
Translator: eric vautier Reviewer: Cristina Bufi-Pöcksteiner
Hello.
I met artificial intelligence,
I was about 12-13 years, I think,
and the turning of an encyclopedia,
I came across this magnificent object brass copper
– this is arithmetic calculators Pascal
otherwise called the Pascaline.
The young Blaise Pascal, 19, 20 years in 1640,
had the idea of making a machine to make arithmetic operations:
of addition, subtraction and even multiplication.
I was really fascinated by the operation of this machine.
In particular, when you make an addition,
you add 999 + 1 and you have a restraint that should move
from beginning to end to reach 1000.
And I saw how inside
these gears allow this deduction to spread
gear in gear, to reach the highest figure.
For me it was the first manifestation of the fact that
something mechanical could make a very important mental operation
since learned to calculate in school.
Suddenly, it seemed that there was no limit to this idea
we can achieve artificial intelligence from mechanical.
My mother was a biologist; it corrected at the same time copies of students.
A drawing of paramecium, she put 0 because they could not draw.
Here is an older copy of a book besides a Spanish biologist,
– his name is Ramon y Cajal – which is the first,
late 19th century, early 20th century,
to understand that the brain was not brains or a whitish mass.
Inside, with a microscope,
we saw neurons – now all know.
And we see very clearly in this diagram,
that the neuron takes its information from hundreds
or thousands of connections in dendrites,
brings this information to the heart – the cell body of the neuron –
and then makes a decision and sends other information through its axon,
to activate itself with a nerve impulses to other neurons.
Ramon y Cajal has seen a neuron was not a wire:
information goes in one direction – the arrows on the diagram –
so he understood the functioning of the neuron.
These are the two actors in this discussion of artificial intelligence.
To your left, the distant descendant of this machine of Pascal,
the last racing machine microprocessor company Intel:
seven billion transistors into something no bigger than this space –
much as human beings on Earth –
a billion operations per second –
with Pascal’s machine, it was an operation every 30 to 40 seconds.
And on your right,
there is a photograph of a piece of cortex of a mouse,
mouse rainbow sky,
that has been modified for us to see clearly,
in different colors, each of the neurons that comprise it.
Our brains to us is 100 billion neurons,
connected by on average 10 000 synapses to other neurons.
That’s biological intelligence.
So today, since say the 1950-1960 years,
was gradually tried to achieve in electronics,
with the software, with computers,
the same as what was, or what our neuron.
It starts from the idea that one can mathematically represent its operation.
To make simple, it is the mathematical formula that you find there,
I will reassure you.
The neuron takes its 10 000 entries,
each of them is modified positively amplified or attenuated,
based learning.
It is he who goes in your life gradually change your synaptic weights
for that information from other neurons
are added together in the heart of the neuron.
And if it exceeds a certain threshold,
the neuron sends a nerve impulse to another neuron,
or to thousands of other neurons.
We did the same in our computers: stacking layers of neurons –
it’s called the Deep Learning, or neural networks.
And we discovered how gradually change
the weight of these neural networks,
implemented in software,
so that they are able to learn new tasks.
It works like that in our brain:
I see a picture, I reflected,
I decide to act by pressing this button.
A task example among others,
is the recognition of objects in an image.
It is a database that contains 14 million images,
you will not see them all, if we are still here tonight.
And it took people, you and me, human beings,
we annotate these 14 million images with over 80,000 concepts.
There are cats, dogs, horses, forks, trucks,
everything you can imagine; all that was annotated by hand.
And from 30 to 40 years,
we try to improve the ability of a computer
to recognize a cat from a dog, for example, in an image.
In 2011, there were 25% errors.
Each year, it progressed by 0.2, 0.3%.
In 2012 came a Toronto team with Deep Learning –
automatic neural networks –
was halved and reached 12% of errors.
The following year, Google with a lot of resources and more artificial neurons,
arrives and sends 6% error.
Today, this task then,
the machine is more powerful than all of us.
It recognizes objects in images
even when lighting conditions are poor, even with few pixels.
Very blurred,
you do not recognize the person, if the machine.
On a specialized task, the machine is better than us.
more poetic vision,
three neural networks have collaborated,
one that recognizes the image, one that synthesizes the text,
and he who has learned to write hand written.
Together, they do what my grandmother knew how:
a caption on a postcard.
A large bridge spanning the width of a river.
Neural networks also make mistakes.
There a quite amusing example.
you acknowledge all left a praying mantis,
right, a dog.
I did not know what it was, I looked:
is a Bichon Frize – I hope someone knew.
By changing a little bit the images imperceptibly –
for you it is an insect, and a dog –
the neural network, which makes no mistake
it will make a:
in both cases, he finds that it’s an ostrich.
(Laughter)
It proves that what is written in the Bible is true,
since you will find in the book of Job:
“God has deprived the ostrich intelligence. “
A network of neurons today proves what is written in the Bible.
That is the low AI.
It’s an AI, an artificial intelligence specialist.
Today, for any specialized task,
if you have the training data,
you pay enough scientific
to organize the neural network and make learning
we can do as well, or even better, than the human specialist task.
For example, recognize a cancer cell in a medical image,
or fly a telecom network better than we are able to do so.
But do not misunderstand the real capacity of these algorithms.
They are certainly very powerful,
but in terms of computational complexity,
this is roughly between the intelligence of the ant and the frog.
That is why we in the best neural networks today.
And the question of strong AI,
a powerful AI and even more powerful than ourselves.
There are lots of people making predictions.
This one, Mr. Kurzweil, a futurist vaguely transhumanist,
transhumanist even completely around the edges,
who in 2010 said:
“In 2040, you’ll IA
millions of times more powerful than a human mind. “
I do not believe for a moment that this will happen.
What he calls the singularity,
that is to say an AI that is more intelligent than a human being,
able to quickly improve itself,
and therefore, exponentially explode us all.
Well, I give you this graph.
Here, the progress of electronics.
Orange every year, computers are improving.
Here, it was planned for years: the progress of electronics.
But our brains progresses.
In fact, this progresses,
This is the estimate that scientists do to the brain’s ability.
In the scientific literature,
Every year, new literatures say
it is more powerful, it makes more operations than previously believed.
It’s going as fast as electronics.
So that will probably never meet.
Seriously, I think the strong AI,
in terms of raw capacity, high calculation,
it’s in a long time, well beyond 2050
and perhaps even beyond.
In fact, this is not the real subject that interests me.
The real issue is to understand what really understanding.
And I like this definition:
“Intelligence is the way to achieve goals,
the world through ‘calculations. “
I pass over the calculations.
With wheels, electronic, or biological, which are calculations.
The important thing is “in the world”, there are plenty of different.
Our world is in this room, one of the microbe,
that of a bacterium at the bottom of a hydrothermal vent,
these are not the same worlds.
The goals are not the same, so the intelligences are varied.
Intelligence is not what’s on this scheme.
It is not the smartest because we have evolved linearly
from monkey and many other things before.
No. Human intelligence, it is extremely varied.
We are able to perceive and recognize shapes in an image,
to recognize chairs and people,
to reason and decide what is much harder,
especially to decide, to me it seems.
It’s acting, which is even harder, because it is in complex worlds.
This is to interact and thus benefit from the intelligence of others,
and to share ours.
And above all is to have knowledge that is extremely varied.
That of a plumber and that of a nuclear physicist have nothing to do.
And it pays to be a plumber when you have a water outage.
So this knowledge, these various worlds,
and cognitive human capabilities, there are dozens,
emotional or not, I will not go into details.
AI today, she did better than us in some very specific areas,
and not in any other.
In fact, we human beings are nothing in this creation.
That’s a little zoom on the wheel all the organisms living on Earth,
all species living on Earth.
It is small, homo sapiens, in the midst of a wheel,
which actually contains species ranging from protozoa to plants,
through fungi and birds.
In fact, there are millions of life forms,
each endowed with forms of intelligence.
Because life, she calculates.
