[Applause]
sym that anywhere you look these days
you can’t help but hear that artificial
intelligence is developing so quickly
that it will make us obsolete and we
must augment ourselves in order to
compete with robots on the job market
just the other day I saw a magazine
cover that showed the logical
progression from a Neanderthal modern
human to a cyber enhanced being with
some mystery hardware strapped to her
hairless head that all brings up a
question when can upload my brain into a
computer my lab works on interfaces with
the nervous system and whenever I get
asked this question which is at every
dinner party I can’t help but wonder
whether this is an ill-posed question
and what we should really be asking is
when can my brain collaborate with a
computer machine intelligence in human
intelligence are fundamentally different
a task of folding a wrinkled sock takes
seconds for an untrained human who is
also eating cereal talking on the phone
listening to the radio while a highly
trained robot will wrestle with a sock
for a few minutes while doing absolutely
nothing else on the other hand a trained
human will spend a few minutes on an
integral of a sine function multiplied
by a polynomial while a single simple
calculator can crank this out in a
fraction of a second the human brain and
the machine brain are fundamentally
different human brain is a generalist a
machine brain is a specialist in this
difference in operation lies
indifference in hardware the building
block of machine brain is a transistor
it can switch billions of times per
second but is only connected to three
neighbors building block of the nervous
system is a neuron it can do only about
a thousand operations per second but it
communicates with 6000 neighbors at once
and trying to upload your brain visits
90 billion neurons
into a machine requires simultaneous
monitoring of a multitude of electrical
chemical mechanical signals across
trillions of synapses at the same time
some say that this translates into an
information density of about a hundred
peda flops which is equivalent to about
a few million of modern laptops we of
course have a few million laptops but
the trick is that unlike your brain
which fits into one head and consumes
about 12 watts a million laptops occupy
a warehouse and consume some megawatts
but that’s not the biggest problem the
biggest problem is that we do not
understand how this multitude of
parallel signals and computations in
your human brain translates into your
human intelligence
worse yet we don’t have the tools to
find out but don’t get me wrong
machines are incredibly powerful many of
you may have used Google Maps and within
fraction of a second machine spits out
three alternate routes from point A to
point B and it can even help you find a
Thai takeout on the way and this task
will take minutes two days for your
human brain yet you consistently rely on
human reviews to choose your Thai
takeout when it comes to menial tasks in
such as procedure that I prescribed for
example welding joints fastening bolts
drilling holes as a manufacturing plant
machine consistently outperforms a human
in both quality speed and precision but
the problems arise when the machine
breaks or simply encounters a joint that
is misaligned and it needs to weld it it
turns out that the most sophisticated
manufacturing plants machines are being
tuned calibrated fixed and trained by
people and it’s the ability of human
brain
to quickly analyze complex environment
pick out patterns make decisions and
perform a wide variety of seemingly
disparate tasks at once is the strength
of our brain so on one hand we have this
compact generalist human intelligence on
the other hand we have this precise
specialist machine intelligence and
wouldn’t it be great if these two types
of intelligence could collaborate in a
way that is more natural than typing
words into a command line in fact our
even in our imagination the interface
between human and machine intelligence
is so poor that science fiction
characters use verbal input to talk to
machines in Iron Man movies for example
Tony Stark talks to super artificial
intelligence Jarvis in order to get help
these calculations shouldn’t jarvis be a
simple module in Tony’s brain well it
turns out we currently don’t have any
tools to connect to the human brain but
how come some of you may have seen
videos of paralyzed persons controlling
robotic arms with their brain signals
but that was in the lab on a highly
prescribed task supervised by a team of
PhDs to give you a flavor of what
happens in real world I’d like to tell
you a story I recently had a chance to
sit in the panel with a science
journalist Miles O’Brien who also
happens to be a single arm amputee and
miles is a one of the most well known
science nerds and you would expect him
to have access to most sophisticated
prosthetic limbs yet he just like many
other single arm amputees does not wear
a prosthetic it turns out that the most
sophisticated artificial arms
pale in comparison to your natural
hardware and what is missing is the
interface to your nervous system simply
put those hands do not feel nor do they
move under them and this problem is
unsolved despite the fact that in order
to connect a prosthetic arm
you only need to communicate with a few
million only fibers in your in the
peripheral nerves in your arm now
imagine how this problem scales when we
talk about trillions of synapses in your
brain that some hope to upload into a
computer and the problem of the neural
interface is not simply a problem of
resolution the difference is between
human computing hardware an artificial
computing hardware come from biology
chemistry and material science your
brain is soft squishy and
three-dimensional electronics that power
modern computing and artificial
intelligence are hard rigid sharp and
flat and despite 30 years and billions
of research dollars machine inspired
brain interfaces continue to fail and
what we need is a brain inspired machine
interface we need an interface that
matches biology chemistry and mechanics
of the nervous system that can match the
resolution of individual synapses and
that can communicate with neurons across
all the natural languages
neurotransmitters electric fields
mechanical forces changes in pH changes
in temperature and many others that we
have not yet discovered at the same time
this interface needs to be able to
communicate back to the artificial
hardware and we do not have such an
interface but we are hard at work trying
to get there in our lab we have
developed nano transducers
that our thousand times smaller than
individual coffee grinds they can be
mixed into IV solution and injected into
your brain just like a drug there they
can receive signals from outside
electronics and convert them into one of
neuronal languages for example heat our
colleagues are busy working on devices
that can record from neurons and
transmit their signals to outside
electronics yet even if it’s the most
sophisticated hardware we can only
communicate with a few thousand neurons
via orders of magnitude away from
trillions of synapses in the meantime
artificial intelligence is marching
along and it’s developing much more
quickly than the hardware required to
interface with your entire brain and
it’s likely going to be a missed
opportunity to not take advantage of
artificial intelligence as a specialist
plug in into your generalist brain and
this type of future is likely much much
closer than trying to develop a
comprehensive understanding of every
synapse in your brain and even if we did
develop such an interface should we
really spend energy on uploading this
compact generalist into an army of
specialist machines a sock folder a
joint welder or Thai takeout reviewer
wouldn’t it be more efficient to for
example add a specialist piano player or
GPS unit to your generalist brain so one
might wonder whether there is even space
in our brain for all the information
that artificial intelligence is capable
of delivering and receiving but did I
mention that brain is not stationary
those synapses are evolving brain is
plastic which means the wiring changes a
every time you learn a new skill and
this process will likely be accelerated
by interfacing to the brain directly and
collaboration with artificial
intelligence may not just deliver
specialist knowledge or skills this
partial interface may make our own human
brain even more powerful and diverse
generalist
so how can we get there the
lightning-fast development of artificial
intelligence is powered by ubiquitous
computing it’s accessible to many and
it’s very democratic development of the
neural interface hardware on the other
hand requires significant infrastructure
training and investment and it also
requires a paradigm shift from machine
inspired electronics to biology driven
design of new materials and
architectures and we cannot simply
outsource this task to a handful of
academic labs it will simply take too
long we already took 30 years and if we
take another 30 we will be missing an
opportunity to gain a collaborator and a
partner in artificial intelligence so
let’s join forces with academia industry
and government to accelerate the
development and deployment of a neural
interface such that our generalist human
intelligence can collaborate with
specialists artificial intelligence to
do what it does best create analyze and
decide
[Applause]
sym that anywhere you look these days
you can’t help but hear that artificial
intelligence is developing so quickly
that it will make us obsolete and we
must augment ourselves in order to
compete with robots on the job market
just the other day I saw a magazine
cover that showed the logical
progression from a Neanderthal modern
human to a cyber enhanced being with
some mystery hardware strapped to her
hairless head that all brings up a
question when can upload my brain into a
computer my lab works on interfaces with
the nervous system and whenever I get
asked this question which is at every
dinner party I can’t help but wonder
whether this is an ill-posed question
and what we should really be asking is
when can my brain collaborate with a
computer machine intelligence in human
intelligence are fundamentally different
a task of folding a wrinkled sock takes
seconds for an untrained human who is
also eating cereal talking on the phone
listening to the radio while a highly
trained robot will wrestle with a sock
for a few minutes while doing absolutely
nothing else on the other hand a trained
human will spend a few minutes on an
integral of a sine function multiplied
by a polynomial while a single simple
calculator can crank this out in a
fraction of a second the human brain and
the machine brain are fundamentally
different human brain is a generalist a
machine brain is a specialist in this
difference in operation lies
indifference in hardware the building
block of machine brain is a transistor
it can switch billions of times per
second but is only connected to three
neighbors building block of the nervous
system is a neuron it can do only about
a thousand operations per second but it
communicates with 6000 neighbors at once
and trying to upload your brain visits
90 billion neurons
into a machine requires simultaneous
monitoring of a multitude of electrical
chemical mechanical signals across
trillions of synapses at the same time
some say that this translates into an
information density of about a hundred
peda flops which is equivalent to about
a few million of modern laptops we of
course have a few million laptops but
the trick is that unlike your brain
which fits into one head and consumes
about 12 watts a million laptops occupy
a warehouse and consume some megawatts
but that’s not the biggest problem the
biggest problem is that we do not
understand how this multitude of
parallel signals and computations in
your human brain translates into your
human intelligence
worse yet we don’t have the tools to
find out but don’t get me wrong
machines are incredibly powerful many of
you may have used Google Maps and within
fraction of a second machine spits out
three alternate routes from point A to
point B and it can even help you find a
Thai takeout on the way and this task
will take minutes two days for your
human brain yet you consistently rely on
human reviews to choose your Thai
takeout when it comes to menial tasks in
such as procedure that I prescribed for
example welding joints fastening bolts
drilling holes as a manufacturing plant
machine consistently outperforms a human
in both quality speed and precision but
the problems arise when the machine
breaks or simply encounters a joint that
is misaligned and it needs to weld it it
turns out that the most sophisticated
manufacturing plants machines are being
tuned calibrated fixed and trained by
people and it’s the ability of human
brain
to quickly analyze complex environment
pick out patterns make decisions and
perform a wide variety of seemingly
disparate tasks at once is the strength
of our brain so on one hand we have this
compact generalist human intelligence on
the other hand we have this precise
specialist machine intelligence and
wouldn’t it be great if these two types
of intelligence could collaborate in a
way that is more natural than typing
words into a command line in fact our
even in our imagination the interface
between human and machine intelligence
is so poor that science fiction
characters use verbal input to talk to
machines in Iron Man movies for example
Tony Stark talks to super artificial
intelligence Jarvis in order to get help
these calculations shouldn’t jarvis be a
simple module in Tony’s brain well it
turns out we currently don’t have any
tools to connect to the human brain but
how come some of you may have seen
videos of paralyzed persons controlling
robotic arms with their brain signals
but that was in the lab on a highly
prescribed task supervised by a team of
PhDs to give you a flavor of what
happens in real world I’d like to tell
you a story I recently had a chance to
sit in the panel with a science
journalist Miles O’Brien who also
happens to be a single arm amputee and
miles is a one of the most well known
science nerds and you would expect him
to have access to most sophisticated
prosthetic limbs yet he just like many
other single arm amputees does not wear
a prosthetic it turns out that the most
sophisticated artificial arms
pale in comparison to your natural
hardware and what is missing is the
interface to your nervous system simply
put those hands do not feel nor do they
move under them and this problem is
unsolved despite the fact that in order
to connect a prosthetic arm
you only need to communicate with a few
million only fibers in your in the
peripheral nerves in your arm now
imagine how this problem scales when we
talk about trillions of synapses in your
brain that some hope to upload into a
computer and the problem of the neural
interface is not simply a problem of
resolution the difference is between
human computing hardware an artificial
computing hardware come from biology
chemistry and material science your
brain is soft squishy and
three-dimensional electronics that power
modern computing and artificial
intelligence are hard rigid sharp and
flat and despite 30 years and billions
of research dollars machine inspired
brain interfaces continue to fail and
what we need is a brain inspired machine
interface we need an interface that
matches biology chemistry and mechanics
of the nervous system that can match the
resolution of individual synapses and
that can communicate with neurons across
all the natural languages
neurotransmitters electric fields
mechanical forces changes in pH changes
in temperature and many others that we
have not yet discovered at the same time
this interface needs to be able to
communicate back to the artificial
hardware and we do not have such an
interface but we are hard at work trying
to get there in our lab we have
developed nano transducers
that our thousand times smaller than
individual coffee grinds they can be
mixed into IV solution and injected into
your brain just like a drug there they
can receive signals from outside
electronics and convert them into one of
neuronal languages for example heat our
colleagues are busy working on devices
that can record from neurons and
transmit their signals to outside
electronics yet even if it’s the most
sophisticated hardware we can only
communicate with a few thousand neurons
via orders of magnitude away from
trillions of synapses in the meantime
artificial intelligence is marching
along and it’s developing much more
quickly than the hardware required to
interface with your entire brain and
it’s likely going to be a missed
opportunity to not take advantage of
artificial intelligence as a specialist
plug in into your generalist brain and
this type of future is likely much much
closer than trying to develop a
comprehensive understanding of every
synapse in your brain and even if we did
develop such an interface should we
really spend energy on uploading this
compact generalist into an army of
specialist machines a sock folder a
joint welder or Thai takeout reviewer
wouldn’t it be more efficient to for
example add a specialist piano player or
GPS unit to your generalist brain so one
might wonder whether there is even space
in our brain for all the information
that artificial intelligence is capable
of delivering and receiving but did I
mention that brain is not stationary
those synapses are evolving brain is
plastic which means the wiring changes a
every time you learn a new skill and
this process will likely be accelerated
by interfacing to the brain directly and
collaboration with artificial
intelligence may not just deliver
specialist knowledge or skills this
partial interface may make our own human
brain even more powerful and diverse
generalist
so how can we get there the
lightning-fast development of artificial
intelligence is powered by ubiquitous
computing it’s accessible to many and
it’s very democratic development of the
neural interface hardware on the other
hand requires significant infrastructure
training and investment and it also
requires a paradigm shift from machine
inspired electronics to biology driven
design of new materials and
architectures and we cannot simply
outsource this task to a handful of
academic labs it will simply take too
long we already took 30 years and if we
take another 30 we will be missing an
opportunity to gain a collaborator and a
partner in artificial intelligence so
let’s join forces with academia industry
and government to accelerate the
development and deployment of a neural
interface such that our generalist human
intelligence can collaborate with
specialists artificial intelligence to
do what it does best create analyze and
decide
[Applause]
Please follow and like us: