Danielle Bassett – “Networks Thinking Themselves” (SFI Community Lecture)


good evening and welcome to the lens ik
I’m Chris campus I’m a professor of the Santa Fe Institute where I research
general laws in biology with the hope of being able to predict things like
evolutionary history and astrobiology the San Fay Institute has been hosting
these community lectures since 1986 and this series has produced lively and
diverse talks on a variety of cutting-edge research topics for 33
years now in addition to being very interesting the series serves as a
reflection of the diverse set of research activities ongoing at the
Institute activities that span the sciences and find commonality between
diverse ideas over the years SFI has brought thought leaders at the forefront
of science to the Santa Fe Institute and to Santa Fe for these lectures and past
lecturers have included Francis Crick Kenneth arrow and Leonard
Susskind amongst many many others the Santa Fe Institute has been able to
produce these high-quality lectures thanks to partnerships with our
community minded local organizations and with that I’d like to give a special
thank you to Thornburg investment management who has generously
underwritten this lecture series for the last five years without their support we
would not be able to host these lectures at all let alone continue to offer them
to Santa Fe at no cost so let’s think Thornburg we’d also like to thank the lensing
Performing Arts Center for their additional support of this series and
lastly we’d like to extend our thanks to a new supporter the Santa Fe reporter for tonight we are lucky enough to have
Danielle Bassett giving our lecture Danielle is a professor at the
University of Pennsylvania where she works on an enormous range of topics
Danny’s work spans condensed matter physics to neuroscience to social
systems and in each of these areas she brings new theoretical and mathematical
tools to bear an important pure and applied problems the questions that she
is interested in are some of the most interesting and modern science what are
human algorithm an algorithm for learning how can interventions be taken
on the dynamics of a brain how do we encode curiosity Danny will talk much
more about these topics tonight and I just want to mention that she has a very
long list of accolades which I couldn’t possibly list here tonight so if I sit
suffice to say that she these include winning a MacArthur Fellowship and
Alfred P Sloan fellowship and the air dish ready prize in network science I’m
very much looking forward to her talk and with that please help me in
welcoming Danielle Bassett thank you very much I’m very excited to be here
tonight thank you for the organizers and the donors who made this possible it’s a
really wonderful set of events and I’m excited to be a part of it
and most of all I’m very excited to meet all of you I think this is probably the
first time we’ve met at least for most of us I often get a little bit nervous
the first time I meet someone so I’ve read somewhere in a book that one a
really great thing to do if you’re nervous meeting someone for the first
time is to tell them a little bit about yourself
so I’ve decided to tell you a little bit about my
so I was home-schooled from kindergarten through 12th grade so all the way
through until I went to college and my parents had this really wonderful home
library that was had at least a thousand books the last time we counted and so I
I grew up really loving reading and loving all kinds of areas of science of
humanities of philosophy and in fact when I became around high school-aged I
really thought that I was going to be a philosopher um so there’s that and then
another interesting tidbit about myself is that my dad is an orthopedic surgeon
and when I was young even sort of seven years old he would come back with videos
of new surgeries they had to learn and he would play them on the local video
you know our videotape just I forget what they used to call them but anyway
and I would I would sit there next to him and watch these videos of these new
different kinds of surgeries that he had to do so I became really fascinated by
the human body and by the the beautiful mechanics of the human body and the idea
that a physician can intervene so carefully and so precisely to make
someone better so I became very interested in biology at that point as
well and then lastly it towards the later years of high school while
homeschooled high school I became very interested in mathematics and in physics
and I just thought it was so amazing that we could take these very simple
formalisms and formulations and use them to understand the world around us I was
just blown away by the fact that we can use these these simple mathematical
constructs to explain really complicated things from the universe to the human
mind so one thing I think that was a common theme across all of these
different interests was the mind so whether it was philosophy of the mind
biology of the minds or mathematics of the mind and ever since then I’ve been
really fascinated with the mind itself I’m trying to tackle it from all of
these different perspectives so when I think about the mind today is a little
bit different than how I thought about it as a seven-year-old when I think
about the mind today I often actually think about Leonardo da Vinci and the
reason that I think about him is that he may
some of the most beautiful artistic representations of the organ of the mind
which is the human brain and when I think about Leonardo da Vinci I often
think about his book called thoughts on art and life and the reason that I think
about that book is that it it provides a really beautiful context for why the
mind does what it does often it’s to create or to engage in relationships
with other individuals and that brings me to thinking about love and friendship
which motivates many of the things that we do in our lives whether it’s science
or art and when I begin to think about love and friendship I think about
Facebook and Twitter and snapchat and obviously Leonardo da Vinci is Feli
rolling over in his grave that I put these two on the same slide so I
apologize but I do want to ask the question is this more than just free
association what do these two things have in common if anything on the left
hand side we have social engineering systems like Facebook I actually didn’t
even know the little icon for snapchat I had to put that in somebody told me I
should really you know get with the program alright so what is what is a
similarity between a social engineering program and our the human brain on the
surface of it these two things are could not be more different however if you
think about them in terms of their underlying mathematical structure they
have some really beautiful similarities that we can actually use to try to
understand the human brain better and that underlying mathematical similarity
is that they both turn out to be networks so what is a network a network
is an object like this which is composed of units which we often called nodes and
then relationship between units which we often call edges now the first very
simple question is how is Facebook or other social engineering systems a
network and that’s pretty straightforward right so a node in the
network is a person and a edge between two people is a friendship and of course
we can put terms on here like the geeks and the jocks and the emo and the goth
which were very relevant when I was younger I’ve been told by undergraduates
who engage in research in my laboratory that these terms are very outdated
so you can put whichever term you prefer from your experience high school
experience probably on those little groupings okay
so the mapping of friendships to networks is very straightforward but
what about the brain how is the brain a network on the surface of it this is a
wet gooey organ that sits inside of your skull and there’s nothing about it that
is really particulate in that way that a social network is so what I’d like to do
is to suggest that we look inside and see whether we can understand
mathematically why the network structure is relevant for understanding the mind
the first question we have to ask is what are the nodes in this network what
are the pieces what are the units of the human mind and that’s obviously a
question that’s been asked not just today but in historical times as well
and was answered in a in the pseudoscience of phrenology so at that
point the idea was that there are specific portions of the brain that code
for specific functions or even specific morphologies of the skull that code for
specific functions as well and so in the age of phrenology we would have regions
that are important for ideology for veneration for firmness for benevolence
and for human nature itself so very very abstract concepts
now the actual terms that are located here and associated to specific regions
of the brain are not the way that we think about it as neuro scientists today
however the idea that a region can code for a single function is something that
we do hold as neuro scientists and why did we hold that what experimental
evidence has been brought to bear to suggest that we often use what’s called
a magnetic resonance imaging machine and these are commonly found in hospitals
and research grade ones are often found either at research hospitals or at
universities and what these machines do is that they allow us to map out the
regions that have become active in the brain when we perform a particular task
so if you read the New York Times or Scientific American or Discover Magazine
you probably often see pictures like what you see on the bottom right which
is a brain with some color on it usually yellows and reds that indicate that’s
the region that becomes active when you do
X in this case it’s playing the piano your motor cortex becomes active now
what’s nice about this is that we can go through and have human volunteers
perform many different tasks inside of the scanner and then we can identify oh
that’s the region that becomes active when you speak that’s the region that
becomes active when you do this that’s the region that becomes active when you
do that and that helps us to map out what these nodes are what these regions
are that code for specific functions but that’s not the end of the story where
are the edges inside of this system now understanding the edges or the
relationships or connections between different parts of the brain is an
endeavor that has also been important over several decades and probably the
most canonical circuit that we know of is the circuit involved in language
processing and so when you think about language you have to realize that
there’s actually no single piece of your brain that becomes active in language
processing or in language production it’s actually a whole set of regions so
information comes in through your visual cortex if you’re reading or through your
auditory cortex if you’re listening as you are doing right now and then it gets
propagated through the rest of the brain to Broca’s area and Wernicke’s area and
then eventually out into motor cortex which allows you to speak back to me or
to raise your hand or to do the other motoric movements that you that you do
in response to speech so in other words there’s a whole constellation of regions
and their relationships the passage of information between them that allows you
to perform language now that’s really nice if we’re talking about language
however there are many functions of a human brain that we do not understand at
that level we don’t understand what the circuits actually are and so often we’re
faced with a question of how could we perhaps find all circuits that are
available to the brain to use and answering that question has actually
only become possible in the last roughly 10 years and that is through this
specific type of imaging which is called diffusion imaging this is again Donna on
a magnetic resonance imaging machine but it maps out the diffusion of water
molecules inside of your brain now water molecules if you didn’t know are
constantly bouncing around in your brain by Brownian
but they are actually constrained along these big bundles of neuronal axons so
for anybody who sort of forgets or has not learned what a neuronal axon is
neurons are cell bodies in the brain and the neuron cell body looks a little bit
like a hand has a center and then it has dendrites coming out of it then it has a
long tail that long tail is called an axon and that allows that cell to
communicate to the other cell that’s way down here okay so what you have in your
brain is huge bundles of these axons that allow for information to be passed
form from one piece of the brain to another and that’s those streamlines are
those threads that you see in the brain right there now what’s fascinating about
this sort of picture is that it can be acquired non-invasively in any human so
meaning it doesn’t hurt you there’s no side effects it’s just an image like an
x-ray for example and what’s also fascinating about this is that these
patterns this the connections inside of your brain are different for every
single person and what scientists are finding over the last couple years is
that that pattern is almost like a fingerprint and it relates to many of
the things or helps us understand many of the things that you find easy to do
in your daily life and also many of the things that you find more difficult and
we’ll get to that in a second all right so those are the edges inside of the
network so here is the network now right so there are nodes which are regions of
the brain that code for a specific function and then there then there are
these wires that connect up different parts of the brain alright so now we
have a network now we understand why it’s like Facebook right but now the
question is how do we study it and how do we use that information to try to
understand how the brain truly works what we often use are many tools that
have been developed in the field of network science which is a very
interdisciplinary and rather emerging area of science that allows us to think
about complex systems by distilling them down into their units and a pattern of
relationships between their units the field pulls on many other traditionally
distinct disciplines like computer science mathematics physics engineering
statistics and even visualization actually so what we do is that we use
these kind of tools to be understand the architecture of a network is now if
I show you these two networks let’s say one of them is my brain and one of them
is my friend’s brain and I asked you are these two networks the same or different
actually I could ask my preschooler that and he would say of course they’re
different but then if I ask well how are they different can you tell me how
they’re different that becomes a lot harder this is they’re both pretty
complicated patterns so what network science does is that allows us to use
mathematics to explicitly quantify what’s different about the pattern on
the left from the pattern on the right and why that becomes important is that
then we can compare a healthy brain with one that has a disease or we can compare
an eight-year-old brain with a 22 year old brain and these comparisons become
important for us to understand cognitive function alright so in truth Network
science is basically a bunch of math that allows us to characterize the
architecture of these systems now I want you to give you an intuition for why
this matters and how it works I want to understand the brain okay so here is the
most fantastic information processing system you can imagine not actually it’s
a very simple toy graph but what I want you to imagine is that we want to use
this to transfer information now if I wanted to start and have information law
on this left side over here and I wanted to send information all the way over to
the other side I have a couple options of how I would want to do that perhaps
and the smartest way to do it would be to follow these red lines so I take one
hop two hops three hops right so I can get information from one side of the
network to the other side in three hops or I could take a more circuitous route
so for perhaps I could go down around here and then around here and
around here a couple times maybe you know and then eventually over here
that’s what your your mind feels like on a red-eye you know it’s like well taking
you a few extra turns around the corner there all right so what we can do is
that we can actually quantify that idea by calculating what’s called the
shortest path from one piece in the network to another so the shortest path
here is through one two three hops and obviously there are many longer paths
but we care about the shorter paths and the question is how would you construct
a brain network if you had the option of constructing one so that it had
relatively short paths to get from any piece of the brain to any other piece of
the brain your brain is an amazing information processing machine that has
to be able to transmit information very freely across across wide ranges of
space all right so we could build a network like this
which is where we have a lot of local connectivity however this is not
particularly good for information transmission right because if I had to
send information from the left side to the right side I’d have to go across
half of the circle okay not so efficient on the right hand side this is a
relatively efficient information transmission system because it can send
information from one side to the other very quickly however it loses the
capacity to have local information processing so there’s no help by
neighboring regions to process information this middle piece right here
