[Applause]
humans love to pretend that objects are
like people we give them names we talk
to them we get attached to them like Tom
hangs here in castaway talking to his
friend of olive oil called Wilson but
when it comes to computers how do you
feel about machines thinking like people
and if machines can think like people
what if one day they begin to think
better than us these are the sorts of
questions that people keep asking me how
far away are we from being outsmarted by
computers how far is the development of
artificial intelligence I’m a researcher
in artificial intelligence at Cambridge
University and it’s been my life’s work
to address these questions artificial
intelligence is making the news
headlines almost daily we read about the
computer deep mind beating the world
grandmaster in the game of go driverless
cars are being tested on our streets
drones are going to be delivering our
parcels films like ex machina and Blade
Runner are portraying our future
populated with robots that are
indistinguishable from us but what does
the future of AI really hoped for us
when I was a child I loved maths what I
really really was fascinated about was
the beauty and the elegance of the
solutions that I saw and sometimes even
found by myself and I was also
interested in people so I wondered how
do you people come up with such
intuitive answers
as I went through my education and work
journey I began to investigate this kind
of human thinking and try to make
computers do the same I want computers
to find solutions that are intuitive to
humans let me give you an example what I
mean by intuitive human thinking here’s
a checkerboard it’s called a mutilated
checkerboard because two diagonal
corners have been removed now a famous
problem asks can you cover this mutilate
the checkerboard with Domino’s there are
many ways that we could try to do that
still no now consider this idea let’s
color this board black and white like a
chess board and now it’s immediately
obvious that we have more black squares
than white squares because when we
mutilated the board we removed two white
squares but we need there to be the same
number of black squares and white
squares because a domino covers one
black and one white square so it’s
obvious we will never be able to cover
this mutilated checkerboard with
dominoes
this is something so easy and simple for
people to see and understand no symbols
no algebra no numbers I’m done now what
do you think how does a computer go
about solving this problem this is how
lots of symbols
lots of algebra lots of mathematical
concepts it’s completely unintelligible
and the intuition behind
human solution is completely lost in it
computers can’t capture this kind of
human thinking yet this is what
fascinates me how do we emulate this
kind of human thinking on machines now
artificial intelligence has long strived
to achieve computer thinking and
behavior that is indistinguishable from
a human so why is it that AI is such a
hot topic now
well there are three main advances there
are enabling us to make huge progress in
AI research and they are the
availability of masses of data that we
all generate all the time the massive
computing power that we have access to
today and the new and more sophisticated
machine learning algorithms that we
haven’t considered before so in terms of
data every click you make is recorded
somewhere every video you watch every
item you buy online everything you say
to your friends on social media and this
data is used to learn something and to
gain insights about you individually so
that you can get personalized
recommendations such as what kind of
film you might like but more than that
it’s also used to learn about groups of
people and their behavior computing
power well you can rent it really
cheaply from anyone of for example
Amazon’s data centers and gain access to
its 80,000 or so computer servers for
comparison the Watson computer that beat
a human in jeopardy
in 2011 used 90 servers so we have
instant access to massive computing
power
but if the neumes and more sophisticated
machine learning algorithms that are at
the heart of the AI revolution unlike in
the past when we build these specialized
systems that knew about problems and
went about solving these specific
problems the machine learning algorithms
of today know nothing about the problems
they only know about learning they only
know about learning
take for example autocorrect on your
phone
previously your spelling mistake would
rarely be picked up and even then only
if it was misspelled in a particular way
and this was because these were
specialized system that had very
specific rules today I can type into my
phone in Slovene and even though my
phone knows nothing about this foreign
language it will be able to learn how I
spell surveen words and even how I
misspell them now this is interesting
because it seems rather counterintuitive
doesn’t it so these machine learning
algorithms know nothing about the
problem instead they are general in that
they can take any data and crunch it and
spot patterns in fact any problem that
can be recast as a statistical algorithm
running on huge datasets can now be
revisited and tackled again using these
new creative approaches so for example
the modern dating apps use new machine
learning algorithms and the result is
your ideal partner did you know
apparently one in three of us use them
but there are huge challenges in trying
to achieve what sometimes is referred to
as general AI a flexible adaptable
learning self improving system that does
all the things that the human does in
ways that are indistinguishable from a
human computer systems today are super
fast are dealing with lots of data
that far exceeds human capabilities but
they do so in ways that are machine
oriented and most of the time
incomprehensible to humans one of the
things that they lack is the ability to
think intuitively like humans not yet
now this brings me back to my love of
maths and people and in my work I look
at how people solve mathematical
problems and then I make computers do
the same in particular I’m interested in
how people solve problems with using
intuitive methods like using diagrams
here’s an example I’m going to add up
all the numbers from 1 to any number
that you give me yes any number so any
number from 0 to 100 let’s say so I I
have a chance even so if look can I hear
a number ok I hear 53 so the sum of all
the numbers from 1 to 53 is 1431 how did
I do that
well I used the theorem this one so I
didn’t really add up all the numbers for
1 to 53 instead I multiplied it 53 times
54 divided by 2 because I used this
theorem it’s much easier and faster that
way so this is the kind of problem that
you may remember from your maths classes
in school so and my children are now
learning about it so let me see show of
hands how many of you remember this from
school ah some some not all let’s see if
I can jog your memory and I will show
you how I was taught this problem in
school we used a diagram
so my job is to add up all the numbers
from to show that adding up all the
numbers from 1 to any number n equals to
n times n plus 1 divided by 2 so I’m
gonna represent because we’re talking
about numbers I’m gonna represent them
with using counters and in fact I’m
gonna start showing one side of the
equation n times n plus 1 and so I will
take a rectangle in fact I’m gonna take
n to be a particular number to be a
number 5 so I’m gonna draw a diagram a
rectangle that is five counters wide and
six counters tall and that’s my n times
n plus 1 now imagine cutting this
rectangle down the diagonal like this
and you can see that what we get is two
identical triangles and each is five
counters tall but my problem halfs my
rectangle it says n times n plus 1
divided by 2 so I’m just gonna take one
of these triangles and this now
represents one side of my equation now
my job is to show that that equals to
the sum of all increasing numbers now
imagine cutting this triangle into rows
like this and now you will notice that I
get a sequence of increasing numbers so
my triangle shows that 1 plus 2 Plus 3
plus 4 plus 5 equals 2 5 times 6 divided
by 2 which is 15 but more than that it
also we can see that my solution is
gonna hold for any number number 5 or
number 3 that you gave me or 99 or 1,001
it will be the same and we can also see
that it’s a correct solution so I wanted
to capture this on computers and wanted
to make them solve problems in exactly
this way
which is which is exactly what I did so
I built a system where these spatial
manipulations are the inferences of the
solution I then used formal methods to
justify why a specific a concrete case
like for number five can capture a
solution for any number and the system
shows them that the solution is correct
so I showed that computers can solve
problems in similar ways that people
solve them but I also wondered could
they discover new unseen solutions
things that people have not understood
before so in order to do that we could
use AI methods like machine learning
that I mentioned before I focused on
problems in mathematics but the general
principles of intuitive thinking can be
applied to any area of human endeavor
and this is where I’m going to be
focusing next making machines discover
new knowledge new intuitive human
solutions so will AI outsmart humans
will these intelligence systems pose a
threat to humans one day I see the
future in the hands of human experts but
heavily supported and helped with these
intelligent systems machines are gonna
work and collaborate with us in order to
help us work more effectively but also
to do things that are beyond our
capability alone should we be scared no
should we restrict or slow down the
development of artificial intelligence
definitely not the positives far
outweigh the negatives in fact I think
it’s more risky not to pursue the
development of artificial intelligence
we are nowhere near a Terminator style
future
but we are only a decade or so away from
the future where intelligent systems
help doctors diagnose better where
personalized drugs are used to treat
diseases more effectively where AI
systems help drivers drive more safely
surgeons operate more accurately where
these system enable us all ask people to
converse across multiple languages this
is perhaps the most exciting time in the
history of mankind the rate at which
technology is advancing is exponential
and with rapid change there will always
be winners and losers so now how what do
we do how do we become winners in this
listen learn to understand embrace the
possibilities and make everything that
we know today
work for us so we can turn our planet
into a better place thank you [Applause]