Our rybosomes,
is a chemical microscopic computer, which continuously calculates.
Ant which moves by releasing pheromones
for guiding the other on the path of the food,
it is a form of individual and collective intelligence.
The bacteria exchange chemical transmitters –
This is a very specialized form of intelligence.
This is an American nutcrackers.
In fact, he has a specialty, just before the end of summer,
10,000 hiding pine seeds to its small reserves for the winter.
It puts them in packs of 5 or 10,
or a thousand caches in the forest,
under a tree, under a sheet, sometimes in the ground.
At the end of winter, he will find his thousand caches in a forest.
With his little bird brain, before you are able,
or that IA is driven by Google to do the same,
it will take a little time.
Only he who knows how, and I think for a while.
Another form of intelligence,
more microscopic,
this is one that will lead these two different species:
a species of fungus, blue, about ten micrometers,
and bacteria, in gray in this image.
In fact, both species talk to each language,
language which is also the most widely spoken on Earth
not English, not French,
language, the terpene,
a chemical molecule that is exchanged between the two species –
and elsewhere, among many other species –
that that one reacts to the behavior of the other.
So there is communication,
so there is intelligence.
A final example, in the basement of a forest,
charts communicate with each other – further chemical molecules –
to warn predators that arrived,
and prepare each reactant.
Better than that, in recent years,
scientists demonstrated that their roots are connected,
and otherwise,
they are connected by fungal mycelium filaments,
able to transport nutrients from one tree to another,
but also carry information in electronic form,
or chemical,
and so again, alert, transmit
and create with what scientists have called the “Wood Wide Web”
the World Wide Web of the trees,
a complex network in each forest in the world –
in any case, in the natural,
much less in the forests planted by man.
So this living ecosystem works far as our brains:
thousands of organizations interacting with decisions, communications,
all coexisting.
My message today,
is that the biosphere, our biosphere,
it consists of intelligence at all scales,
ours, with our big brains extremely capable,
but also that of the trees or bacteria.
And if you look at the complexity of the biosphere in computing power,
ability to calculate, having made some comparisons,
between the largest chip I showed you and our human brain,
multiply by at least 10 million computing power
to reach this equivalence.
And if it is 10 billion people,
one can imagine that together
one is 10 billion times more intelligent one.
Sometimes one wonders.
But if we look at the total capacity
all species and all organisms that inhabit the Earth,
there is more intelligence in bacteria, in total –
There are many –
that there is in the sum of human beings.
So the power 10 22 organisms on Earth calculated together
they have these interactions,
smart they are in very different worlds, worlds that are,
for once,
relatively inaccessible yet to artificial intelligence.
My point today, I want you to keep with you on and off,
This is not so much the idea that, maybe one day
IA such as Odyssey space,
who decides to take power and eliminate all of us;
is the fact that IA, said strong even if it existed,
and it would be in a very long time,
would be confined to specialty niches
which are those that a computer can have.
The fact that we have the sense that we have an ecosystem,
that we lived with a biosphere that surrounds us,
this biosphere can be analyzed as a vast intelligence –
in any case, an ecosystem of intelligences.
She will co-evolve with AI which certainly
will become more and more powerful in terms of computing power,
certainly, will be able to tackle tasks more and more sophisticated,
perhaps even one day to have some forms of creativity –
if you saw the DeepDream software, you say:
“Here, AI has taken LSD. “
So it reminds us of things.
And so we coexistence
between forms of intelligence on scales
and very diverse skill levels,
in a diverse world.
And that is fascinating to watch.
No reason to be afraid.
Like any technology, things are going to happen for good and evil.
The real fear to have
is that we must have of ourselves, the way we manage this world.
What I want you to keep,
IA is that this so long as it remains in computers,
as long as it is not rooted in reality,
in directions in emotions, in chemistry, in physics,
it will lack much of the intelligence of our world.
She will not know what is happening on the ocean floor,
between the roots of the trees,
and that what is really important is the coexistence,
coevolution of this IA, which will evolve and will improve,
and our biosphere in which we live.
