Hello everyone, and welcome to this first episode of the news of the AI.
We will go through the news of the week of artificial intelligence together, and all in less than five minutes.
Let’s start now ! The first big news is Amazon who are opening a store with 4000 robots in Bretigny-sur-Orge in France.
So 4000 robots is huge, but no worry the warehouse is not totally automated.
In fact, the robots here will only help to simplify the work of the employees.
As we can see, the robots are actually fetching products, and take them to the employees who can put them in the package and send them.
The French Minister Cédric O wants to be totally confident about the new implementation of this warehouse by Amazon.
“There are six times more robots in Germany and Germany is in full employment and so there is a big digital transformation which comes with other transformation as the ecological transformation.
What we want to do in France, what we want to achieve is to ensure that France is part of this transformation.
” But we know that Amazon has already hurt the Retail sector in France.
And is this warehouse the sign of a new snub to all French retailers?There will only be time to tell us.
Also Amazon did not give any indication on the future of this warehouse.
In the end, will it be totally automated?We do not know more for now.
Anyway, one thing is sure is that Amazon has arrived with his robots in France, and I can not wait to see more.
In this week’s news, also in artificial intelligence, at Station F, there was another edition of ‘France is AI’.
‘France is AI’ is an event in artificial intelligence, in France.
Several researchers, several professionals, come to explain their novelties.
We saw a lot of interventions from Deep Mind, Facebook, Google, but also schools and French universities who came to present the research they are doing, and who expose them to the public.
And that’s really interesting, for example this work at CentraleSupélec, on medical imaging.
DeepMind spoke to us about multi-agents learning”.
“I’m a scientist at Deep Mind and I’m also a professor of Computer Science at the University of Liverpool, UK.
Today I’m going to talk about the topic of multi-agent system evaluation and training and I hope to convince you that this is a very important topic for the success of our field.
” Because in the end, multi-tasking, multi-agents, is how we represent out world at best.
So imagine, you’re driving, you have a prediction engine that will give you the shortest way it can be.
Could be your Waze, or your Google Maps.
And in front of you, there is an accident.
So, there is a traffic jam being created and the GPS will suggest you take the exit to find another route.
And there you have two cases: either you say to yourself:”ok I trust the GPS, and I go because all the others will stay in the traffic jams and I could go out.
” And so, you will be alone on the road, or you say to yourself:”No I stay because maybe all the other people will go out and in the end I will be alone on the road here.
“So here your future in the end, it does not depend solely on your action it depends on your action in the middle of a bunch of other agents that are here cars and their actions.
So it was super interesting.
There was a conference of Yoshua, who presented us with his new passion, his new research topic that is to develop consciousness in an artificial intelligence.
And this is the third news we will see right away.
So Yoshua wants to find a consciousness in the intelligence artificial.
But for that he says that Deep Learning today is a little broken.
Why? Because in Deep Learning, a model hoped that the test data has the same distribution as the training data.
But in real life, it’s not always the case.
Today, we want our models to generalize.
A model that generalizes what does it mean? It means that we make him learn about a certain dataset and we put it in nature.
And in the end, he will succeed in giving good predictions.
No matter in fact, the nature of the event or the consequences that will be in the event.
For example, if I have prices data for a house I want to be able to predict the price of a house, in Paris, Marseille, Rennes, New York or Tokyo.
And that’s very, very complicated to do.
So Yoshua wants to find a little awareness of artificial intelligence.
In fact, it is necessary that the machine can understand the effects of each event to deduce the causes, without necessarily having to to learn and that’s a bit like that human thinks finally and acts.
For example, when we drive, we do not wait to have fifteen accidents to learn that we must not have an accident.
So for a machine it’s the same.
For example, the autonomous car must understand by itself that if she has an accident, she will be broken.
And so she will do everything to avoid accidents.
And so the game is, , if we had to schematize it, it is ta machine should understand that it must not fall, otherwise she will break.
But we do not need to break fifteen autonomous cars to make an autonomous car understand that it must not break.
And that’s it, in fact, the consciousness he wants to develop in the machines.
Go to finish I show you the new toy developed by OpenAI:it is a hand that solves a Rubik’s cube.
So we remember the work in 2018 OpenAI, develop a robot with dexterity.
So we saw a robot that handled cubes.
Now they have succeeded in making this happen, they have this machine full of dexterity solved a Rubik’s cube.
So, we see that the hand is a bit slow.
The robot was able to solve a Rubik’s cube in 4 minutes and 8 seconds.
I am going to leave you on these images of a robot that solves a Rubik’s cube.
See you next week for a new episode of the news of the AI.
Bye Bye !.