
Translator: Priscilla RA Reviewer: eric vautier
AI, already, there is still something to seize:
some are very old.
All of you who have taken planes,
have already entrusted their life to an AI.
It’s called autopilots.
There is nothing to be afraid of,
it’s been working for years, and better than the man.
There was also a new change.
There are emerging AIs,
that rely on scientific principles
which will be detailed later so that you understand the issues,
and that make it possible to do extraordinary things.
Typically, we made a big project with the SNCF.
In a very precise way, thanks to that,
you are less often late when you go to work
because we are able to anticipate breakdowns before they occur.
We also do very interesting things with AI
in the field of medical research.
We can detect weak signals,
the explanatory causes of cancer recurrence can be analyzed.
So, it’s a true societal added value.
However, I also see that there are effects of AI
which are more diffuse, less well known
and which, they, can have harmful effects.
My point here is not to sell you AIs,
it’s to make you aware
AI issues
to collectively take ownership of the subject
and you want to debate it.
So, we said that the theme of the text is to open the box,
so, I’ll open the AI box.
New AIs work on a simple principle
called Machine Learning.
Machine Learning is sending data to an algorithm,
to tell him:
“This data, you learn from this base of learning. “
And after a while,
the neural network algorithm or another type of algorithm,
will be able to reproduce in real conditions
what he learned before.
We have two techniques in terms of machine learning.
There is a first more classical technique
called supervised learning.
Typically, you want to create an algorithm
who will recognize elephants in the savannah,
you send him 10,000 pictures of elephants saying “This is an elephant”,
and then you send him 1,000,000 rhinoceros photos
saying, “It’s not an elephant. “
This is called supervised learning.
It’s growing a lot right now,
because we now have the computing capabilities to do it.
It relies on neural networks
which are known since the 80s.
It did not work very well
because we did not have the computing skills to do it.
There is another family of algorithms called
unsupervised learning.
So here you say to the machine:
“I have data, do me distance calculations
to try to see if you can do groupings. “
It corresponds to what you do in SVT, with phylogenetic trees.
You say, “This population is close to this one,
so I put them very close in the tree.
And this population is further from her,
so there is a stronger gap between the two. “
It basically works like this.
It’s not mysterious.
But suddenly, it has consequences that can be problematic.
So, the first problem,
is that AI is the title of the text, we can hack them easily,
that is, we can mislead them
consciously or unconsciously.
Another effect, which is also problematic,
it is because they are enclosing,
since we will see that, working by affinities,
they tend to offer you
what you already like or what you already read.
Which can have a problematic effect in the public debate.
And finally, these AI, we will not lie, will transform the world of work,
because they will substitute for some people
to perform the data processing tasks
which until now were operated by people,
or even start driving cars.
Which begs the question: “What happens to the taxi driver? “
So, in terms of the naivety of AI, there has been a pretty funny experience
which was made by Microsoft just 6 months ago.
They deployed an AI on Twitter
who was supposed to marry the behavior of a 16-year-old girl.
It turned out that there are some clever people
who detected that it was a learning AI.
As a result, they sent him lots of racist and misogynistic messages.
So at the end of the day, she was more of a 16-year-old girl,
but someone abominable.
So, you have to know that an AI believes everything that is said to him.
You say to him, “This is an elephant,”
even if it’s a rhino,
she will understand that it is an elephant.
If we are many to hacker saying anything,
she will end up saying anything seriously.
This is a first subject.
It can be funny but in practice,
some hijackings are really embarrassing.
There is another effect:
it is the effect of the labeling of the data.
So, for supervised learning, you say:
“This is an elephant, this is a rhinoceros,”
everything is fine.
Now, the way we label data …
I took a controversial example
but which shows that there is a real subject around
of the labeling of the data.
I, by habit, tend to say,
“This is a white baby, this is a black baby”.
But in fact, in doing so, I introduce a bias
which is due to my own experience of the subject.