Thank you.
(Applause)
Hello.
I met artificial intelligence,
I was about 12-13 years, I think,
and the turning of an encyclopedia,
I came across this magnificent object brass copper
– this is arithmetic calculators Pascal
otherwise called the Pascaline.
The young Blaise Pascal, 19, 20 years in 1640,
had the idea of making a machine to make arithmetic operations:
of addition, subtraction and even multiplication.
I was really fascinated by the operation of this machine.
In particular, when you make an addition,
you add 999 + 1 and you have a restraint that should move
from beginning to end to reach 1000.
And I saw how inside
these gears allow this deduction to spread
gear in gear, to reach the highest figure.
For me it was the first manifestation of the fact that
something mechanical could make a very important mental operation
since learned to calculate in school.
Suddenly, it seemed that there was no limit to this idea
we can achieve artificial intelligence from mechanical.
My mother was a biologist; it corrected at the same time copies of students.
A drawing of paramecium, she put 0 because they could not draw.
Here is an older copy of a book besides a Spanish biologist,
– his name is Ramon y Cajal – which is the first,
late 19th century, early 20th century,
to understand that the brain was not brains or a whitish mass.
Inside, with a microscope,
we saw neurons – now all know.
And we see very clearly in this diagram,
that the neuron takes its information from hundreds
or thousands of connections in dendrites,
brings this information to the heart – the cell body of the neuron –
and then makes a decision and sends other information through its axon,
to activate itself with a nerve impulses to other neurons.
Ramon y Cajal has seen a neuron was not a wire:
information goes in one direction – the arrows on the diagram –
so he understood the functioning of the neuron.
These are the two actors in this discussion of artificial intelligence.
To your left, the distant descendant of this machine of Pascal,
the last racing machine microprocessor company Intel:
seven billion transistors into something no bigger than this space –
much as human beings on Earth –
a billion operations per second –
with Pascal’s machine, it was an operation every 30 to 40 seconds.
And on your right,
there is a photograph of a piece of cortex of a mouse,
mouse rainbow sky,
that has been modified for us to see clearly,
in different colors, each of the neurons that comprise it.
Our brains to us is 100 billion neurons,
connected by on average 10 000 synapses to other neurons.
That’s biological intelligence.
So today, since say the 1950-1960 years,
was gradually tried to achieve in electronics,
with the software, with computers,
the same as what was, or what our neuron.
It starts from the idea that one can mathematically represent its operation.
To make simple, it is the mathematical formula that you find there,
I will reassure you.
The neuron takes its 10 000 entries,
each of them is modified positively amplified or attenuated,
based learning.
It is he who goes in your life gradually change your synaptic weights
for that information from other neurons
are added together in the heart of the neuron.
And if it exceeds a certain threshold,
the neuron sends a nerve impulse to another neuron,
or to thousands of other neurons.
We did the same in our computers: stacking layers of neurons –
it’s called the Deep Learning, or neural networks.
And we discovered how gradually change
the weight of these neural networks,
implemented in software,
so that they are able to learn new tasks.
It works like that in our brain:
I see a picture, I reflected,
I decide to act by pressing this button.
A task example among others,
is the recognition of objects in an image.
It is a database that contains 14 million images,
you will not see them all, if we are still here tonight.
And it took people, you and me, human beings,
we annotate these 14 million images with over 80,000 concepts.
There are cats, dogs, horses, forks, trucks,
everything you can imagine; all that was annotated by hand.
And from 30 to 40 years,
we try to improve the ability of a computer
to recognize a cat from a dog, for example, in an image.
In 2011, there were 25% errors.
Each year, it progressed by 0.2, 0.3%.
In 2012 came a Toronto team with Deep Learning –
automatic neural networks –
was halved and reached 12% of errors.
The following year, Google with a lot of resources and more artificial neurons,
arrives and sends 6% error.
Today, this task then,
the machine is more powerful than all of us.
It recognizes objects in images
even when lighting conditions are poor, even with few pixels.
Very blurred,
you do not recognize the person, if the machine.
