“An AI that would unplug us? Science fiction”, says a researcher

• In Cannes, a brand new exhibition, World artificial intelligence Cannes festival, is entirely devoted to artificial intelligence "capable of simulating human intelligence", according to the definition of Larousse.

• These technologies are present today "in almost everything related to digital technology", according to researcher Antoine Bordes.

• They advance research, develop, but will not be able to reach a level of “general” intelligence, in all areas, like human intelligence, he tells us.

They are everywhere.

On our computers.

In our smartphones.

A simple internet search.

An automatic translation… Artificial intelligence (AI) is taking up more and more space in our lives.

Without, necessarily, that we are fully aware.

They are also mobilizing more and more professionals around the world.

In Cannes, a brand new exhibition, the World artificial intelligence Cannes festival, is entirely devoted to these technologies “capable of simulating human intelligence”, according to the definition of Larousse.

This Saturday, this event, organized in the congress center of the Croisette, is even open to the general public.

The opportunity to take stock of the development of these AIs.

How far can they go?

What are their limits ?

interviewed researcher Antoine Bordes.

At the CNRS until 2014, this specialist has since been the co-director of Fair (Facebook artificial intelligence research), Meta's fundamental research laboratory in artificial intelligence.

What does the birth of this new event in Cannes symbolize?

There were already shows that dealt with artificial intelligence, such as the Mobile world congress in Barcelona, ​​on smartphones, the CES in Las Vegas, which is much more generalist, or even Vivatech in Paris.

But this event in Cannes is really the first of its kind to be entirely dedicated to Artificial Intelligence, from research to consumers.

It is an innovation that shows the place it occupies today in the economy.

How fast has technology evolved in recent years?

There are areas that are still very complicated for machines.

But for some, considered classic, such as object recognition, machine translation, speech recognition, progress has been really very fast.

Much faster than the experts projected.

What are the applications in which AI is used today?

Let's start from the opposite, it will go faster.

Today, in everything related to digital, what are the applications that do not use AI?

There are very very few.

It is used to create new experiences, new entertainment by offering new tools to creators and brands.

It is obviously used to organize platforms like ours;

for example, to protect them by detecting content that should not be there.

The whole ecosystem is affected by AI to varying degrees.

AI is also advancing science…

It actually helps enormously in scientific discovery.

In biology, in chemistry.

For medical research, too, where it enables extremely rapid progress, whether in data processing or in the way of generating hypotheses.

I can give two examples.

One of which we developed called Fast MRI.

It's an algorithm that allows image reconstruction to speed up this medical imaging procedure, which is long and can be a bit traumatic.

Usually, it is a test that lasts 15 to 30 minutes.

We manage to divide this time by about four by taking fewer images and asking the AI ​​to complete the sequence.

It's been clinically tested with doctors.

It's completely open source like all the research we do and which will be able to be used by manufacturers.

Our colleagues at DeepMind [Google] have also developed software that can predict the structure of proteins from their amino acid sequence.

It will be a fundamental building block for biology tomorrow.

Let's talk about the future.

AI will continue to grow over time, but will its development have a limit?

Researcher Yann LeCun, who works with us, often explains that AI is more like animal intelligence than human intelligence.

That is to say, it could be much more advanced in specific segments, but not in general.

It will be more and more efficient to perform certain categories of tasks.

On the other hand, seeing the same AI doing a cooking recipe as well, analyzing proteins or even translating all languages, seems more complicated to me.

To have an intelligence that we could call "general", which would be able to do everything, we are still very far from it.

So, yes, in that sense, I think there is a limit.

So there is no risk of seeing, as in some science fiction films, the machine acquire a conscience and turn against its creator?

There was a debate on Twitter that there could be sentient AIs.

But I think the community is still quite skeptical about all this.

At the same time, we haven't really defined what the consciousness of machines could be… It should be done.

In the meantime, we are creating regulatory frameworks to define their use, the data on which they are based and the applications with which we have the right to use them.

And an AI that would be so intelligent that it would be able to go beyond its limits in a subtle way and manage to unplug us in order to be able to stay connected... this is a theory that is precisely science fiction.

Compared to this animal intelligence we are talking about, some think that an AI may want to defend itself from being disconnected in the name of a survival instinct... This seems very fanciful to me.

high tech

Cannes: Transformed into NFT, the city's heritage will soon be auctioned off to invade the metaverse

Health

Nice: A new blood and early screening for lung cancer soon to be tested throughout France

• high tech

• Nice

• walking sticks

• Artificial intelligence

• Paca