It may be that if we ask the question
to someone from another continent,
she says to me, “This is a baby, this is a white baby. “
The problem of input labeling bias,
is that suddenly the algorithm will match your way of thinking.
It will therefore reproduce the cognitive biases that you have.
It will eventually amplify the problem you were supposed to correct.
It must be known that the AI,
by its ability to integrate the biases we all have,
can lead to rather detrimental effects of reinforcing prejudices.
It is necessary to know that in a concrete way, to prevent these phenomena,
companies like Amazon
have their image data labeled
and their product data
by people spread across five continents,
saying to himself:
“By taking people from different cultures and countries,
I will be able to correct this cognitive bias. “
This is the second detrimental effect: the reinforcement of the biases we all have.
Then there is an even more annoying effect
which has been discussed a lot since the recent elections.
It’s actually the AI mechanisms
use so-called affinity mechanisms.
That is to say, we will recognize in the population where you are
people thinking like you
and we will offer you articles
that people who think like you have found well.
The problem of this effect, we do not realize it,
but when you go to a newsagent,
if you buy Humanity, you are aware
that there are people who buy the Figaro.
On the Earth, everyone is different,
you are confronted with this otherness permanently in the real world.
In the digital world,
it’s a little different.
On Facebook, you have either friends,
people who do not exist,
therefore otherness does not exist.
And the affinity mechanism leads you to read
that things that you thought were right
and to strengthen you in your own prejudices.
And that has an extremely embarrassing effect.
It’s just that,
AI, instead of giving you to think
and to open you horizons,
it locks you into your thinking
or in the way of thinking of your own group.
And this disappearance of otherness produces ugly things.
I saw that after Trump’s election.
It was Silicon Valley Americans
who were astonished to have no one
having voted Trump among their Facebook friends.
How come
considering that the majority of people have in any case voted Trump?
This reinforcing bubble effect makes you lose sight of
that there are lots of people in your country and around you
who do not think like you and do not agree with you.
This disappearance has another effect:
it is that suddenly, one can have a reinforcement of absurd beliefs.
Since we have the Internet,
we see incredible communities being created.
So, there are those who are convinced that clouds of smoke in the sky
are a plot of the CIA.
There are those who are convinced that the Earth is flat.
Platters are an important community on the net.
We must not believe that it is the case only of enlightened people.
They are aware,
they just locked themselves in a bubble
who ends up telling them:
” Around me,
everyone thinks like me,
so I’m right to think what I think. “
We had hope.
Internet would become the super universal encyclopedia
and global knowledge would progress.
We realize today
that this is not necessarily the case.
And there is another detrimental effect.
By dint of living in your bubble,
when in the real world you are confronted with otherness,
it scares you.
So you have a violent reaction:
“Who is this person who does not think like me? “
You are not used to debate.
So we find in the digital world a number of debates that are not very constructive,
where basically people are exchanging names of birds
and it contributes to the increase of conflictuality.
Instead of having a healthy debate about the issues
because you are confronted with different opinions continuously,
you are locked in your bubbles.
If someone does not think like you,
the human reflex is to take the club and hit hard
to be sure to be right.
Watch out for the enclosing effect.
I do not want to ask you,
but on Facebook do you have people
who have political opinions very different from you?
Ask yourself the question.
If you do not have in your Facebook friends
that people who think like you,
you are building a bubble.
Fortunately, there are plenty of solutions to get by,
otherwise I would not be here tonight.
The solutions are not so complex to put in place.
On the one hand, from the moment we have a Machine Learning algorithm,
we can do as in scientific research.
Service operators can be asked to publish the data
that helped create the model and the nature of the model
for it to be open to criticism.
Has there been a bias in the selection of images?
It can be questionable.
Of course, we do not have to ask companies
to entrust their industrial secrets,
but we can very well say:
“After a year, you make public
your algorithm and your learning data,
to see if there was any bias. “
And we look at it like a researcher.
So, it’s a first thing.