has the best of both worlds which is that you have local processing here as
well as these short these long-distance connections that allow you to transmit
information to the other side very very quickly so when we look at humans we
actually find that many that all of us have this sort of architecture and in
fact it’s not just humans then we also see that across many different species
so we’ve studied humans we’ve studied macaque monkeys we have studied mice
Drosophila which is a fly and the elegans which is a nematode or worm and
what we found is that very very consistently there tend to be mostly
local paths but then these short these long-distance connections which allow
short cuts of information transmission across the entire brain and that’s very
consistent across all the species but then if we just or that we’ve studied
this that we’ve studied I should say but then
if we just look inside of the human and we look at the variation in this feature
across humans we can actually see that that variation is related to the way
that brain works so this is a really beautiful study not from my lab but from
another lab that shows that this prevalence of short paths in the network
is actually positively related to full scale IQ so individuals who have higher
IQ tend to have relatively short paths through their brain networks and
individuals with lower IQ tend to have longer paths within their brain networks
and that’s really interesting because it suggests that this network
representation of a human brain is not only an interesting correspondence to
social networks but it’s relevant for understanding how the system works as an
information processing system okay now why do you think short paths would make
for better brains this so this is a simulation where we’re just trying to
illustrate the fact that if you want to transmit information across this entire
system you have to have on average relatively short paths throughout the
entire thing but I also hope that this movie makes you start thinking about not
just where the paths are but how information gets transmitted and so I
want to make it I want to make an analogy to traffic and to roadway
systems because I think it’s helpful to help us understand how the brain works
so on the left-hand side these structural wires that I told you about
earlier I think of them as sort of the highways of the brain but knowing where
the highways are is not as important at least to me living around Philadelphia
as knowing what the traffic is like on each of the highways right I really want
to know what the traffic is so that I can choose the correct path for myself
to get to Penn in time for my class all right so we want to make the just we
want to distinguish between the structural highways and the traffic that
exists on top of them now what I would really love to do and say to you tonight
is that we have a perfect measurement of traffic on human brains that’s
non-invasive the answer is we actually don’t have that however what we can do
is make an indirect inference of where the traffic is in
brain and the way that we do that is that we say well let’s measure the
activity of every single brain region in this second and let’s measure it again
at this second and let’s measure it again at this second and every second
we’ll take a picture of the pattern of activity across the entire brain and
then what we’ll do is that will say if two regions of the brain are changing in
activity with one another then there possibly sharing information with each
other so in other words if two regions are increasing in activity together and
then decreasing in activity together and then increasing again together and
decreasing again together so they’re following each other in their activity
we say that those two regions are functionally connected they’re probably
responding to the same stimulus or possibly sharing information among them
so that’s our measurement of traffic for a human brain and it’s a non-invasive
measurement so then once we do that we can construct a second type of Network
for the human which is the traffic network we call it a functional Network
and an interesting question is how should I study that network to better
understand how cognitive processes function in a healthy human or in a non
healthy human and I actually want to go back to the illustration from social
networks because I think it actually helps us very much to understand what to
do here so here’s an example of a social network that I want to draw on so this
is a picture of Caltech Facebook friends and so each dot here is a different
student at Caltech the color of the dot indicates the house that the student
lived in and so what you can generally see is a strong clustered structure or
we often call it community structure or modular structure and you often find
that over here most of the people who live in the yellow house tend to be
friends with other people in the yellow house that makes sense people in the
purple house here tend to also be friends with other people in the purple
house there are these really interesting people in the center that don’t tend to
be friends with their own house but tend to be friends with at least one other
person in almost every house so those are really fascinating because they can
be important brokers of information between the different houses and that
actually becomes very important when we think about the brain as well because
there these areas in the brain as well that
sit between different clusters the other important feature I wanted to point out
over here is that there’s a little red guy right inside of the purple house so
he’s not friends with anyone in the red house but he’s friends with everyone in
the purple house and we’re pretty sure he should move so here is the picture of
what that sort of structure looks like in a human brain so here each of the the
circles is a region of the brain and the color code from the left hand side is
the same as the color code on the right so you can map across so for example
these regions over here are very strongly connected with one another
they’re all located in the back of your brain which is the occipital cortex that
helps you with visual processing or processing information that comes in
through your eyes there are also regions of the brain over here in purple that
are important for audition so listening and also for a language and there are
regions of the brain that are important for many other functions as well what we
basically find is that there are these separate groups that seem to performing
perform specific kinds of functions and that’s shown by their pattern of
activity they’re dynamic pattern of activity and I like to liken this to an
orchestra where we have many different kinds of functions in an orchestra right
we have the brass we have the strings we have the woodwinds we have the
percussion and they tend frequently to be communicating with one another in the
sense that they’re playing similar melodies specifically the strings but
that’s also true for some of the other systems as well so we have that
separation and yet together it creates this resonant beautiful architecture
which is what we see in the human mind as well okay so this observation that
there are a really beautiful clustered structure inside of the brain and that
those clusters seem to code for specific kinds of functions is something that’s
motivated many new questions across a wide variety of areas so I wanted to
specifically highlight these because it helps to showcase the interdisciplinary
nature of the investigations that are going on motivated by this work so
number one in psychology we ask how do these networked modules change in
different brain States number two in neuroscience what
rolls do neurophysiological processes play in the in these modules number
three in medicine how our networks altered in psychiatric disease or in
neurological disorders in mathematics we try to understand what graph models are
most like the brain and why in statistics we’re trying to understand
how we can infer true signal from noise in the data in physics we’re trying to
understand what role that network structure plays and material properties
that’s actually extremely important for traumatic brain injury so if you have
this structural wiring in your brain right that’s that’s very heterogeneous
that actually is a conduit for force from a traumatic brain injury in a way
that differs across depending on where the individual was hit and so that’s
this work is playing a very important role in understanding the effect of
traumatic brain injury on cognition and then in art what role does the network
structure play in the creative process and why so I think that all of these
questions are very exciting and I could certainly finish off the talk with these
questions but I actually want to place the statement out there that there’s one
big elephant in the room from all of this work and that is that if I really
want to understand human cognition there is no single picture I could show
you that would explain human cognition and that’s because a picture ignores the
dimension of time human cognition is dynamic it’s changing constantly right
so if we want to understand how the human mind works we have to have an
approach that’s dynamic in nature how are these networks changing as you think
about going home tonight or as you go to work tomorrow how did they change over
development how do they change with age how do they change when you’re anxious
versus less anxious so this is really where the work needs to go so how can we
do that well one way is that we can take those same images of the brain and those
same patterns of activity that we acquire from magnetic resonance imaging
and we can say well let’s extract the functional Network so that traffic
network just from the first two min of data okay and then let’s look at the
next two minutes of data and construct a network of what paths were used there
and then the next two minutes of data and the next two minutes of data that
allows us to actually create an avi of the network reconfiguring over time
showing different powers of information processing and communication as a
function of time so where have we been using that we’ve been using that
technique to understand long term reconfigurations and human learning and
we’ve been focusing initially on motor skill learning and so that’s relevant
for a child learning to ride a tricycle is relevant for anyone who plays a
musical musical instrument that’s motor learning it’s also relevant for anyone
who plays a sport that’s also motor learning so how does this happen
how does motor learning happen in the brain and can we watch it happen in real
time and see how these networks are changing the way that we did that Alysha
lee in the lab is through a very highly stylized experiment which i’ve been told
is very similar to guitar hero for anybody who has played guitar hero
before so the idea is basically that human participants volunteers go into
the scanner and they see this pseudo musical staff here you can see there’s
no clef therefore it’s doing a musical and UK also are holding a response
button box in their hand that has four buttons
so this says press the purple button then the blue one then the green one
then the red one then the blue one then the purple one etc okay and then they
practice these passages over and over and over and over again over the course
of three days and to ensure that they’re learning we calculate the movement time
which is the time from the first button press to the last button press and we
see how that changes as a function of time what you can see is that over the
three days of practice there’s a significant decrease in the time it
takes them to play one of these little passages and that’s similar to what any
of you would attest to if you play a musical instrument
or if you play a sport or if you learn to ride a tricycle when you were a child
okay so what we did in this experiment was to ask what is happening in the
brain as people are learning so here we know that behavior is changing we know
that output is changing says that what is happening inside that allows for
to change so what we did is exactly what I described before which is that we took
a two-minute window and extracted that pattern of functional relationships
between brain regions and then we did it again the next two minutes and again and
again we had 75 networks in a row for each person that mapped out that
reconfiguration of the brain over time and what we found is this beautiful
architecture which is that those strong modules that I showed you before are
still present but they tend to reconfigure at their boundaries so while
the modules themselves are relatively strong what happens at their
intersection changes a lot so you’ll have regions of the brain that initially
are communicating with one module and then change over time so look at this
peach one is becoming more and more strongly connected to the yellow module
so it’s switching over to its communication pattern to the yellow one
I like to think a little bit about dance partners and swapping dance partners in
this part of the talk because that’s that’s really what these regions are
doing is that initially they’re partnering with one other region or set
of regions and then at some point in the dance they switch over and they partner
with a different set and that allows for a change in the pattern of communication
that is then supportive of learning new information so what we found over the
last couple years is that brains that are very flexible like this that’s that
change these dance partners frequently throughout time are correlated with
individual differences in in how well that individual learns so the more
flexible the brain the better visual motor learning is the more cognitively
flexible that person is the higher working memory performance the person
has and better statistics in planning and reasoning so I think it’s also you
know we can we can think a little bit about the orchestral example here too
because every now and then you have the flute not playing with other woodwinds
but playing with the strings instead right or every now and then you have the
trumpets playing with the bass or the harp and that’s that change in
communication can really change the timbre of the music in the same way that
or in an analogous way to the way the brain is changing its pattern of
communication all right so that motivated me to ask the question
well if flexibility is related to these wonderful things like working memory and
learning and cognitive flexibility is there a way for us to change maybe our
lifestyle or something else to enable better more flexible brains and so what
we did is that we studied this wonderful data set that was actually acquired by a
fantastic professor at Stanford University his name is Russell pulled
rack and what he did is that he actually scanned himself twice a week for an
entire year on Tuesdays he was fasting I think and on Thursdays he had eaten
breakfast and then he also did an entire cognitive battery which is a set of
tests that test his cognitive function that day and filled out a bunch of
questionnaires so not only did he give us a year of his life doing this but
then he deposited all of that data online for any scientist anywhere in the
world to freely use and study so big kudos to restful track I think that’s
really an amazing thing for a scientist to do and what we found studying his
data is that on the days when he was rested and fed he tended to have a more
positive attitude was also much more attentive and scored more highly on the
attention tests and he also had higher brain flexibility on those days so that
this is a correlative finding but it suggests a causal hypothesis we could
test which is that if we alter the amount of rest or food that someone has
in the morning that would change their attitude and that would also change
their brain flexibility now if I tell this story of this study to a elementary
school teacher they say oh well of course I know that when my kids are not
rested and fed they don’t do very well but it’s really nice to see some
empirical backing from neuroscience to support that now I also want to suggest
that there are some individuals who do not have the capacity to change their
lifestyle in this particular way and so we still want to know whether there was
an intervention that we could come up with that would be able to enhance
flexibility for them and so what we’ve done over the last couple of years is
collaborate with andreas Meyer Lindenberg at the Central Institute of
Mental Health and Mannheim Germany and he has this wonderful dataset where
we’re healthy individuals we’re either taking placebo or we’re taking
dextromethorphan which is a drug it as an NMDA receptor antagonist and we find
that there’s actually significantly greater flexibility when the individuals
are on txm so that suggests that there is a pharmacological intervention that
we could offer to enhance flexibility and I’m really interested in asking
whether we could use that prior to rehabilitation particularly after stroke
and see if we can increase our speed recovery particularly of motor skills
after stroke so that’s the direction of that work at the moment okay now I think
all of that is amazing and exciting and I love this area of science but I wanted
to step back and ask some really big questions here in the last couple
minutes and that is what is learning and to what degree do our laboratory
experiments help us to understand the type of learning that occurs for example
in a classroom or maybe every day at work for you
so this type of experiment that I talked to you about is very relevant for
understanding motor skill learning specifically but is it relevant for
understanding learning broadly and the kind of learning that each of you are
engaging with by coming to a community lecture on science right all right well
to understand what is learning I’m gonna ask to ask the broader question which is
what is knowledge what we’re seeking when we are trying to learn so we’re
seeking knowledge so what is knowledge and if we better understand what
knowledge is perhaps we can change the kind of experiments that we do to tackle
that question more precisely alright so I wanted to quote John Dewey here in his
1916 democracy and education where he says knowledge is a perception of those
connections of an object which determine its applicability in a given situation
thus we get at a new event indirectly instead of immediately by invention
ingenuity and resourcefulness and ideally perfect knowledge would
represent such a network of interconnections that any past
experience would offer a point of advantage from which to get
at the problem in a new experience all right beautiful right so knowledge is a
network it’s a network of interconnections between concepts
between ideas between problems now if that’s the case then how do we gain this
knowledge network if we’re interested in learning how do we how do we grab that I
think we do it in two ways one is bi-curious ‘ti so again from view ii
curiosity is not an accidental isolated possession it’s a necessary consequence
of the fact that an experience is a moving changing thing involving all
kinds of connections with other things curiosity is but the tendency to make
those conditions perceptible so certainly curiosity drives us to
understand the connections why why is this related to this or what is related
to this over here how can i how can i build how can i grow what it is that i
know but we also learn knowledge networks by example and that’s
particularly important for classroom situations so to give you an
illustration of how we learn knowledge networks by example I wanted to tell you
a little story which is recounted in Robert McFarland’s book called landmarks
which is a really wonderful book I highly recommend it it’s a story about
Roger Deakin who was an English writer and documentary maker on UK waterways
specifically so the story is about a mentee who invites Roger who was his
mentor to Cambridge University where the mentee is currently studying and the
mentees idea is well if I invite my very famous mentor to come and give a talk at
Cambridge everyone else at Cambridge will think more highly of me because my
mentor gave a really great talk ok so that was his goal and now here is his
memoir of that now-infamous day he says I stared dedicatedly at my shoes
embarrassed that my friend was failing to perform in front of my academic peers
it was only later that I realized it wasn’t a failure to perform but a
failure to conform Cambridge seminars expect rigour and logic in their
presentations braced subtlety of exposition and explanation tested proofs
of cause and but Roger but water doesn’t do rigor in
that sense and neither did Roger though his writing was often magnificently
precise in his poetry for Roger watered flowed fast and wildly through culture
it was protein it was slip shaped and so that was how he followed it slipshod and
shipshape at once moving from a word here to an idea there too fast for his
notes or his audience to keep up with joining his watery subjects by means of
an invisible network of tunnels and drains feel really bad for the mentee
right on the other hands I’m fascinated by Roger
so is this person he was just an amazing well he’s an amazing scholar and he
gives lectures in this architecture that’s very similar to his subject so he
gives watery lectures lectures that sort of meander around sometimes slow too
fast sometimes too slow you know it’s fascinating
okay so that makes me wonder our lecturers in general perhaps are they
could they be thought of as walks through networks and maybe it’s not just
lectures maybe it’s books the book that you’re currently reading is that a walk
through a network of ideas of concepts so perhaps you’re reading a linear
algebra textbook uh-huh and the architecture of that
linear algebra textbook is highly structured right and so that walk
through those ideas is very highly structured I say that because we’re
currently studying linear algebra textbooks to understand their network
architecture so there’s a backstory to that or perhaps you’re reading a book on
UK waterways and you may see more of this tortuous information transmission
or perhaps you’re reading a history book now in history there’s often more of
this tendril like nature right because you’re following a story through time
and it’s relatively linear often or at least that’s how we can often understand
why things happened in history the way that they happened all right so that
makes me ask the question of is there perhaps an optimal way of walking
through a network in lectures or in books or in papers so let’s suppose that
you are currently reading a book and the aura book chapter specifically and the
author has these 15 ideas that they hope to present to you if it’s a if it’s a
fictional book you know perhaps this is all information about the protagonist
this is information about a counterplot this is information about the history of
the protagonist in their prior life any of the author has to go through all of
these pieces of information but they have to walk through it in some sort of
way that makes you on the other side understand what that broader
architecture was and what makes this really difficult is that language is
formally linear you only read one word before you read the next word right and
they can only write one word before they write the next word so everything that
you take in from a lecture or from a book or from a paper is a linear
transmission of information but it’s a transmission of what’s probably a much
higher dimensional much more complicated object so this is the picture of what’s
happening and what I’m really fascinated by this I think is what learning really
is so here’s the brain of the speaker or a writer it has this beautiful network
structure of ideas concepts problems and how they relate to one another the brain
of the speaker or writer Maps that into the one dimension
of time either in writing or in audio so that the brain of the listener or reader
on the other side can optimally reconstruct what that beautiful higher
dimensional object was how do we do that we do that regularly we all right we all
communicate with one another we all speak to each other we all take very
complicated ideas and map them into one D how do we do it
it’s amazing and also how does this person on the other side reconstruct
this complicated object from a linear stream of information this actually is a
beautiful function of the human mind that’s important for many many behaviors
beyond listening to a lecture or reading a book so in fact the problem of
inferring patterns of pairwise dependencies like this between from
incoming streams of data allows us to learn language to begin with this is how
it’s thought that infants actually learn language helps us to segment visual
images to parse tonal groupings in music to parse spatial scenes to infer social
networks around us and to perceive distinct concepts so it’s pervasive in
all of our lives now what we’ve been doing over the last couple years is
trying to understand is there a particular kind of network architecture
that’s easy to map into one dimension and that’s also easy for the other
person to reconstruct so is it perhaps this circular one is it best if I were
to give a lecture that’s circular that means two different things to different
people is it best if I give a lecture that has more of a torus shape or that’s
very densely filled in or that has mostly triangle so I close every gap in
knowledge I for my clothes immediately I relate every concept to every other
concept as much as possible what is the optimal way of presenting information
and sharing that architecture with one another what we’ve found over the last
couple years is that in fact this architecture this very highly modular
one that I showed you earlier tends to be very easy
learned by many different people across many different ages and so we actually
think that there might be something we haven’t I should say we haven’t
comprehensively evaluated every network architecture that’s out there in the
world but our initial evidence suggests that this modular architecture is very
very easy for humans to learn and perhaps that would help us to present
information in classrooms to write books to write papers and to communicate with
one another better but I think probably one of the
most interesting features or corollaries of this observation is that what I told
you earlier is that the human mind the human brain looks very modular itself so
it is composed of these dense clusters of connectivity between brain regions
and then what we’re finding is that actually the network architectures that
we are very that we find very easy to learn have this modular structure as
well and that reminds me of this beautiful passage from Aristotle’s
metaphysics where he says mind thinks itself because it shares the nature of
the object of thought for it becomes an object of thought in coming into contact
with and thinking about its objects so that mind an object of thought are the
same that raises all sorts of interesting questions is the
architecture of an optimally learn about network over here a topological
reflection of the optimally developed a neural network over here or is this
correspondence in structure simply just happenstance I think it’s probably not
happenstance I think it’s probably more than that perhaps it tells us something
important about the nature of modeling in the human brain now that’s a rather
large question and I’m actually going to leave it there I would like to
acknowledge the people who are very important in this work particularly
these individuals here who performed a lot of the experiments for the data that
I showed you today the team itself is quite large we have I’m very grateful
for funding from many different organizations that have supported the
work over the last few years I want to thank you for listening again it was
very nice to meet you I would love to take yeah yes
does this help alright so in the scale that you used that led to flexibility in
terms of the way the brain thinks was the scale repeated in the same way every
time or was the scale itself flexible or varied that’s a really great question
there were six different sequences so they practice six different ones but
they were fixed they were not probabilistic or flexible in any way one
person did six different little passages yeah but those passages were were fixed
there was no no change so we’re certainly not teaching improvisation I wonder if you could review that the
model that was the most connective that you talked about a model that was easier
to understand either lecture or in a written form whatever it would be more
easily translatable could you repeat or kind of review that this this modular
structure that I show on the right-hand side so what we’re finding is that we
can develop a particular laboratory experiment where we show information
that has that architecture so it’s a stream of information that’s actually a
walk on that graph so if you imagine you’re a person who can walk on pages
you would stand at one of those black nodes and that and then you would walk
anywhere that there’s a blue line you would be allowed to walk okay
but you couldn’t walk in the white space and so if you actually take a random
walk through that graph that provides you with a passage of this concept and
this concept and this concept in this concept right and so what w