On a specialized task, the machine is better than us.
more poetic vision,
three neural networks have collaborated,
one that recognizes the image, one that synthesizes the text,
and he who has learned to write hand written.
Together, they do what my grandmother knew how:
a caption on a postcard.
A large bridge spanning the width of a river.
Neural networks also make mistakes.
There a quite amusing example.
you acknowledge all left a praying mantis,
right, a dog.
I did not know what it was, I looked:
is a Bichon Frize – I hope someone knew.
By changing a little bit the images imperceptibly –
for you it is an insect, and a dog –
the neural network, which makes no mistake
it will make a:
in both cases, he finds that it’s an ostrich.
(Laughter)
It proves that what is written in the Bible is true,
since you will find in the book of Job:
“God has deprived the ostrich intelligence. “
A network of neurons today proves what is written in the Bible.
That is the low AI.
It’s an AI, an artificial intelligence specialist.
Today, for any specialized task,
if you have the training data,
you pay enough scientific
to organize the neural network and make learning
we can do as well, or even better, than the human specialist task.
For example, recognize a cancer cell in a medical image,
or fly a telecom network better than we are able to do so.
But do not misunderstand the real capacity of these algorithms.
They are certainly very powerful,
but in terms of computational complexity,
this is roughly between the intelligence of the ant and the frog.
That is why we in the best neural networks today.
And the question of strong AI,
a powerful AI and even more powerful than ourselves.
There are lots of people making predictions.
This one, Mr. Kurzweil, a futurist vaguely transhumanist,
transhumanist even completely around the edges,
who in 2010 said:
“In 2040, you’ll IA
millions of times more powerful than a human mind. “
I do not believe for a moment that this will happen.
What he calls the singularity,
that is to say an AI that is more intelligent than a human being,
able to quickly improve itself,
and therefore, exponentially explode us all.
Well, I give you this graph.
Here, the progress of electronics.
Orange every year, computers are improving.
Here, it was planned for years: the progress of electronics.
But our brains progresses.
In fact, this progresses,
This is the estimate that scientists do to the brain’s ability.
In the scientific literature,
Every year, new literatures say
it is more powerful, it makes more operations than previously believed.
It’s going as fast as electronics.
So that will probably never meet.
Seriously, I think the strong AI,
in terms of raw capacity, high calculation,
it’s in a long time, well beyond 2050
and perhaps even beyond.
In fact, this is not the real subject that interests me.
The real issue is to understand what really understanding.
And I like this definition:
“Intelligence is the way to achieve goals,
the world through ‘calculations. “
I pass over the calculations.
With wheels, electronic, or biological, which are calculations.
The important thing is “in the world”, there are plenty of different.
Our world is in this room, one of the microbe,
that of a bacterium at the bottom of a hydrothermal vent,
these are not the same worlds.
The goals are not the same, so the intelligences are varied.
Intelligence is not what’s on this scheme.
It is not the smartest because we have evolved linearly
from monkey and many other things before.
No. Human intelligence, it is extremely varied.
We are able to perceive and recognize shapes in an image,
to recognize chairs and people,
to reason and decide what is much harder,
especially to decide, to me it seems.
It’s acting, which is even harder, because it is in complex worlds.
This is to interact and thus benefit from the intelligence of others,
and to share ours.
And above all is to have knowledge that is extremely varied.
That of a plumber and that of a nuclear physicist have nothing to do.
And it pays to be a plumber when you have a water outage.
So this knowledge, these various worlds,
and cognitive human capabilities, there are dozens,
emotional or not, I will not go into details.
AI today, she did better than us in some very specific areas,
and not in any other.
In fact, we human beings are nothing in this creation.
That’s a little zoom on the wheel all the organisms living on Earth,
all species living on Earth.
It is small, homo sapiens, in the midst of a wheel,
which actually contains species ranging from protozoa to plants,
through fungi and birds.
In fact, there are millions of life forms,
each endowed with forms of intelligence.
Because life, she calculates.
Our rybosomes,
is a chemical microscopic computer, which continuously calculates.
Ant which moves by releasing pheromones
for guiding the other on the path of the food,
it is a form of individual and collective intelligence.
The bacteria exchange chemical transmitters –
This is a very specialized form of intelligence.
This is an American nutcrackers.
In fact, he has a specialty, just before the end of summer,
10,000 hiding pine seeds to its small reserves for the winter.
It puts them in packs of 5 or 10,
or a thousand caches in the forest,
under a tree, under a sheet, sometimes in the ground.
At the end of winter, he will find his thousand caches in a forest.
With his little bird brain, before you are able,
or that IA is driven by Google to do the same,
it will take a little time.
Only he who knows how, and I think for a while.
Another form of intelligence,
more microscopic,
this is one that will lead these two different species:
a species of fungus, blue, about ten micrometers,
and bacteria, in gray in this image.
In fact, both species talk to each language,
language which is also the most widely spoken on Earth
not English, not French,
language, the terpene,
a chemical molecule that is exchanged between the two species –
and elsewhere, among many other species –
that that one reacts to the behavior of the other.
So there is communication,
so there is intelligence.
A final example, in the basement of a forest,
charts communicate with each other – further chemical molecules –
to warn predators that arrived,
and prepare each reactant.
Better than that, in recent years,
scientists demonstrated that their roots are connected,
and otherwise,
they are connected by fungal mycelium filaments,
able to transport nutrients from one tree to another,
but also carry information in electronic form,
or chemical,
and so again, alert, transmit
and create with what scientists have called the “Wood Wide Web”
the World Wide Web of the trees,
a complex network in each forest in the world –
in any case, in the natural,
much less in the forests planted by man.
So this living ecosystem works far as our brains:
thousands of organizations interacting with decisions, communications,
all coexisting.
My message today,
is that the biosphere, our biosphere,
it consists of intelligence at all scales,
ours, with our big brains extremely capable,
but also that of the trees or bacteria.
And if you look at the complexity of the biosphere in computing power,
ability to calculate, having made some comparisons,
between the largest chip I showed you and our human brain,
multiply by at least 10 million computing power
to reach this equivalence.
And if it is 10 billion people,
one can imagine that together
one is 10 billion times more intelligent one.
Sometimes one wonders.
But if we look at the total capacity
all species and all organisms that inhabit the Earth,
there is more intelligence in bacteria, in total –
There are many –
that there is in the sum of human beings.
So the power 10 22 organisms on Earth calculated together
they have these interactions,
smart they are in very different worlds, worlds that are,
for once,
relatively inaccessible yet to artificial intelligence.
My point today, I want you to keep with you on and off,
This is not so much the idea that, maybe one day
IA such as Odyssey space,
who decides to take power and eliminate all of us;
is the fact that IA, said strong even if it existed,
and it would be in a very long time,
would be confined to specialty niches
which are those that a computer can have.
The fact that we have the sense that we have an ecosystem,
that we lived with a biosphere that surrounds us,
this biosphere can be analyzed as a vast intelligence –
in any case, an ecosystem of intelligences.
She will co-evolve with AI which certainly
will become more and more powerful in terms of computing power,
certainly, will be able to tackle tasks more and more sophisticated,
perhaps even one day to have some forms of creativity –
if you saw the DeepDream software, you say:
“Here, AI has taken LSD. “
So it reminds us of things.
And so we coexistence
between forms of intelligence on scales
and very diverse skill levels,
in a diverse world.
And that is fascinating to watch.
No reason to be afraid.
Like any technology, things are going to happen for good and evil.
The real fear to have
is that we must have of ourselves, the way we manage this world.
What I want you to keep,
IA is that this so long as it remains in computers,
as long as it is not rooted in reality,
in directions in emotions, in chemistry, in physics,
it will lack much of the intelligence of our world.
She will not know what is happening on the ocean floor,
between the roots of the trees,
and that what is really important is the coexistence,
coevolution of this IA, which will evolve and will improve,
and our biosphere in which we live.
Thank you.
(Applause)
Please follow and like us: