
hello I’m building a model of the bee
brain and I was at a conference recently
talking about what I do and I got a
question do bees have brains and I went
off on this mortified pearl clutching
tirade of how could anybody in 2017
possibly imagine that bla bla bla bla
bla bla bla
and as usual it took my husband to talk
me down and point out that there are
plenty of good reasons why intelligent
well-educated normal people emphasis not
you may not have thought much about bees
and their brains so this is my attempt
to explain what I’m doing and why now
that’s a model of the sorry that’s an
image of the bee brain created by
Angelique Polk and two key you into this
scale that black bar at the bottom is
one millimeter so we’re looking at a
millimetric Li scaled structure but what
I want you to focus on is first of all
the intrinsic beauty of this but also
its complexity so the bee brain is is
small but the bee brain is not simple
the bee brain like all brains is made up
of neurons neurons are just nerve cells
which are cells that are specialized to
make connections and transmit
information and even in a brain this
small there’s just shy of 1 million
neurons now I’m making a model of this
now that doesn’t mean I’m sitting in a
shared making it out of matchsticks I I
would like to do that but I wouldn’t get
paid to do that so my model looks like
this so this is a model I’ve made with
my colleague Alex cope of part of the B
brain it’s a high-level abstraction it’s
a computational model where each neuron
or group of neurons is represented as a
connection that transmits information so
it’s an it’s a model of the brain as an
information flow Network now why am I do
this well I wanted to think back to you
the last time you had an encounter with
a B however that went for you whether
you may have got stung that was probably
your fault but what you will have
realized in that is that bees are smart
they are incredibly smart and to show
you just how smart look at this video
from last chick ker Ollie Luke Ola and
Clint Perry
so this bee has learned that if it moves
that yellow ball into the yellow circle
the well beneath the ball fills up with
nectar and it gets a drink and that’s a
bee now if you wanted to you could
describe that as a form of tool use and
all of that intelligence all of that
smarts come somehow from the bee brain
and I want to understand this be level
of intelligence but to do that I’ve got
to understand how the bee brain works
and that’s a mighty question even for
something like a bee brain so how can we
go about dealing with a question with
that kind of complexity well one
approach would be we could take the
brain to simpler bits we could work out
what the bits do and then put it back
together the problem with that is for
brains what matter are the connections
when you take it apart there comes a
point where you realize all you’ve done
is get good at breaking it so in other
approaches we could map it we could we
could learn the structure of the brain
because a map is a form of understanding
and the nice thing about contemporary
neuroscience is we can do that we’ve got
techniques now to build incredibly
detailed maps of the brain but brains
are dynamic structures it’s dynamic
information flow and a map therefore
doesn’t give you the dynamics it’s only
part of the story
so another approaches we could observe
brains working and with contemporary
neuroscience we can do that we can look
at living neurons in living brains of
living animals that doing their thing
the problem with that is there were so
many neurons with so many connections
that you really quickly almost start
drowning in the volume of data that
you’re generating there is so much
detail it’s hard to step back
and find the meaning of your system from
the mass of detail of all of these newer
ones with all of these complex
connections and that’s where the
modeling approach really comes in so
model is is really scientific jargon for
best guess it’s a hypothesis so when
you’re building a model what you do is
you take everything that you know about
the brain and you ask how could this
work as a system the things that are
known are what model is call constraints
because the rule of modeling is you
can’t make anything up you can’t imagine
connections that don’t exist you can’t
imagine neurons that aren’t there so
within those constraints of the biology
that constrains the kind of model you
can make and then you ask yourself okay
given these constraints how could this
work as a system and that becomes your
model and once you’ve got your model
then your job is to test the model and
that’s also not straightforward to test
a model you’ve got it on a different
thought experiment you ask yourself if
this models right what have we learned
from it what can we infer from it and
then you test whether that inference is
actually true so to give you an example
I want to talk about the work of Monday
I’m suin Avastin who gave a brilliant
TED talk a couple of years ago now
swinney studies insect flight control
and he had a model of how insects
control their behavior in flight is
their coming into land but one of the
weird consequences of Sweeney’s model
was that if his model was right he
proposed that insects aren’t judging
distance in meters or kilometers they’re
judging distance by how much they’ve
seen which sounds kind of weird to us so
Sweeney designed an experiment to test
