the intense urge to veer left; a very positive one
indicates a desire to turn right.
The "somatic map" is computed by taking the slug's
"Incentive Salience" level, and applying it to the signals
being received by the
slug's left and
right chemoreceptor stalks respectively. (That's what those
antenna-looking things on the slug's head are: they're its
nose. For more information, hover over the Hermi/Flab/βine
switches above to learn about how the Pleurobranchaea
navigates by scent.) Whichever side has the greater incentive,
that's the direction in which the slug turns.
The Vhermi and Vflab gauges measure
a pair of internal values that represent the slug's ability
to learn from experience.
Vhermi tracks a reward signal,
and represents a slug's eagerness to veer toward
the green-cloud odor signature of
a defenseless, nutritious
Vflab tracks a punishment signal,
and represents a slug's eagerness to veer away from
the red-cloud odor signature
a spiny, aggressive
In general, these gauges quickly both converge to 1 —
the slug quickly learns the obvious conclusion: that Hermi
scent trails are good to follow, while Flab scent trails are
good to steer clear of.
A noteworthy temporary exception to this rule happens when
a very hungry slug, out of desperation, tracks a red scent trail...
and eats not a Flab but its Batesian mimic, a
The Faux smells just like a Flab, but it lacks the Flab's
defensive spines. The Cyberslug expects to pay the price of
a certain punishment when it bites into the source of a "red"
odor signature, but instead it receives nothing but pure
This causes the Cyberslug to reduce its avoidance response from
the Flabellina (Vflab), sometimes by dramatic margins.
If it happens multiple times, the Cyberslug may lose its fear
of Flabellina altogether.
From an ecological standpoint, this illustrates the tenuous balance
that Batesian mimics must maintain with the creatures they're mimicking.
The Faux is a weak, helpless thing whose sole protection is that
predators recognize that it "smells like"
something else that's dangerous. If the Faux were to get too numerous,
then its predators would correctly "unlearn" the association between its odor
and danger. This, in turn, would not only be bad for the Faux, but for
the Flabellina as well.
After all, Batesian mimicry can be seen as a form
of parasitism; the Faux gains an "unfair" advantage over the Flabellina
by warding off predators with the Flabellina's scent but without paying
the metabolic cost of having to grow toxic spines. And, like most parasitic
relationships, it's harmful to the "host" — after all, if predators
like Pleurobranchaea learn not to fear the Flabellina's scent
because of the preponderance of Faux,
then that not only undermines the whole point of Faux's mimicry, but it
also doesn't bode well for the Flabellina.
This effect is worth seeing for yourself. Go up to the row of dials
marked "Hermi/Flab/Faux", and adjust the populations of the various prey
species. Then reset the simulation by hitting "Stop" and then "Play". Watch the
"Eaten" counters down at the bottom. Notice how, when Faux
are rare, the slug will prefer to eat Hermis, even if they're much less
common than anything else. But if the population of Faux is high, then
notice how Vflab never gets particularly high,
and the prowling Cyberslug
devours Hermi, Flab, and Faux alike.