HomeArticlesLES NEWS DE L’IA #2- AI for Humanity, Alphastar, Starcraft, Facebook, Tensormask, MIT, Robot
LES NEWS DE L’IA #2- AI for Humanity, Alphastar, Starcraft, Facebook, Tensormask, MIT, Robot
November 8, 2019
Hello everyone and welcome to the second episode of artificial intelligence news. This week was rich in novelties, especially in technological innovations and we will see all this together, always with the same promise to do all that, in less than 5 minutes. Let’s go! Please do not hesitate to subscribe, to comment, to like the videos. It’s always appreciated and it helps me in the SEO of the channel, thank you! So the first news was on October 30, the Institute of France: the Global Forum on AI for humanity, with, in particular, a closing speech by the President of the Republic, Emmanuel Macron who recalled the French objectives, but also objectives in terms of European policy regarding the artificial intelligence.
He particularly reminded the duty to have a transparency on artificial intelligence, understanding artificial intelligence, not just be content with the fact that it’s a black box, as we often hear. He also reminded to the world that France will acquire, the largest health DataHub in the world.
And so he put the individual data sovereignty at the heart of his speech.
Let’s listen him. “This call for responsibility is also a call for individual responsibility, but also for cooperation. We must, indeed, work more together. For this reason, we need to create more bridges between public policies, public decisions, research and civil society. For these reasons, we need to reconnect science and public decisions. ” So now, we expect a lot from France and even from Europe, to be able to counter the US giants and Chinese giants in the artificial intelligence market. We will see what will happen in the next few years, especially that Emmanuel Macron announced a reform of the research system in France. We know that there are many French researchers who leave for other countries, especially from a salary point of view, the gap between French and American researchers is huge when we compare them on salary.
And that’s really problematic if we want to have a sovereignty regarding the artificial intelligence.
So we’ll see what the next few years reserve for us, but in any case Emmanuel Macron posted a clear goal. Like last year, it is to make France and Europe, champions in the field of artificial intelligence.
And it is the time to hurry when we look at the advance of other countries. And the second news is a totally technological news but that is a very big step forward.
We know that artificial intelligence has been democratized when we started to see AI videos. You know, videos where we could see boxes around people, or around cars, and we could see the computer predicted whether it was a car or a person. Facebook has worked a lot on these detection systems and they have published a new paper, which explains a new technique for analyze images.
In fact, we knew about object recognition techniques, so it was about drawing boxes around objects in pictures.
But, there is another problem that appeared, it was: “OK, but if there are several objects, in several planes on an image, how are we doing detect each object?” So after, we introduced the techniques of image segmentation.
And especially in 2016, Facebook had published a very good paper on this, including the technique of neural network, which was called the mask RCNN which is one of the most used techniques to make the segmentation of image.
And in fact, this technique boiled down to the fact, when to still detect the box around the objects you want, and after segmenting the image in this box.
And with the paper that has been published this week Facebook, they wish to no longer detect the box, but only directly detect the segments in the image, and thus detect the boundaries of each object in the image.
And that’s a small revolution. We will certainly see during the next weeks, the next months, new papers and new applications appearing, with this technique. You see that, it’s still the proof that when we think we managed to do a task in intelligence artificial, we wait, and there are better techniques who appears.
So that’s called TensorMask, and it’s a promising technique. So in third news, well we do not present the company anymore, it’s the company DeepMind, the company that was bought by Google few years ago. We knew them because they became famous by succeeding to beat the champion of Go, with their artificial intelligences ‘Alpha Go 1’ and ‘Alpha Go 2’, which is an improved version.
We had heard that they had managed to beat Starcraft champions too.
So Starcraft is a strategy video game where you have to move little guys, and go to conquer the bases of the opponents.
But this week they did even better. They announced that ‘AlphaStar’ so which is their artificial intelligence on Starcraft, had managed to pass the rank of Grand Master.
So, that means that AlphaStar is stronger than 99.8% of human players. Well, that means that there are still 0.2% of human players who are able to beat AlphaStar to video games. (LOL) But it’s still a huge score.
Especially when we know that DeepMind has bridged AlphaStar.
They narrowed the number of actions she is able to take per second, the number of image refresh, and they also added latency when it sends a command, compared to when it is executed on the Starcraft server.
All this to get closer, to how human deal with video games behind a computer. So here you will ask me:
“But why ?why making an AI that beats everybody to video games?” In fact, the answer in on the DeepMind homepage, they want to find solutions to problems that seem away from the problems of reality that we have, but in fact after, with these resolutions, they want to transfer them to real-world applications.
For example, we have an artificial intelligence learning technique who is able to beat humans to strategy games, who is able to find strategies hyper performing.
Tomorrow we can take this same architecture and put it to help doctors choose specific treatments, to specific diseases. So this is another big hit for DeepMind. So, the 4th news, and it’s pretty funny it comes to us from the MIT. So we knew that drones could build houses.
You know, we program drones and they will build houses, they will assemble them. Imagine, tomorrow your home is built alone.
And it’s a bit what have imagined the MIT scientists, by creating cube-shaped robots.
It reminds us a bit of the video game Minecraft.
So here in fact, we will ask robots to form a cube and they will form this cube by jumping on top of each other. Wow ! I leave you with these images of robots that jump on each other, and I hope you enjoyed this second episode.
Tell me in the comments section if that is the case.
See you next Sunday for a new episode of the News of Artificial Intelligence.
See ya’ !