how insects judge distance and found yet
actually insects aren’t judging distance
by distance they’re judging distance by
how much they’ve seen now that finding
didn’t prove that Sweeney’s flight
control model was right but it did mean
there was no reason for Sweeney to
imagine his model was wrong
and when you’re modeling being wrong is
good because of all the possible million
ways that I could imagine a brain to
work if I’m wrong about something I can
categorically rule out all the
possibilities I have now proven to be
wrong and I’ve also learned a new
constraint so what I can do is recast my
model taking into account this new
constraint and try again and by
iterating that process you incrementally
step closer and closer towards the truth
of the system but along the way you’re
also learning what matters for your
system from what’s just detail so along
the way you’re approaching the truth but
you’re also learning the meaning from
the detail now the beauty of this be
brain modeling project is what we will
learn from it
so we will learn how intelligent
self-directed autonomous behavior
emerges from the circuits of the B brain
now that’s just B level sure but this
project is not going to stop with a B
because once we’ve got a model of a B
brain we can use that as its self a
model to explore how larger and more
complicated brains could work including
our own now I’m not saying that the
human brain is just a Big B brain that’s
absolutely not what I’m saying
but our brain like the B brain is made
out of neurons and the neurons in our
brain the way they work is actually very
similar to the way the neurons in a pea
brain work and the way our neurons make
connections and transmit information is
actually very similar to the way neurons
make connections and transmit
information in the B brain but here’s
the kicker our brain is 80 thousand
times larger than the B brain so our
brain contains eighty thousand million
neurons that’s an eight with ten zeros
after it so just take a moment to
appreciate the scale of that number now
that is a legitimately dizzying level of
complexity and I would argue that we’re
never going to understand a system that
complex by brute
for Salone we’ve got to be smart and one
way that we can be smart is we could
take our understanding of the be brain
and use that as a model for certain
aspects of the human brain so the B
brain model will become our first best
guess it will become our way in to start
to understand the complexity of the
human brain and ultimately that’s why
I’m building a model of the B brain it’s
not just to understand B’s in their
behavior it’s because ultimately
modeling the B brain is going to help us
understand the human mind thank you
you
brain and I was at a conference recently
talking about what I do and I got a
question do bees have brains and I went
off on this mortified pearl clutching
tirade of how could anybody in 2017
possibly imagine that bla bla bla bla
bla bla bla
and as usual it took my husband to talk
me down and point out that there are
plenty of good reasons why intelligent
well-educated normal people emphasis not
you may not have thought much about bees
and their brains so this is my attempt
to explain what I’m doing and why now
that’s a model of the sorry that’s an
image of the bee brain created by
Angelique Polk and two key you into this
scale that black bar at the bottom is
one millimeter so we’re looking at a
millimetric Li scaled structure but what
I want you to focus on is first of all
the intrinsic beauty of this but also
its complexity so the bee brain is is
small but the bee brain is not simple
the bee brain like all brains is made up
of neurons neurons are just nerve cells
which are cells that are specialized to
make connections and transmit
information and even in a brain this
small there’s just shy of 1 million
neurons now I’m making a model of this
now that doesn’t mean I’m sitting in a
shared making it out of matchsticks I I
would like to do that but I wouldn’t get
paid to do that so my model looks like
this so this is a model I’ve made with
my colleague Alex cope of part of the B
brain it’s a high-level abstraction it’s
a computational model where each neuron
or group of neurons is represented as a
connection that transmits information so
it’s an it’s a model of the brain as an
information flow Network now why am I do
this well I wanted to think back to you
the last time you had an encounter with
a B however that went for you whether
you may have got stung that was probably
your fault but what you will have
realized in that is that bees are smart
they are incredibly smart and to show
you just how smart look at this video
from last chick ker Ollie Luke Ola and
Clint Perry
so this bee has learned that if it moves
that yellow ball into the yellow circle
the well beneath the ball fills up with
nectar and it gets a drink and that’s a
bee now if you wanted to you could
describe that as a form of tool use and
all of that intelligence all of that
smarts come somehow from the bee brain
and I want to understand this be level
of intelligence but to do that I’ve got
to understand how the bee brain works
and that’s a mighty question even for
something like a bee brain so how can we
go about dealing with a question with
that kind of complexity well one
approach would be we could take the
brain to simpler bits we could work out
what the bits do and then put it back
together the problem with that is for
brains what matter are the connections
when you take it apart there comes a
point where you realize all you’ve done
is get good at breaking it so in other
approaches we could map it we could we
could learn the structure of the brain
because a map is a form of understanding
and the nice thing about contemporary
neuroscience is we can do that we’ve got
techniques now to build incredibly
detailed maps of the brain but brains
are dynamic structures it’s dynamic
information flow and a map therefore
doesn’t give you the dynamics it’s only
part of the story
so another approaches we could observe
brains working and with contemporary
neuroscience we can do that we can look
at living neurons in living brains of
living animals that doing their thing
the problem with that is there were so
many neurons with so many connections
that you really quickly almost start
drowning in the volume of data that
you’re generating there is so much
detail it’s hard to step back
and find the meaning of your system from
the mass of detail of all of these newer
ones with all of these complex
connections and that’s where the
modeling approach really comes in so
model is is really scientific jargon for
best guess it’s a hypothesis so when
you’re building a model what you do is
you take everything that you know about
the brain and you ask how could this
work as a system the things that are
known are what model is call constraints
because the rule of modeling is you
can’t make anything up you can’t imagine
connections that don’t exist you can’t
imagine neurons that aren’t there so
within those constraints of the biology
that constrains the kind of model you
can make and then you ask yourself okay
given these constraints how could this
work as a system and that becomes your
model and once you’ve got your model
then your job is to test the model and
that’s also not straightforward to test
a model you’ve got it on a different
thought experiment you ask yourself if
this models right what have we learned
from it what can we infer from it and
then you test whether that inference is
actually true so to give you an example
I want to talk about the work of Monday
I’m suin Avastin who gave a brilliant
TED talk a couple of years ago now
swinney studies insect flight control
and he had a model of how insects
control their behavior in flight is
their coming into land but one of the
weird consequences of Sweeney’s model
was that if his model was right he
proposed that insects aren’t judging
distance in meters or kilometers they’re
judging distance by how much they’ve
seen which sounds kind of weird to us so
Sweeney designed an experiment to test
how insects judge distance and found yet
actually insects aren’t judging distance
by distance they’re judging distance by
how much they’ve seen now that finding
didn’t prove that Sweeney’s flight
control model was right but it did mean
there was no reason for Sweeney to
imagine his model was wrong
and when you’re modeling being wrong is
good because of all the possible million
ways that I could imagine a brain to
work if I’m wrong about something I can
categorically rule out all the
possibilities I have now proven to be
wrong and I’ve also learned a new
constraint so what I can do is recast my
model taking into account this new
constraint and try again and by
iterating that process you incrementally
step closer and closer towards the truth
of the system but along the way you’re
also learning what matters for your
system from what’s just detail so along
the way you’re approaching the truth but
you’re also learning the meaning from
the detail now the beauty of this be
brain modeling project is what we will
learn from it
so we will learn how intelligent
self-directed autonomous behavior
emerges from the circuits of the B brain
now that’s just B level sure but this
project is not going to stop with a B
because once we’ve got a model of a B
brain we can use that as its self a
model to explore how larger and more
complicated brains could work including
our own now I’m not saying that the
human brain is just a Big B brain that’s
absolutely not what I’m saying
but our brain like the B brain is made
out of neurons and the neurons in our
brain the way they work is actually very
similar to the way the neurons in a pea
brain work and the way our neurons make
connections and transmit information is
actually very similar to the way neurons
make connections and transmit
information in the B brain but here’s
the kicker our brain is 80 thousand
times larger than the B brain so our
brain contains eighty thousand million
neurons that’s an eight with ten zeros
after it so just take a moment to
appreciate the scale of that number now
that is a legitimately dizzying level of
complexity and I would argue that we’re
never going to understand a system that
complex by brute
for Salone we’ve got to be smart and one
way that we can be smart is we could
take our understanding of the be brain
and use that as a model for certain
aspects of the human brain so the B
brain model will become our first best
guess it will become our way in to start
to understand the complexity of the
human brain and ultimately that’s why
I’m building a model of the B brain it’s
not just to understand B’s in their
behavior it’s because ultimately
modeling the B brain is going to help us
understand the human mind thank you
you
Please follow and like us: