so can machines think this idea the
singularity the time in history where a
computer program will be so smart that
it will be able to create another
computer program even smarter and so on
and so forth until humanity’s demise
ensues it’s not likely Alan Turing
devised the test to test this idea ask
the question can machines think his test
was let’s have a computer trying to fool
a human into thinking it itself is human
by means of conversation a simple and
you may think ill-defined test but would
it surprise you to know that almost 70
years on this test has not yet been
passed not close generic AI strong AI
the artificial intelligence that can do
everything humans can do and more and
better it’s nowhere near here however
arguably one of the biggest leaps in
that direction is a technology that’s
called deep learning in fact when in the
news you hear about developments in AI
more often than not you are hearing
about developments or applications of
deep learning
you see whilst other AIS are capable of
solving small problems such as playing a
game or finding the path in a maze
what’s exciting and new about deep
learning is that it’s one technology
that can solve a range of different
problems
it is what we call a classifier it’s one
of the best classifiers yet but what is
that it can be boiled down to cats and
dogs
imagine you want a computer to be able
to distinguish between cats and dogs you
could teach it about the structure about
what they look like you could teach them
specifically to recognize specific
patterns but what’s revolutionary about
deep learning is that it’s entirely
based on samples the only thing you need
is many many samples many many pictures
of cats and dogs and the Internet has
plenty of errors and the system will
learn on its own to distinguish between
them and you may think well this is a
very simple example and you’d be
surprised the number of different
problems in the world that can be
reduced to just classify for example
classifying between a good chess
position and a bad chess position
classifying between an empty road and a
road with pedestrians on it if you’re
building a self-driving car so deep
learning is versatile and yes it can be
used to solve and play games better than
humans you may have heard recently about
alphago defeating the world’s best go
player and then recently again defeating
the world’s best go player again the
other world best brain deep learning has
another very important characteristic
which is because we only need samples to
train it
we don’t need expert
the engineers that built alphago are not
expert go players they barely needs to
know the rules and the system will learn
on its own and is this same technology
that can be applied to a number of
problems which capture the imagination
of the media from lip-reading to medical
diagnosis and what is fascinating is
that even though it can solve all these
problems deep learning has a number of
limitations we’ve been here today about
how to future-proof your life we’ve been
hearing from our first speaker about
having to learn things and about other
speakers about being involved with new
technology and the additional
contribution I like to make is that
sometimes new technology can be
overrated deep learning is powerful it’s
exciting it also has some limitations as
I said you need lots of pictures of cats
and dogs and we have plenty of those but
when your problem becomes more complex
you need more and more data and big data
quickly becomes the need for a big
computing system which requires a lot of
energy and a lot of money and that’s not
the only limitation it was shown
recently that you can trick deep
learning you can show it an image of a
cat we love this and then we can change
every pixel in the image by a very small
amount in a very clever way in such a
way that the resulting image to us still
looks like a cat but the deep learning
system will think it’s a dog
or a table or a Mac or whatever you want
to trick it into believing deep learning
is limited and it’s exciting and it’s
versatile and it’s powerful but all
those adjectives are shared I contend by
pretty much every disruptive technology
we’ve encountered here’s an example it’s
called a computer or a personal computer
and let me ask you this do you
understand every single electronic
component and its function inside your
computer or your phone I know I don’t
and that doesn’t stop me using them it
doesn’t stop me
learning about them every day more and
more as I need if computers were
invented today and people were saying
they will make your job obsolete
they will render your role irrelevant
well how would you react panic
complacency I contend and this is my
challenge to you that we treat deep
learning in the same way that we treat
our computers we learn it and it will be
difficult but so was using spreadsheets
the first time around it may not even be
that efficient but then again neither
were word processes when they first
appeared the good news is that because
the only thing we need to train a deep
learning system is examples samples of
data it is relatively nice Treacher
relatively simple to actually implement
how simple nine nine lines of code
that’s how many lines of code it takes
me to write a deep learning program so
here is the idea that I think it’s worth
sharing let us do away with this elitism
of skill let us make the machine
learning experts less unique
let’s democratize the use of technology
including machine learning go away and
find out what those nine lines of code
are and if you don’t know how to code
learn how to code and if you can do it
in fewer than nine lines get in touch my
point is don’t panic learn to code and
you
[Applause]
singularity the time in history where a
computer program will be so smart that
it will be able to create another
computer program even smarter and so on
and so forth until humanity’s demise
ensues it’s not likely Alan Turing
devised the test to test this idea ask
the question can machines think his test
was let’s have a computer trying to fool
a human into thinking it itself is human
by means of conversation a simple and
you may think ill-defined test but would
it surprise you to know that almost 70
years on this test has not yet been
passed not close generic AI strong AI
the artificial intelligence that can do
everything humans can do and more and
better it’s nowhere near here however
arguably one of the biggest leaps in
that direction is a technology that’s
called deep learning in fact when in the
news you hear about developments in AI
more often than not you are hearing
about developments or applications of
deep learning
you see whilst other AIS are capable of
solving small problems such as playing a
game or finding the path in a maze
what’s exciting and new about deep
learning is that it’s one technology
that can solve a range of different
problems
it is what we call a classifier it’s one
of the best classifiers yet but what is
that it can be boiled down to cats and
dogs
imagine you want a computer to be able
to distinguish between cats and dogs you
could teach it about the structure about
what they look like you could teach them
specifically to recognize specific
patterns but what’s revolutionary about
deep learning is that it’s entirely
based on samples the only thing you need
is many many samples many many pictures
of cats and dogs and the Internet has
plenty of errors and the system will
learn on its own to distinguish between
them and you may think well this is a
very simple example and you’d be
surprised the number of different
problems in the world that can be
reduced to just classify for example
classifying between a good chess
position and a bad chess position
classifying between an empty road and a
road with pedestrians on it if you’re
building a self-driving car so deep
learning is versatile and yes it can be
used to solve and play games better than
humans you may have heard recently about
alphago defeating the world’s best go
player and then recently again defeating
the world’s best go player again the
other world best brain deep learning has
another very important characteristic
which is because we only need samples to
train it
we don’t need expert
the engineers that built alphago are not
expert go players they barely needs to
know the rules and the system will learn
on its own and is this same technology
that can be applied to a number of
problems which capture the imagination
of the media from lip-reading to medical
diagnosis and what is fascinating is
that even though it can solve all these
problems deep learning has a number of
limitations we’ve been here today about
how to future-proof your life we’ve been
hearing from our first speaker about
having to learn things and about other
speakers about being involved with new
technology and the additional
contribution I like to make is that
sometimes new technology can be
overrated deep learning is powerful it’s
exciting it also has some limitations as
I said you need lots of pictures of cats
and dogs and we have plenty of those but
when your problem becomes more complex
you need more and more data and big data
quickly becomes the need for a big
computing system which requires a lot of
energy and a lot of money and that’s not
the only limitation it was shown
recently that you can trick deep
learning you can show it an image of a
cat we love this and then we can change
every pixel in the image by a very small
amount in a very clever way in such a
way that the resulting image to us still
looks like a cat but the deep learning
system will think it’s a dog
or a table or a Mac or whatever you want
to trick it into believing deep learning
is limited and it’s exciting and it’s
versatile and it’s powerful but all
those adjectives are shared I contend by
pretty much every disruptive technology
we’ve encountered here’s an example it’s
called a computer or a personal computer
and let me ask you this do you
understand every single electronic
component and its function inside your
computer or your phone I know I don’t
and that doesn’t stop me using them it
doesn’t stop me
learning about them every day more and
more as I need if computers were
invented today and people were saying
they will make your job obsolete
they will render your role irrelevant
well how would you react panic
complacency I contend and this is my
challenge to you that we treat deep
learning in the same way that we treat
our computers we learn it and it will be
difficult but so was using spreadsheets
the first time around it may not even be
that efficient but then again neither
were word processes when they first
appeared the good news is that because
the only thing we need to train a deep
learning system is examples samples of
data it is relatively nice Treacher
relatively simple to actually implement
how simple nine nine lines of code
that’s how many lines of code it takes
me to write a deep learning program so
here is the idea that I think it’s worth
sharing let us do away with this elitism
of skill let us make the machine
learning experts less unique
let’s democratize the use of technology
including machine learning go away and
find out what those nine lines of code
are and if you don’t know how to code
learn how to code and if you can do it
in fewer than nine lines get in touch my
point is don’t panic learn to code and
you
[Applause]
Please follow and like us: