WEBVTT

00:00:00.000 --> 00:00:03.204
(Toby Walsh) I wanna talk 
about artificial intelligence: it's --

00:00:03.674 --> 00:00:05.089
I'm a professor of artificial intelligence

00:00:05.089 --> 00:00:09.271
and its a great time, 2015, 
to be working in AI.

00:00:09.544 --> 00:00:10.544
We're making real palpable progress

00:00:10.544 --> 00:00:12.820
and there's loads of money 
being thrown at us.

00:00:12.820 --> 00:00:17.000
Google just spent 
five hundred million dollars --

00:00:17.000 --> 00:00:21.989
pounds buying an AI startup called Deep Mind a couple of weeks ago,

00:00:21.989 --> 00:00:25.172
today they announced that they were 
going to spend a billion dollars

00:00:25.671 --> 00:00:28.469
setting up an AI lab in Silicon Valley.

00:00:29.001 --> 00:00:31.591
IBM is betting about 
a third of the company

00:00:31.830 --> 00:00:34.858
on their cognitive AI computing effort.

00:00:35.084 --> 00:00:38.216
So it's really interesting time to be working in AI.

00:00:38.631 --> 00:00:41.782
But the first thing I wanted to 
help inform you about

00:00:42.250 --> 00:00:46.792
is what is the state of art, 
what progress have we made in AI

00:00:46.792 --> 00:00:51.187
because Hollywood paints 
all these pictures,

00:00:51.187 --> 00:00:53.488
these mostly dystopian pictures of AI.

00:00:53.495 --> 00:00:58.672
Whenever the next science fiction movie
comes out, I put my head in my hands

00:00:58.672 --> 00:01:01.755
and think Oh my God, what do people think
that we're doing?

00:01:01.755 --> 00:01:08.729
So, I wanted to start by just giving you
a feel for what actually is really capable.

00:01:09.081 --> 00:01:13.872
So a couple of years ago, IBM Watson
won the game show Jeopardy 1,

00:01:13.872 --> 00:01:19.737
the million dollar prize in the game show,
Jeopardy P, the reigning human champions.

00:01:19.737 --> 00:01:23.357
Now, you might think, well that's just
a party trick, isn't it?

00:01:23.357 --> 00:01:27.625
It's a -- pour enough of Wikipedia 
and the internet into a computer,

00:01:27.625 --> 00:01:30.543
and it can answer general knowledge
questions.

00:01:30.543 --> 00:01:33.204
Well, you guys are being a bit unfair to
IBM Watson,

00:01:33.204 --> 00:01:35.881
there are the cryptic questions
they are answering.

00:01:35.881 --> 00:01:40.878
But just to give you a real feel for 
what is technically possible today,

00:01:40.878 --> 00:01:43.586
something that was announced
just two days ago:

00:01:43.834 --> 00:01:51.869
some colleagues of mine at NII in Japan
passed the University Entrance Exam

00:01:51.869 --> 00:01:53.433
with an AI program.

00:01:53.433 --> 00:01:56.326
Now, I thought long and hard about
putting up a page of maths.

00:01:56.326 --> 00:01:58.007
I thought, well, I'm going to get --

00:01:58.007 --> 00:02:00.007
half of the audience is going to 
leave immediately

00:02:00.007 --> 00:02:01.354
if I put up a page of math

00:02:01.354 --> 00:02:03.133
But I wanted you to see, just to feel

00:02:03.962 --> 00:02:07.403
the depth of questions 
that they were answering.

00:02:07.403 --> 00:02:11.507
So this is from the maths paper, you know,
a non trivial, sort of,

00:02:11.507 --> 00:02:15.148
if you come from the UK, 
A-level-like math question

00:02:15.148 --> 00:02:16.468
that they were able to answer

00:02:17.272 --> 00:02:21.889
Here is a physics question
about Newtonian dynamics

00:02:21.889 --> 00:02:23.552
that they were able to answer.

00:02:23.552 --> 00:02:28.545
Now they got 511 points, out of
a maximum of 950.