There is a trend right now.
There are movements
pushing for the transparency of algorithms.
But we will come to counteract the effects we have seen.
There is another effect:
it’s like AIs like Tay that you’ve seen,
they are AIs who do what is called active learning,
that is to say, she continues to learn continuously.
The problem is that these are very very hard.
So, from the moment we are put an Active Learning AI,
we need safeguards.
So, there has to be a regulatory authority that says:
“You can not put a chatbot on Twitter,
because he’s going to be hacked fast
by a community of not very constructive people. “
So we need regulatory control at that level.
The other effect that begins to be questioned on Facebook,
it is not necessarily to use the sole criterion of popularity
an article or a thought within your group
to propose it to you.
That’s what creates affinity mechanisms.
People you like say the same thing as you.
So, it creates a bubble.
So, we can use other criteria as well
to temper these ranking rules.
We can also say:
“This article comes from a reliable source, I go back. “
Or we can say:
“It’s a popular article of people
who have strong opinions,
who are not from your community
but I propose it to you anyway,
to confront what happens on a daily basis, “
to know, if I buy Humanity, others buy Le Figaro and me, Les Échos,
and the Economic Alternative is bought by others.
I am well aware of all that.
There is no reason that in the digital world this otherness disappears.
Afterwards, there is a question of confronting individuals with divergent opinions.
In the digital world,
if you do not agree,
you delete people on Twitter, you delete people on FB,
it is not very constructive.
And finally, there was a project called Open AI,
which is pushed by GAFA publishers, people like Elon Musk
who say, “Beware, AI poses a real challenge to humanity,
we need to build an ethical framework around that. “
At the moment, there are reflections
on the charts of good use of the data,
and if data scientists should not also have
a code of ethics such as doctors and lawyers.
I think we will come to that.
This posed, we found solutions
for the problems I gave you there.
But there is another problem that is not solved at this stage.
Tomorrow we will have beautiful cars that will be driving alone.
There will be fewer road accidents.
Instead of 2,000,000 deaths a year in the world,
we will only have 5,000 left.
It’s awesome !
We will not prevent that from happening.
But have we thought of all those who live
today’s activity of driving vehicles:
paramedics, taxi drivers, truck drivers?
They will find themselves without work in a near time horizon,
and that corresponds to hundreds of thousands of jobs in France.
If we do not think about the transition,
we will find ourselves in a situation where many people will be losers,
who will be without work,
and will depend on public assistance to live.
If we do not organize this transformation,
it’s going to happen what’s always going on
when hundreds of thousands of people lose their jobs:
there will be violent reactions
like demonstrations or strikes hard enough,
or the coming to power of not very constructive people.
So there, think this transformation …
I have a little statistic.
The 29 states where Donald Trump came in first,
are those where the driver trades are
in the first basins of private employment.
Most people understood
that they would be replaced by machines.
We must stop lying to each other.
We have to make sure that it goes well instead of trying to prevent it.
Indeed, there are several ways to counteract this problem.
There are a number of jobs …
We can imagine that they are easily substitutable.
I imagine there is a majority of engineers here.
It does not matter to tell you that at 40,
you will follow training to change jobs.
There is a part of the population
who is able to do lifelong training.
That’s part of the answer to the question.
It will not treat everyone.
There is another way of seeing things.
It is also to distinguish between employment,
which is the work for which you are paid,
and the work you can do next.
Tomorrow we will probably have people who will be craftsmen,
who will work in fablabs,
who will make their own fridge, their radio machine, their computer.
It will not be linked to a paid activity.
It’s not a job, but it’s a job.
They are not at home doing nothing.
We will see the emergence too
a transformation of the world of employment into a world of work,
where the remuneration will not be automatic for the tasks performed.
So we must also rethink the organization of our societies
around work rather than around the job.
Employment is what you get paid for.
Work is thousands of other things:
volunteering activities, organizing in a fablab, babysitting,
that create value for society as a whole
and that can be considered as work.
So we change a little paradigm,
we substitute employment for work.
We can also imagine, and it’s not necessarily shocking,
make massive compensation plans for employees
affected by this transformation.
After all, we will all have economies of scale.
Connected cars being driven alone,
taxis will cost less,
freight traffic will cost less,
so we will all benefit.
There is nothing shocking to anyone
that part of that profit created by automation,
that it goes to compensate people victims of automation.
There is no reason not to do it.
Finally, and it’s news,
but that was not my point when I launched the TED,
is that, indeed, there are reflections in progress
in different countries on universal income.
It should be known that originally, universal income was an idea
rather defended by the people of Silicon Valley who said:
“Be careful, with what we see in the labs,
we will have a big problem.
There will be millions of people out of work.
We do not know how to treat that,
so much to say right now
that we find the means to compensate them,
by thinking that it will help to ward off the emergency. “
It is also a way of remunerating work by saying to oneself:
“The job is gone but it does not matter, there is still work. “
So that’s it, the different solutions.
They are sometimes contradictory to each other.
One can not necessarily go with the other.
Some have a political lens and some do not.
But in any case, it remains possible to think this transformation.
What can not be done,
it’s as if everything would be fine
and we will continue to find delivery driver jobs
because it’s not true.
There will be no delivery driver jobs in 50 years.
It’s over, it’s gone.
So, here, think the next world.
So, my point is not to say that it is extremely new.
The solution is us, all of us collectively,
what do we decide to do about these AIs?
Basically, this is not the first time
in the history of humanity we ask ourselves questions
on the consequence of a technological innovation.
“Science without conscience is only ruin of the soul” is not new.
I just wanted to tell you tonight
that AIs put this old debate back on the table,
and only because there will be a contradictory debate
and the societal choices that will be made that
It will be alright.
If we leave the subject in the only hands
scientists like me,
it may go wrong.
However, you can call me to find solutions.
(Laughter)
That’s it, thank you very much.
(Applause)
AI, already, there is still something to seize:
some are very old.
All of you who have taken planes,
have already entrusted their life to an AI.
It’s called autopilots.
There is nothing to be afraid of,
it’s been working for years, and better than the man.
There was also a new change.
There are emerging AIs,
that rely on scientific principles
which will be detailed later so that you understand the issues,
and that make it possible to do extraordinary things.
Typically, we made a big project with the SNCF.
In a very precise way, thanks to that,
you are less often late when you go to work
because we are able to anticipate breakdowns before they occur.
We also do very interesting things with AI
in the field of medical research.
We can detect weak signals,
the explanatory causes of cancer recurrence can be analyzed.
So, it’s a true societal added value.
However, I also see that there are effects of AI
which are more diffuse, less well known
and which, they, can have harmful effects.
My point here is not to sell you AIs,
it’s to make you aware
AI issues
to collectively take ownership of the subject
and you want to debate it.
So, we said that the theme of the text is to open the box,
so, I’ll open the AI box.
New AIs work on a simple principle
called Machine Learning.
Machine Learning is sending data to an algorithm,
to tell him:
“This data, you learn from this base of learning. “
And after a while,
the neural network algorithm or another type of algorithm,
will be able to reproduce in real conditions
what he learned before.
We have two techniques in terms of machine learning.
There is a first more classical technique
called supervised learning.
Typically, you want to create an algorithm
who will recognize elephants in the savannah,
you send him 10,000 pictures of elephants saying “This is an elephant”,
and then you send him 1,000,000 rhinoceros photos
saying, “It’s not an elephant. “
This is called supervised learning.
It’s growing a lot right now,
because we now have the computing capabilities to do it.
It relies on neural networks
which are known since the 80s.
It did not work very well
because we did not have the computing skills to do it.
There is another family of algorithms called
unsupervised learning.
So here you say to the machine:
“I have data, do me distance calculations
to try to see if you can do groupings. “
It corresponds to what you do in SVT, with phylogenetic trees.
You say, “This population is close to this one,
so I put them very close in the tree.
And this population is further from her,
so there is a stronger gap between the two. “
It basically works like this.
It’s not mysterious.
But suddenly, it has consequences that can be problematic.
So, the first problem,
is that AI is the title of the text, we can hack them easily,
that is, we can mislead them
consciously or unconsciously.
Another effect, which is also problematic,
it is because they are enclosing,
since we will see that, working by affinities,
they tend to offer you
what you already like or what you already read.
Which can have a problematic effect in the public debate.
And finally, these AI, we will not lie, will transform the world of work,
because they will substitute for some people
to perform the data processing tasks
which until now were operated by people,
or even start driving cars.
Which begs the question: “What happens to the taxi driver? “
So, in terms of the naivety of AI, there has been a pretty funny experience
which was made by Microsoft just 6 months ago.
They deployed an AI on Twitter
who was supposed to marry the behavior of a 16-year-old girl.
It turned out that there are some clever people
who detected that it was a learning AI.
As a result, they sent him lots of racist and misogynistic messages.
So at the end of the day, she was more of a 16-year-old girl,
but someone abominable.
So, you have to know that an AI believes everything that is said to him.
You say to him, “This is an elephant,”
even if it’s a rhino,
she will understand that it is an elephant.
If we are many to hacker saying anything,
she will end up saying anything seriously.
This is a first subject.
It can be funny but in practice,
some hijackings are really embarrassing.
There is another effect:
it is the effect of the labeling of the data.
So, for supervised learning, you say:
“This is an elephant, this is a rhinoceros,”
everything is fine.
Now, the way we label data …
I took a controversial example
but which shows that there is a real subject around
of the labeling of the data.
I, by habit, tend to say,
“This is a white baby, this is a black baby”.
But in fact, in doing so, I introduce a bias
which is due to my own experience of the subject.
It may be that if we ask the question
to someone from another continent,
she says to me, “This is a baby, this is a white baby. “
The problem of input labeling bias,
is that suddenly the algorithm will match your way of thinking.
It will therefore reproduce the cognitive biases that you have.
It will eventually amplify the problem you were supposed to correct.
It must be known that the AI,
by its ability to integrate the biases we all have,
can lead to rather detrimental effects of reinforcing prejudices.
It is necessary to know that in a concrete way, to prevent these phenomena,
companies like Amazon
have their image data labeled
and their product data
by people spread across five continents,
saying to himself:
“By taking people from different cultures and countries,
I will be able to correct this cognitive bias. “
This is the second detrimental effect: the reinforcement of the biases we all have.
Then there is an even more annoying effect
which has been discussed a lot since the recent elections.
It’s actually the AI mechanisms
use so-called affinity mechanisms.
That is to say, we will recognize in the population where you are
people thinking like you
and we will offer you articles
that people who think like you have found well.
The problem of this effect, we do not realize it,
but when you go to a newsagent,
if you buy Humanity, you are aware
that there are people who buy the Figaro.
On the Earth, everyone is different,
you are confronted with this otherness permanently in the real world.
In the digital world,
it’s a little different.
On Facebook, you have either friends,
people who do not exist,
therefore otherness does not exist.
And the affinity mechanism leads you to read
that things that you thought were right
and to strengthen you in your own prejudices.
And that has an extremely embarrassing effect.
It’s just that,
AI, instead of giving you to think
and to open you horizons,
it locks you into your thinking
or in the way of thinking of your own group.
And this disappearance of otherness produces ugly things.
I saw that after Trump’s election.
It was Silicon Valley Americans
who were astonished to have no one
having voted Trump among their Facebook friends.
How come
considering that the majority of people have in any case voted Trump?
This reinforcing bubble effect makes you lose sight of
that there are lots of people in your country and around you
who do not think like you and do not agree with you.
This disappearance has another effect:
it is that suddenly, one can have a reinforcement of absurd beliefs.
Since we have the Internet,
we see incredible communities being created.
So, there are those who are convinced that clouds of smoke in the sky
are a plot of the CIA.
There are those who are convinced that the Earth is flat.
Platters are an important community on the net.
We must not believe that it is the case only of enlightened people.
They are aware,
they just locked themselves in a bubble
who ends up telling them:
” Around me,
everyone thinks like me,
so I’m right to think what I think. “
We had hope.
Internet would become the super universal encyclopedia
and global knowledge would progress.
We realize today
that this is not necessarily the case.
And there is another detrimental effect.
By dint of living in your bubble,
when in the real world you are confronted with otherness,
it scares you.
So you have a violent reaction:
“Who is this person who does not think like me? “
You are not used to debate.
So we find in the digital world a number of debates that are not very constructive,
where basically people are exchanging names of birds
and it contributes to the increase of conflictuality.
Instead of having a healthy debate about the issues
because you are confronted with different opinions continuously,
you are locked in your bubbles.
If someone does not think like you,
the human reflex is to take the club and hit hard
to be sure to be right.
Watch out for the enclosing effect.
I do not want to ask you,
but on Facebook do you have people
who have political opinions very different from you?
Ask yourself the question.
If you do not have in your Facebook friends
that people who think like you,
you are building a bubble.
Fortunately, there are plenty of solutions to get by,
otherwise I would not be here tonight.
The solutions are not so complex to put in place.
On the one hand, from the moment we have a Machine Learning algorithm,
we can do as in scientific research.
Service operators can be asked to publish the data
that helped create the model and the nature of the model
for it to be open to criticism.
Has there been a bias in the selection of images?
It can be questionable.
Of course, we do not have to ask companies
to entrust their industrial secrets,
but we can very well say:
“After a year, you make public
your algorithm and your learning data,
to see if there was any bias. “
And we look at it like a researcher.
So, it’s a first thing.
There is a trend right now.
There are movements
pushing for the transparency of algorithms.
But we will come to counteract the effects we have seen.
There is another effect:
it’s like AIs like Tay that you’ve seen,
they are AIs who do what is called active learning,
that is to say, she continues to learn continuously.
The problem is that these are very very hard.
So, from the moment we are put an Active Learning AI,
we need safeguards.
So, there has to be a regulatory authority that says:
“You can not put a chatbot on Twitter,
because he’s going to be hacked fast
by a community of not very constructive people. “
So we need regulatory control at that level.
The other effect that begins to be questioned on Facebook,
it is not necessarily to use the sole criterion of popularity
an article or a thought within your group
to propose it to you.
That’s what creates affinity mechanisms.
People you like say the same thing as you.
So, it creates a bubble.
So, we can use other criteria as well
to temper these ranking rules.
We can also say:
“This article comes from a reliable source, I go back. “
Or we can say:
“It’s a popular article of people
who have strong opinions,
who are not from your community
but I propose it to you anyway,
to confront what happens on a daily basis, “
to know, if I buy Humanity, others buy Le Figaro and me, Les Échos,
and the Economic Alternative is bought by others.
I am well aware of all that.
There is no reason that in the digital world this otherness disappears.
Afterwards, there is a question of confronting individuals with divergent opinions.
In the digital world,
if you do not agree,
you delete people on Twitter, you delete people on FB,
it is not very constructive.
And finally, there was a project called Open AI,
which is pushed by GAFA publishers, people like Elon Musk
who say, “Beware, AI poses a real challenge to humanity,
we need to build an ethical framework around that. “
At the moment, there are reflections
on the charts of good use of the data,
and if data scientists should not also have
a code of ethics such as doctors and lawyers.
I think we will come to that.
This posed, we found solutions
for the problems I gave you there.
But there is another problem that is not solved at this stage.
Tomorrow we will have beautiful cars that will be driving alone.
There will be fewer road accidents.
Instead of 2,000,000 deaths a year in the world,
we will only have 5,000 left.
It’s awesome !
We will not prevent that from happening.
But have we thought of all those who live
today’s activity of driving vehicles:
paramedics, taxi drivers, truck drivers?
They will find themselves without work in a near time horizon,
and that corresponds to hundreds of thousands of jobs in France.
If we do not think about the transition,
we will find ourselves in a situation where many people will be losers,
who will be without work,
and will depend on public assistance to live.
If we do not organize this transformation,
it’s going to happen what’s always going on
when hundreds of thousands of people lose their jobs:
there will be violent reactions
like demonstrations or strikes hard enough,
or the coming to power of not very constructive people.
So there, think this transformation …
I have a little statistic.
The 29 states where Donald Trump came in first,
are those where the driver trades are
in the first basins of private employment.
Most people understood
that they would be replaced by machines.
We must stop lying to each other.
We have to make sure that it goes well instead of trying to prevent it.
Indeed, there are several ways to counteract this problem.
There are a number of jobs …
We can imagine that they are easily substitutable.
I imagine there is a majority of engineers here.
It does not matter to tell you that at 40,
you will follow training to change jobs.
There is a part of the population
who is able to do lifelong training.
That’s part of the answer to the question.
It will not treat everyone.
There is another way of seeing things.
It is also to distinguish between employment,
which is the work for which you are paid,
and the work you can do next.
Tomorrow we will probably have people who will be craftsmen,
who will work in fablabs,
who will make their own fridge, their radio machine, their computer.
It will not be linked to a paid activity.
It’s not a job, but it’s a job.
They are not at home doing nothing.
We will see the emergence too
a transformation of the world of employment into a world of work,
where the remuneration will not be automatic for the tasks performed.
So we must also rethink the organization of our societies
around work rather than around the job.
Employment is what you get paid for.
Work is thousands of other things:
volunteering activities, organizing in a fablab, babysitting,
that create value for society as a whole
and that can be considered as work.
So we change a little paradigm,
we substitute employment for work.
We can also imagine, and it’s not necessarily shocking,
make massive compensation plans for employees
affected by this transformation.
After all, we will all have economies of scale.
Connected cars being driven alone,
taxis will cost less,
freight traffic will cost less,
so we will all benefit.
There is nothing shocking to anyone
that part of that profit created by automation,
that it goes to compensate people victims of automation.
There is no reason not to do it.
Finally, and it’s news,
but that was not my point when I launched the TED,
is that, indeed, there are reflections in progress
in different countries on universal income.
It should be known that originally, universal income was an idea
rather defended by the people of Silicon Valley who said:
“Be careful, with what we see in the labs,
we will have a big problem.
There will be millions of people out of work.
We do not know how to treat that,
so much to say right now
that we find the means to compensate them,
by thinking that it will help to ward off the emergency. “
It is also a way of remunerating work by saying to oneself:
“The job is gone but it does not matter, there is still work. “
So that’s it, the different solutions.
They are sometimes contradictory to each other.
One can not necessarily go with the other.
Some have a political lens and some do not.
But in any case, it remains possible to think this transformation.
What can not be done,
it’s as if everything would be fine
and we will continue to find delivery driver jobs
because it’s not true.
There will be no delivery driver jobs in 50 years.
It’s over, it’s gone.
So, here, think the next world.
So, my point is not to say that it is extremely new.
The solution is us, all of us collectively,
what do we decide to do about these AIs?
Basically, this is not the first time
in the history of humanity we ask ourselves questions
on the consequence of a technological innovation.
“Science without conscience is only ruin of the soul” is not new.
I just wanted to tell you tonight
that AIs put this old debate back on the table,
and only because there will be a contradictory debate
and the societal choices that will be made that
It will be alright.
If we leave the subject in the only hands
scientists like me,
it may go wrong.
However, you can call me to find solutions.
(Laughter)
That’s it, thank you very much.
(Applause)
Please follow and like us: