artificial intelligence is just like
politics everybody loves to talk about
it but at the same time very few people
actually understand how it exactly works
and what it is and if you’re like me you
used to think that artificial
intelligence was a science fiction
concept but then lately you hear quite a
lot of serious people researchers
talking about it and you don’t quite get
it well we do associate AI with movies
with Terminator with Star Wars and that
makes it a little bit fictional to us a
little bit unrealistic another problem
is that AI is a very broad concept and
it arise from our phones through
self-driving cars to something that can
change our lives dramatically and so all
of this is very confusing and we as
humans tend to be afraid of the things
we do not understand so my goal is to
help you understand AI a little bit
better to be a little bit less scared
and a little bit more excited about its
development so what is AI artificial
intelligence sometimes called AI is the
theory and development of computer
systems which are supposed to be doing
the same tasks as we humans can do and
that includes machine vision natural
language processing language translation
and so on but does that actually mean
that our human intelligence is nothing
more than just the sum of the abilities
to hear see and speak well I hope not
because that would be just very sad so
let’s think what else do we know about
intelligence being intelligent is
something that we wish other people
think about us
we know from school that dolphins are
intelligent animals right and nowadays
even if we go to an electro store we can
find a lot of intelligent home
appliances they’re confusing again and
so let’s take those as an example an
electric ion can react to one particular
state becoming too hot
then it switches itself off which is
supposedly making it intelligent but do
you think this is intelligence I don’t
think so because the Aaron has only one
program puta net it’s pre-programmed for
one scenario getting two holes turn
itself off and that’s it and it hasn’t
even learned that through an experience
so it cannot call it truly intelligent
machine and we as humans can never do
things right basing on the precise
instructions only we try to really give
up on that but it doesn’t always work
quite as we expected because
instructions are never enough to learn
how to do something if you remember how
all of us were learning how to walk
being a baby was it just your mother
telling you that you know now you stand
still you put your weight on the left
leg you meant your right leg you put it
forward boom you know how to walk not
really
would have been nice but not yet because
the trick is baby can never learn basing
on the instructions only baby needs to
practice all the previous steps and
master each of them firstly needs to
learn how to sit then how to stand then
how to walk but after it learned how to
walk and that’s the beauty maybe can do
absolutely anything with his legs baby
can run baby can jog jab walk backwards
anything you can imagine and baby
doesn’t need a million of very precise
instructions for each case for each task
because it understood the concept right
but at the same time everything which we
call AI these days is something like an
advanced iron it can only do one thing
these things are sometimes seems
complicated to us it can do these things
in the same level of humans and
sometimes even better than humans but
still every particular time it’s just
one narrow task and that’s why the
current state-of-the-art AI is called
narrow AI if you ask an AI which can
play chess to play poker with you for
example it wouldn’t be able to do so
because these two are completely
different tasks the trick is that all
this seems all these things seem very
complicated to us
as humans but at the same time they’d
have very clear rules on how to progress
with those and that’s why these things
are simple for AI but at the same time
the things we do without thinking the
natural things such as vision motion
perception something which is almost
impossible to teach AI to do because
these are the skills we cannot define in
zeros and ones and they only seem
natural to us because we as humans got
that kind of as a gift over the
thousands of years of human evolution
and that’s why it’s easy and that’s why
it’s natural so while modern AI can do
only one narrow task what we try to do
in our research group good AI is to
create an AI which would possess the
ability to learn because this we believe
is going to change the rules of the game
dramatically how do we make an AI learn
how does it learn thing
the answer is gradually just like humans
step by step just as baby from the very
beginning to more complicated tasks one
of the approaches to teaching AI is
actually very similar to making kid do
something to working with kids at school
if you have your baby and you want to
motivate it to do something to do some
tasks for example do the homework then
you motivate the baby to take the action
by giving a positive reward like a bar
of chocolate you did a homework that’s
really good to carry on and so after
that baby or an AI realizes that this is
good it’s on the right track and it
keeps repeating the action to receive
the positive reward as long as needed
before it actually learned something
this particular approach is called
reinforcement learning but the
approaches are numerous the key is to
keep the direction from the simple tasks
to the more complicated ones and to make
sure that AI doesn’t forget the things
it learns in the first place but keeps
reusing those skills over and over to
obtain even new ones even more skills
and after that instead of having an AI
which can play one game we will be able
to have an AI which learned how to play
any game and then it can learn how to
play video games and then it can design
its own video game and then it can