00:02:28.939 --> 00:02:32.391
That's more than the average score
of Japanese students

00:02:32.391 --> 00:02:34.007
sitting the entrance exam.

00:02:34.007 --> 00:02:38.542
They would have got into most 
Japanese universities with a score of 511.

00:02:38.908 --> 00:02:40.621
That's what's possible today.

00:02:40.621 --> 00:02:44.134
Their ambition in 10 years' time is to get
into Tokyo, University of Tokyo,

00:02:44.134 --> 00:02:46.434
which is one of the best universities
in the world.

00:02:48.488 --> 00:02:50.475
So, this is why I put up 
a picture of Terminator,

00:02:50.475 --> 00:02:53.468
because whenever I talk to the media
about what we do in AI,

00:02:53.476 --> 00:02:55.769
they put up a picture of Terminator,
right?

00:02:55.769 --> 00:02:58.098
So I don't want you to worry 
about Terminator,

00:02:58.098 --> 00:03:02.117
Terminator is at least 
50 to 100 years away.

00:03:02.582 --> 00:03:05.985
and there are lots of reasons why
we don't have to worry about it,

00:03:05.985 --> 00:03:07.248
about Terminator.

00:03:07.248 --> 00:03:08.979
I'm not going to go and spend
too much time

00:03:08.979 --> 00:03:11.348
on why you don't have to worry about Terminator.

00:03:11.676 --> 00:03:13.334
But there is actually things 
that you should worry about,

00:03:13.334 --> 00:03:15.352
much nearer than Terminator.

00:03:17.007 --> 00:03:20.160
Many people have said, 
Stephen Hawkins has said

00:03:20.160 --> 00:03:22.713
that, you know, AI is going to spell
the end of the human race.

00:03:23.295 --> 00:03:27.961
Elon Musk chimed in afterwards, said
"It's our biggest existential threat."

00:03:29.793 --> 00:03:32.841
Bill Gates followed on by saying,
"Elon Musk was right."

00:03:33.498 --> 00:03:39.519
Lots of people have said that's a --
AI is a real existential threat to us.

00:03:39.519 --> 00:03:42.887
I don't want you to worry about
the existential threat that AI

00:03:42.887 --> 00:03:44.900
or Terminator is going to bring.

00:03:45.953 --> 00:03:47.711
There's actually a very common confusion,

00:03:48.121 --> 00:03:51.139
which is that it's not the AI 
that's going to be the existential threat,

00:03:51.139 --> 00:03:52.139
it's autonomy, it's autonomous systems.

00:03:54.386 --> 00:03:56.042
What's in the (check) --

00:03:56.042 --> 00:03:57.972
it isn't it going to wake up 
any time in the morning and say:

00:03:57.972 --> 00:04:02.109
"You know what? I'm tired 
of playing Jeopardy,

00:04:02.109 --> 00:04:04.173
"I want to play Who Wants to Be a Millionaire!

00:04:05.299 --> 00:04:07.468
"Or wait a second, I'm tired of playing 
game shows,

00:04:07.468 --> 00:04:09.393
I want to take over the universe."

00:04:09.821 --> 00:04:12.195
It's just not in its code, there is no way,

00:04:12.195 --> 00:04:15.065
it's not given any freedom to think 
about anything other

00:04:15.065 --> 00:04:17.000
than maximizing its Jeopardy score.

00:04:18.219 --> 00:04:21.209
And it has no desires, no other desires than,

00:04:21.228 --> 00:04:23.482
other to improve its maximum scores.

00:04:23.482 --> 00:04:26.256
So I don't want you to worry about Terminator,

00:04:27.350 --> 00:04:29.102
but I do want you to worry about jobs.

00:04:29.974 --> 00:04:32.166
Because lots of people, 
lots of very serious people,

00:04:32.166 --> 00:04:35.076
have been saying 
hat AI is going to end jobs,

00:04:35.076 --> 00:04:38.157
and that is a very great consequence 
for anyone working in education,

00:04:38.157 --> 00:04:42.269
because, certainly, the jobs that are going 
to exist in the future

00:04:42.269 --> 00:04:44.925
are going to be different 
than the jobs that exist today.

00:04:46.880 --> 00:04:49.724
Now, who has an odd birthday?

00:04:51.251 --> 00:04:52.938
Well, I haven't told you 
what an odd birthday is yet,

00:04:52.938 --> 00:04:54.990
so someone has an odd birthday, like me.

00:04:54.990 --> 00:04:57.858
OK. Who was born on an odd-number 
day of the month?

00:04:57.858 --> 00:04:59.995
I was born on the 11th of April, right?

00:05:00.824 --> 00:05:03.368
Come on, it's half the room, 
I know it's half the room.

00:05:03.368 --> 00:05:04.923
(Laughter)

00:05:04.923 --> 00:05:08.690
OK.Well, you want to have 
an odd birthday, by the way,

00:05:09.059 --> 00:05:12.504
because that means, in 20 years' time, 
you will be a person with a job.

00:05:13.426 --> 00:05:16.277
As opposed to the even people, 
who won't have jobs.

00:05:16.277 --> 00:05:20.662
That's certainty -- if you believe 
lots of serious people,

00:05:22.141 --> 00:05:25.010
you might have missed this news 
on Friday the 13th,

00:05:25.010 --> 00:05:27.778
I thought this was a rather
depressing news story

00:05:27.778 --> 00:05:29.748
....... (check) comparison otherwise ...... (check)

00:05:29.777 --> 00:05:31.527
but the chief economist, Bank of England,

00:05:31.922 --> 00:05:36.230
went on the record saying 50% of jobs 
were under threat in the UK.

00:05:36.948 --> 00:05:40.420
And he's not the first serious person
who should know what he's talking about

00:05:40.420 --> 00:05:42.736
who said similar things.

00:05:43.029 --> 00:05:46.255
There was a very influential Merrill Lynch
report that came out a month or to ago

00:05:46.270 --> 00:05:49.430
saying very similar things about 
the impact of AI,

00:05:49.430 --> 00:05:51.375
robotics, automation on jobs.

00:05:53.439 --> 00:05:56.278
And some of this goes back to, I think, 
one of the first reports

00:05:56.278 --> 00:05:59.288
that really hit the press, 
that really got people's attention,

00:05:59.288 --> 00:06:02.075
was a report that came out of 
the Oxford Martin School.

00:06:02.075 --> 00:06:05.702
They predicted that 47% of jobs 
in the United States

00:06:06.110 --> 00:06:08.550
were under threat of automation 
in the next 20 years.

00:06:10.234 --> 00:06:13.594
We followed that up 
with a very similar study and analysis

00:06:13.880 --> 00:06:15.961
for jobs in Australia, where I work.

00:06:16.865 --> 00:06:19.323
And because it's a slightly different 
profile of workers,

00:06:19.323 --> 00:06:24.329
of the work force in Australia, we came up
with a number of around 40%.

00:06:24.329 --> 00:06:27.129
These are non trivial numbers, right? 
40-50%.

00:06:27.589 --> 00:06:32.497
No, just an aside: 47%, I don't know 
why they didn't say 47.2%, right?

00:06:32.497 --> 00:06:35.551
You can't believe a number 
that's far too precise

00:06:35.551 --> 00:06:37.842
when you're predicting the future,
but nevertheless,

00:06:37.842 --> 00:06:39.532
the fact that it's of this sort of scale,

00:06:39.532 --> 00:06:44.601
you've got to take away: it wasn't 4%, 
it was roughly about half the jobs.

00:06:47.784 --> 00:06:50.495
Now, let's put some context to this.

00:06:50.495 --> 00:06:52.413
I mean, is this really a credible claim?

00:06:52.413 --> 00:06:55.549
The Number One job 
in the United States today: truck driver.

00:06:55.962 --> 00:06:57.526
Now you might have noticed,

00:06:57.536 --> 00:07:00.079
the autonomous cars
are coming to us very soon.

00:07:00.506 --> 00:07:03.579
We're going to be having -- tried
the first trial of autonomous cars

00:07:03.579 --> 00:07:06.760
on the roads, public roads of Australia, 
three weeks ago.

00:07:07.371 --> 00:07:09.657
The Google Car has driven 
over a million kilometers

00:07:09.657 --> 00:07:12.542
-- or the Google cars, rather, 
have driven over a million kilometers,

00:07:12.907 --> 00:07:15.050
autonomously, on the roads of California.

00:07:15.893 --> 00:07:18.553
In 20 years' time, we are going to have 
autonomous cars.

00:07:18.553 --> 00:07:20.255
We're also going to have 
autonomous trucks.

00:07:21.033 --> 00:07:24.769
So if you are in the Number One profession
in the United States,

00:07:24.769 --> 00:07:28.480
you have to worry that your job 
is not going to be automated away.

00:07:29.712 --> 00:07:32.494
The Number Two job in the United States 
is salesperson.

00:07:33.343 --> 00:07:36.248
Again, since we use the internet,

00:07:36.248 --> 00:07:39.618
we've actually mostly automated 
that process ourselves,

00:07:39.618 --> 00:07:43.382
but it's clear that a lot of those jobs 
are going to be disappearing.

00:07:43.382 --> 00:07:46.736
So I think these claims 
have a lot of credibility.

00:07:48.896 --> 00:07:53.104
There's actually a nice dinner party game 
that my colleagues in AI play

00:07:53.104 --> 00:07:55.736
at the end of our conferences,
where we sit around

00:07:55.736 --> 00:07:58.250
and the game is, you have to name a job

00:07:59.115 --> 00:08:02.671
and then, someone has to put up 
some credible evidence

00:08:02.671 --> 00:08:05.748
that we're actually well on the way 
to actually automating that.

00:08:05.748 --> 00:08:08.152
And this game is almost impossible to win.

00:08:08.737 --> 00:08:10.423
If I had more time, 
I'd play the game with you.

00:08:11.293 --> 00:08:15.260
The only -- about the only winning answer
is politician.

00:08:15.650 --> 00:08:17.931
(Laughter)

00:08:17.931 --> 00:08:20.747
They will certainly regulate that
they'll be the last to be automated.

00:08:20.747 --> 00:08:22.731
But that's about 
the only winning answer we have.

00:08:24.049 --> 00:08:28.836
So -- and it's not just technology 
that is the cause of this.

00:08:28.836 --> 00:08:33.035
There's many other, really, 
sort of rather unhelpful trends.

00:08:33.035 --> 00:08:35.558
If you were trying to set up 
the world's economy,

00:08:35.558 --> 00:08:38.842
you would not put these things 
all down on the table at the same time:

00:08:38.842 --> 00:08:40.825
the global, 
ongoing global financial crisis,

00:08:40.825 --> 00:08:43.840
which seems like 
it will never disappear, I think;

00:08:44.755 --> 00:08:48.013
the fact that we're all living longer: 
this is great, great news for us

00:08:48.013 --> 00:08:50.333
but bad news for employment;

00:08:50.961 --> 00:08:54.425
the impact of globalization, the fact that 
we can outsource our work

00:08:54.425 --> 00:08:56.191
to cheaper economies.

00:08:56.191 --> 00:09:00.191
All of these things 
are compounding the impact

00:09:00.191 --> 00:09:02.957
that technology is having 
on the nature of work.

00:09:04.749 --> 00:09:09.725
And this transformation is going to be 
different than the last one,

00:09:09.725 --> 00:09:11.357
the Industrial revolution.

00:09:12.021 --> 00:09:15.591
There's no hard and fast 
rule of economics that says:

00:09:16.911 --> 00:09:20.039
"As many jobs need to be created 
by a new technology as destroyed."

00:09:20.039 --> 00:09:22.952
Every time we have a new technology, 
of course, new jobs are created.

00:09:22.952 --> 00:09:25.883
There's lots of, there's thousands, 
hundreds of thousands of new jobs

00:09:26.204 --> 00:09:27.848
enabled by technology today.

00:09:28.355 --> 00:09:32.484
But there's no reason that they have to 
balance exactly those that are destroyed.

00:09:33.092 --> 00:09:36.199
In the last -- in the last revolution, 
that did happen to be the case.

00:09:36.853 --> 00:09:40.930
A third of the population was working 
out in the fields, in agriculture.

00:09:40.930 --> 00:09:44.552
Now, worldwide, 
it's 3 or 4% of the world's population

00:09:44.552 --> 00:09:45.657
working in agriculture.

00:09:45.657 --> 00:09:48.029
Those people are working 
in factories and offices now.

00:09:48.029 --> 00:09:51.920
We employ far more people than we did 
at the turn of the 19th century.

00:09:52.629 --> 00:09:55.993
But this one looks different, this 
information revolution looks different.

00:09:55.993 --> 00:10:00.233
It looks like it has the potential 
to take away more jobs, perhaps,

00:10:00.233 --> 00:10:01.529
than it does.

00:10:01.529 --> 00:10:03.684
And one of the other things is that 
we used to think

00:10:03.684 --> 00:10:05.293
it was the blue-collar jobs.

00:10:06.509 --> 00:10:10.937
And that's true: if you go 
to a car factory today, sure enough,

00:10:10.937 --> 00:10:12.578
there are robots 
that are doing the painting,

00:10:12.578 --> 00:10:14.800
there are robots 
that are doing the welding.

00:10:14.800 --> 00:10:16.584
But nowadays, it's white-collar jobs:

00:10:16.584 --> 00:10:20.006
it's journalists, it's lawyers, 
it's accountants,

00:10:20.006 --> 00:10:21.667
these jobs that are under threat.

00:10:21.667 --> 00:10:25.444
These graphs here show 
the percentage change in employment

00:10:25.444 --> 00:10:31.244
and the change in employment rates.

00:10:31.660 --> 00:10:34.672
And it's the middle, the middle class, 
white-collar professions

00:10:34.672 --> 00:10:37.923
that we thought that you would go 
to university to make yourself safe,

00:10:37.926 --> 00:10:40.551
but it seems to be the ones 
that are most under threat.

00:10:40.551 --> 00:10:42.646
If you are a ....... (check) 
it's probably --

00:10:43.277 --> 00:10:46.172
you're too cheap to be replaced 
by something automated.

00:10:46.172 --> 00:10:49.963
But if you're a more expensive person, 
and this means (check)

00:10:49.963 --> 00:10:53.427
that the rich are getting richer and 
inequalities that we are seeing in society

00:10:53.427 --> 00:10:55.281
that are distressing our societies to day,

00:10:55.281 --> 00:10:58.284
seem to be magnified 
by these technological changes.

00:10:59.899 --> 00:11:01.761
And there is so many frightening graphs,

00:11:01.761 --> 00:11:03.723
Go and read 
Thomas Picketty (check), I encourage you.

00:11:04.079 --> 00:11:05.980
Go and look at one of his books
and you can see here

00:11:05.980 --> 00:11:09.619
that we're seeing 
a constant improvement in productivity.

00:11:10.079 --> 00:11:12.860
Technology is buying us 
those improvements in productivity,

00:11:12.860 --> 00:11:17.888
is increasing our wealth,
but there's a leveling off of employment.

00:11:18.462 --> 00:11:20.187
And so, the challenge, then, is how

00:11:20.753 --> 00:11:23.651
-- it's a question for society, 
not for a technologist like myself --

00:11:23.651 --> 00:11:25.777
how do we all benefit 
from this rising tide,

00:11:25.777 --> 00:11:30.648
not so that it is the rich get richer 
and the rest of us get further behind.

00:11:31.613 --> 00:11:36.443
So, many parts of many jobs 
looks likely (check) to be automated.

00:11:36.767 --> 00:11:39.624
One confusion is this: people say 
these jobs are going to disappear.

00:11:39.624 --> 00:11:43.009
Actually, it seems to be more likely that 
many parts of your job will be automated.

00:11:43.009 --> 00:11:45.892
But that still means that there is 
perhaps less employment around.

00:11:46.201 --> 00:11:48.860
So how can you make yourself 
more future-proof?

00:11:49.382 --> 00:11:51.291
Well, I have two pieces of advice 
as a technologist,

00:11:51.291 --> 00:11:53.950
in terms of what's going to be 
technically possible in AI.

00:11:54.215 --> 00:11:57.886
Either you've got to embrace the future, 
so become like me,

00:11:58.605 --> 00:12:01.725
become someone who's working 
on trying to invent that future.

00:12:02.218 --> 00:12:03.671
And if you're not technically minded, 
that's fine:

00:12:03.671 --> 00:12:07.127
I've got the other part of the equation, 
the other answer to your question

00:12:07.127 --> 00:12:10.437
which is completely at the other end 
of the spectrum, which is:

00:12:10.786 --> 00:12:14.011
focus on those things that find as AI, (check)
the hardest things,

00:12:14.011 --> 00:12:15.672
making computers more creative,

00:12:15.672 --> 00:12:18.719
making computers that can understand 
your emotional state,

00:12:19.269 --> 00:12:23.111
focusing on emotional terms 
and not intellectual intelligence.

00:12:24.765 --> 00:12:26.866
So, how safe is education?

00:12:26.866 --> 00:12:28.728
The room is here, full of people 
working in education.

00:12:28.728 --> 00:12:30.104
How safe are your jobs?

00:12:30.104 --> 00:12:33.452
Well these are the numbers 
from that Oxford Martin report I wrote.

00:12:33.789 --> 00:12:35.096
So if you're a telemarketeer,

00:12:35.096 --> 00:12:37.908
99% chance that 
you're going to be automated.

00:12:37.908 --> 00:12:39.162
Not surprising, right?

00:12:39.162 --> 00:12:41.208
Easy to automate, it's down the phone.

00:12:42.589 --> 00:12:45.322
Some of the numbers 
I just don't want you to take away:

00:12:45.322 --> 00:12:48.195
they used machine only, they used AI 
to actually generate the report

00:12:48.195 --> 00:12:50.240
and I don't believe some of the numbers,

00:12:50.240 --> 00:12:53.141
I don't believe: 
"Bicycle repairmen: 94%."

00:12:53.141 --> 00:12:54.550
There's no chance in hell

00:12:54.550 --> 00:12:56.804
that the bicycle repair person 
is going to be automated:

00:12:56.804 --> 00:12:58.916
far too cheap and intricate a job.

00:12:58.916 --> 00:13:03.840
"Parking lot attendant: 87%."
I don't know why it's not 100%:

00:13:03.840 --> 00:13:06.271
well, you're not going to have
parking lot attendants, for sure.

00:13:06.770 --> 00:13:08.290
But look: luckily,

00:13:08.290 --> 00:13:10.408
you and me are 
right at the bottom of the list,

00:13:10.743 --> 00:13:12.974
down at the 0's and 1%, right?

00:13:13.451 --> 00:13:15.381
I think those numbers 
probably underestimate

00:13:15.981 --> 00:13:19.203
how replaceable or 
how irreplaceable we are,

00:13:19.203 --> 00:13:23.315
but nevertheless, 
you can take some heart away

00:13:23.315 --> 00:13:25.418
from the sort of numbers you see there.

00:13:26.488 --> 00:13:28.217
And the reason being?

00:13:28.217 --> 00:13:30.873
The first reason is, because 
we're dealing with people,

00:13:30.873 --> 00:13:32.436
we're trying 
to understand people,

00:13:32.436 --> 00:13:36.469
understand their motivations,
what are their mental blocks.

00:13:36.469 --> 00:13:38.404
These are things that are 
really hard to get computers,

00:13:38.404 --> 00:13:41.445
I can tell you, really hard to get 
computers, program computers to do. 13:41