design anything you can think of and
then it can think of something to design
and then design it and so on and so on
this should be truly exciting and this
is what we call general purpose AI we
try to achieve general purpose AI as
fast as possible and this is because
future is coming and future is coming
much faster than we all think AI
research is now in its very beginning
the general purpose AI is not yet
created and we don’t know how long is it
going to take
but that’s why in the beginning of the
research the cost of sharing the ideas
is very low while at the same time the
benefit for the bright minds to
cooperate cooperate together and crack
this problem is very big another nice
thing is that computing power is so far
very well controlled and we are not yet
that dependent on the computers to wake
up one morning to realize that you know
they’ve taken the whole world over and I
know this was meant to be a joke but I
know that most of you get really scared
at the moment because AI seem scary to a
lot of people right and people believe
that AI is likely to take their jobs and
to replace them as humans well I’ve been
thinking about this as well a lot and I
believe that there were many cases in
history when something like that a new
technology was also taken in a similar
way I would like to introduce you one
piece of technology which was supposed
to replace thousands of people’s on
their jobs and everybody hated that and
nobody wants it bad introduced what do
you think that might be there are many
options right but this is one of them
remember there were actually people out
there on the cross rose navigating the
traffic out in the cold risking their
lives every day do you think it’s better
that now we can rely on the technology
to coordinate the traffic and those
people can stay in the warm offices to
coordinate that or even do something
completely different
I believe this is the beauty of the
technology and it is very important to
understand that AI is nothing more than
just a tool just something to help us do
different things better something to
augment our capabilities but not a whole
different scary thing how many of you
have been possibly dreaming of studying
history or arts becoming an artist but
then ended up being for example an
accountant because that’s something more
practical and that gets paid better well
I believe that there are many cases like
that and once the invoices we’ll be able
to sort out themselves we as humanity
will have a chance to appreciate the art
more and do the things we really feel
like doing but this is the bright side
but at the same time there is more to
the motivation of having AI as soon as
possible and this is because our brains
are reaching their limits and that’s
unfortunately very serious with
terabytes of data being generated around
us every day with our computers with our
cell phones we just don’t catch up with
that as humanity we cannot progress on
many research problems and we cannot
progress with a space exploration for
example because space is abstract and
remote and you know that big and our hat
it’s just of this size and then our
brains are just limited to the size of
our hat while at the same time computers
can grow as big as we can imagine that
would allow them to be more fast more
efficient than us humans what is more
important is that computer transistors
are also much more reliable than our
brain neurons because I can complain
that I have a headache pretty much every
week you know and I cannot work when I
have headache and computers can work 24
/ 7 on their peak performance every day
non-stop which I believe is exciting and
it would be a shame not to use that
finally it is ok when computer breaks
but it’s not ok if anybody dies during
the research and that’s unfortunately
what’s holding us back in the progress a
lot that’s the cost of the new discovery
is in many cases measure
by humans lives when we talk about space
exploration again we cannot send a
cosmonaut further than for a distance of
a human’s life time because that
wouldn’t just be fair you know and
that’s why we cannot progress on those
problems and humans right now are the
only creatures in the whole universe
that can make decisions in the unknown
circumstances and if we manage to train
our general-purpose AI to make those
decisions if we manage to make it
explore space or do many other things
then I believe we can progress much
further in the scientific progress for
example in the medicine it would be
amazing to have access to the knowledge
of all the doctors in the universe at
once you know having an algorithm which
would be able to create an optimal
treatment court for every patient basing
on this knowledge and that would allow
us to save even more lives and that’s
what AI can do for us and then the
cherry on the cake is that if a I can
possibly solve any problem do you think
it can solve the problem of solving
problems there is actually a research
term for that we call that point
singularity and that looks very exciting
but it’s not coming anytime soon
I guess however we must start right now
to make that happen and as you might
imagine now creating something like that
is a very difficult task and in my
research group
we’re just 20 people so how can
something like that you know be created
by just many people well the good thing
is that AI is very hyped these days
that’s the reason you’re listening to me
right now and that’s the reason why many
people feel excited to be part of the
research another good thing is that
nobody needs a very particular set of
skills to be able to become an AI
researcher people with very different
backgrounds can come together to work on
AI and all of that
creates a perfect momentum to leverage
something called citizen science citizen
science is a concept of opening and
outsourcing the research problems to the
general crowd to people who might be
interested to participate and contribute
to the progress and for that we have
recently launched a competition called
general AI challenge this is a
competition where people from all over
the world could contribute to the
development the first round was devoted
to the problem of gradual learning of
learning step by step and people were
programming their own AI agent able to
solve some tasks today we were
announcing the winners of this first
round and it’s exciting to see that they
come from very different backgrounds we
actually have an astronaut there we have
a quantum physicist we have people of
all different kinds and it’s exciting to
see how all of them can contribute let’s
go now we’re preparing a second round
which would be devoted to the question
of how can we avoid AI development
becoming a race between different
researchers and how to deploy AI in
businesses safely so basically no matter
whether you have programming background
or Social Sciences background each of
you gets an opportunity to join us on
this exciting journey to give us a hand
in creating safe and beneficial
general-purpose
AI exciting rate what I wants you to
keep in my mind what I want you to keep
in your mind after my talk are just
three things the first one the zero one
is the general AI challenge which you
can participate in and then the first
one is that our brains are incredible
and the most precious ability we possess
is the ability to learn that comes from
experience and that’s how we learn
things an AI is learning things in the
very same way too step by step what is
even more important is that AI is just a
tool for us to be better for us to do
things better and it’s not a separate
species the second thing is that
general-purpose AI can be worth more
than anything because this is
ultimately elaborate to any technology
any science any process it can optimize
plenty of things but in order to create
that the collaboration between
researchers needs to start right now and
then finally right here right now is the
best moment in the whole history to
start working on the AI because the
ingredients to having a good a AI are
essentially just three it’s having
computing power it’s having data and
it’s having algorithms and so guess what
computing power has never been cheaper
than today algorithms for inspiration
are being open sourced each day and were
droning with data with terabytes of it
being generates at each second so we
have everything we need and if you have
ever been dreaming of being born in some
other time to witness some amazing
technology emerging or new discoveries
being made then with AI you actually
don’t have to dream because the Golden
Age of AI is right here it’s right now
and everybody can join us on this
journey I am planning to have a very
cool story to tell my kids about how
their mommy helped make the future
happen and each of you can have one too
so join us and thank you
[Applause]
[Music]
[Applause] you
politics everybody loves to talk about
it but at the same time very few people
actually understand how it exactly works
and what it is and if you’re like me you
used to think that artificial
intelligence was a science fiction
concept but then lately you hear quite a
lot of serious people researchers
talking about it and you don’t quite get
it well we do associate AI with movies
with Terminator with Star Wars and that
makes it a little bit fictional to us a
little bit unrealistic another problem
is that AI is a very broad concept and
it arise from our phones through
self-driving cars to something that can
change our lives dramatically and so all
of this is very confusing and we as
humans tend to be afraid of the things
we do not understand so my goal is to
help you understand AI a little bit
better to be a little bit less scared
and a little bit more excited about its
development so what is AI artificial
intelligence sometimes called AI is the
theory and development of computer
systems which are supposed to be doing
the same tasks as we humans can do and
that includes machine vision natural
language processing language translation
and so on but does that actually mean
that our human intelligence is nothing
more than just the sum of the abilities
to hear see and speak well I hope not
because that would be just very sad so
let’s think what else do we know about
intelligence being intelligent is
something that we wish other people
think about us
we know from school that dolphins are
intelligent animals right and nowadays
even if we go to an electro store we can
find a lot of intelligent home
appliances they’re confusing again and
so let’s take those as an example an
electric ion can react to one particular
state becoming too hot
then it switches itself off which is
supposedly making it intelligent but do
you think this is intelligence I don’t
think so because the Aaron has only one
program puta net it’s pre-programmed for
one scenario getting two holes turn
itself off and that’s it and it hasn’t
even learned that through an experience
so it cannot call it truly intelligent
machine and we as humans can never do
things right basing on the precise
instructions only we try to really give
up on that but it doesn’t always work
quite as we expected because
instructions are never enough to learn
how to do something if you remember how
all of us were learning how to walk
being a baby was it just your mother
telling you that you know now you stand
still you put your weight on the left
leg you meant your right leg you put it
forward boom you know how to walk not
really
would have been nice but not yet because
the trick is baby can never learn basing
on the instructions only baby needs to
practice all the previous steps and
master each of them firstly needs to
learn how to sit then how to stand then
how to walk but after it learned how to
walk and that’s the beauty maybe can do
absolutely anything with his legs baby
can run baby can jog jab walk backwards
anything you can imagine and baby
doesn’t need a million of very precise
instructions for each case for each task
because it understood the concept right
but at the same time everything which we
call AI these days is something like an
advanced iron it can only do one thing
these things are sometimes seems
complicated to us it can do these things
in the same level of humans and
sometimes even better than humans but
still every particular time it’s just
one narrow task and that’s why the
current state-of-the-art AI is called
narrow AI if you ask an AI which can
play chess to play poker with you for
example it wouldn’t be able to do so
because these two are completely
different tasks the trick is that all
this seems all these things seem very
complicated to us
as humans but at the same time they’d
have very clear rules on how to progress
with those and that’s why these things
are simple for AI but at the same time
the things we do without thinking the
natural things such as vision motion
perception something which is almost
impossible to teach AI to do because
these are the skills we cannot define in
zeros and ones and they only seem
natural to us because we as humans got
that kind of as a gift over the
thousands of years of human evolution
and that’s why it’s easy and that’s why
it’s natural so while modern AI can do
only one narrow task what we try to do
in our research group good AI is to
create an AI which would possess the
ability to learn because this we believe
is going to change the rules of the game
dramatically how do we make an AI learn
how does it learn thing
the answer is gradually just like humans
step by step just as baby from the very
beginning to more complicated tasks one
of the approaches to teaching AI is
actually very similar to making kid do
something to working with kids at school
if you have your baby and you want to
motivate it to do something to do some
tasks for example do the homework then
you motivate the baby to take the action
by giving a positive reward like a bar
of chocolate you did a homework that’s
really good to carry on and so after
that baby or an AI realizes that this is
good it’s on the right track and it
keeps repeating the action to receive
the positive reward as long as needed
before it actually learned something
this particular approach is called
reinforcement learning but the
approaches are numerous the key is to
keep the direction from the simple tasks
to the more complicated ones and to make
sure that AI doesn’t forget the things
it learns in the first place but keeps
reusing those skills over and over to
obtain even new ones even more skills
and after that instead of having an AI
which can play one game we will be able
to have an AI which learned how to play
any game and then it can learn how to
play video games and then it can design
its own video game and then it can
design anything you can think of and
then it can think of something to design
and then design it and so on and so on
this should be truly exciting and this
is what we call general purpose AI we
try to achieve general purpose AI as
fast as possible and this is because
future is coming and future is coming
much faster than we all think AI
research is now in its very beginning
the general purpose AI is not yet
created and we don’t know how long is it
going to take
but that’s why in the beginning of the
research the cost of sharing the ideas
is very low while at the same time the
benefit for the bright minds to
cooperate cooperate together and crack
this problem is very big another nice
thing is that computing power is so far
very well controlled and we are not yet
that dependent on the computers to wake
up one morning to realize that you know
they’ve taken the whole world over and I
know this was meant to be a joke but I
know that most of you get really scared
at the moment because AI seem scary to a
lot of people right and people believe
that AI is likely to take their jobs and
to replace them as humans well I’ve been
thinking about this as well a lot and I
believe that there were many cases in
history when something like that a new
technology was also taken in a similar
way I would like to introduce you one
piece of technology which was supposed
to replace thousands of people’s on
their jobs and everybody hated that and
nobody wants it bad introduced what do
you think that might be there are many
options right but this is one of them
remember there were actually people out
there on the cross rose navigating the
traffic out in the cold risking their
lives every day do you think it’s better
that now we can rely on the technology
to coordinate the traffic and those
people can stay in the warm offices to
coordinate that or even do something
completely different
I believe this is the beauty of the
technology and it is very important to
understand that AI is nothing more than
just a tool just something to help us do
different things better something to
augment our capabilities but not a whole
different scary thing how many of you
have been possibly dreaming of studying
history or arts becoming an artist but
then ended up being for example an
accountant because that’s something more
practical and that gets paid better well
I believe that there are many cases like
that and once the invoices we’ll be able
to sort out themselves we as humanity
will have a chance to appreciate the art
more and do the things we really feel
like doing but this is the bright side
but at the same time there is more to
the motivation of having AI as soon as
possible and this is because our brains
are reaching their limits and that’s
unfortunately very serious with
terabytes of data being generated around
us every day with our computers with our
cell phones we just don’t catch up with
that as humanity we cannot progress on
many research problems and we cannot
progress with a space exploration for
example because space is abstract and
remote and you know that big and our hat
it’s just of this size and then our
brains are just limited to the size of
our hat while at the same time computers
can grow as big as we can imagine that
would allow them to be more fast more
efficient than us humans what is more
important is that computer transistors
are also much more reliable than our
brain neurons because I can complain
that I have a headache pretty much every
week you know and I cannot work when I
have headache and computers can work 24
/ 7 on their peak performance every day
non-stop which I believe is exciting and
it would be a shame not to use that
finally it is ok when computer breaks
but it’s not ok if anybody dies during
the research and that’s unfortunately
what’s holding us back in the progress a
lot that’s the cost of the new discovery
is in many cases measure
by humans lives when we talk about space
exploration again we cannot send a
cosmonaut further than for a distance of
a human’s life time because that
wouldn’t just be fair you know and
that’s why we cannot progress on those
problems and humans right now are the
only creatures in the whole universe
that can make decisions in the unknown
circumstances and if we manage to train
our general-purpose AI to make those
decisions if we manage to make it
explore space or do many other things
then I believe we can progress much
further in the scientific progress for
example in the medicine it would be
amazing to have access to the knowledge
of all the doctors in the universe at
once you know having an algorithm which
would be able to create an optimal
treatment court for every patient basing
on this knowledge and that would allow
us to save even more lives and that’s
what AI can do for us and then the
cherry on the cake is that if a I can
possibly solve any problem do you think
it can solve the problem of solving
problems there is actually a research
term for that we call that point
singularity and that looks very exciting
but it’s not coming anytime soon
I guess however we must start right now
to make that happen and as you might
imagine now creating something like that
is a very difficult task and in my
research group
we’re just 20 people so how can
something like that you know be created
by just many people well the good thing
is that AI is very hyped these days
that’s the reason you’re listening to me
right now and that’s the reason why many
people feel excited to be part of the
research another good thing is that
nobody needs a very particular set of
skills to be able to become an AI
researcher people with very different
backgrounds can come together to work on
AI and all of that
creates a perfect momentum to leverage
something called citizen science citizen
science is a concept of opening and
outsourcing the research problems to the
general crowd to people who might be
interested to participate and contribute
to the progress and for that we have
recently launched a competition called
general AI challenge this is a
competition where people from all over
the world could contribute to the
development the first round was devoted
to the problem of gradual learning of
learning step by step and people were
programming their own AI agent able to
solve some tasks today we were
announcing the winners of this first
round and it’s exciting to see that they
come from very different backgrounds we
actually have an astronaut there we have
a quantum physicist we have people of
all different kinds and it’s exciting to
see how all of them can contribute let’s
go now we’re preparing a second round
which would be devoted to the question
of how can we avoid AI development
becoming a race between different
researchers and how to deploy AI in
businesses safely so basically no matter
whether you have programming background
or Social Sciences background each of
you gets an opportunity to join us on
this exciting journey to give us a hand
in creating safe and beneficial
general-purpose
AI exciting rate what I wants you to
keep in my mind what I want you to keep
in your mind after my talk are just
three things the first one the zero one
is the general AI challenge which you
can participate in and then the first
one is that our brains are incredible
and the most precious ability we possess
is the ability to learn that comes from
experience and that’s how we learn
things an AI is learning things in the
very same way too step by step what is
even more important is that AI is just a
tool for us to be better for us to do
things better and it’s not a separate
species the second thing is that
general-purpose AI can be worth more
than anything because this is
ultimately elaborate to any technology
any science any process it can optimize
plenty of things but in order to create
that the collaboration between
researchers needs to start right now and
then finally right here right now is the
best moment in the whole history to
start working on the AI because the
ingredients to having a good a AI are
essentially just three it’s having
computing power it’s having data and
it’s having algorithms and so guess what
computing power has never been cheaper
than today algorithms for inspiration
are being open sourced each day and were
droning with data with terabytes of it
being generates at each second so we
have everything we need and if you have
ever been dreaming of being born in some
other time to witness some amazing
technology emerging or new discoveries
being made then with AI you actually
don’t have to dream because the Golden
Age of AI is right here it’s right now
and everybody can join us on this
journey I am planning to have a very
cool story to tell my kids about how
their mommy helped make the future
happen and each of you can have one too
so join us and thank you
[Applause]
[Music]
[Applause] you
Please follow and like us: