Meet Elon Musk's Open AI trainee Kevin Frans when he was 15 years old

There is nothing other than "high youth" for me.

Since being founded by Elon Musk nearly two years ago, OpenAI's research-based research lab has published a series of research reports - just like its name implies: OpenAI - OpenAI . However, a report published last week was different from the previous one: its main research author was just a high school student.

Kevin Frans.

He is Kevin Frans , and he is currently applying for college. Frans designed neural net , the system that tech giants still design and use to identify your voice, face - when you're only 15 years old. My inspiration comes directly from the artificial intelligence machines capable of mastering intellectual games such as Go and Poker, and based on that, I built my own system, every time a small puzzle piece.

"I really like the way people create computers that do things that were still impossible before," Frans said with a fresh smile when asked by Wired reporter. One of the systems I created here: an interactive website, which automatically colored a drawing, colored in manga style.

Frans first approached OpenAI when confronted with one of the many research issues that this group still faces. I had a little progress, but when the research didn't work, I sent an email to OpenAI researcher John Schulman for advice. After discussing an optimization algorithm, allowing multiple computers to analyze and solve problems, Schulman read about Frans' blog and had a pleasant surprise.

"I didn't expect those emails to come from a guy in high school," he said.

Frans had a chance to meet Schulman directly when he interviewed for an internship at OpenAI. When I was accepted to work in Mission County, San Francisco, Frans was the only trainee who didn't have a degree or went to college. In the new workplace, she conducts research and handles problems that still hinder the development of robots and artificial intelligence: how do these machines apply knowledge, old data to solve new problem.

People do not take much effort to do this. Once we learn a recipe or experience something, we'll do the same thing again and again without much difficulty. But with machine-learning software is different, with each new problem, it has to go through the training process: persistent, even if this new problem has many things in common with the old problem.

Frans' new study collaborates with Schulman and three other researchers from California Berkeley University to show promising progress in this area of ​​research.

"If it can solve this problem, it will not only be a leap in the robot industry but will also affect other factors in artificial intelligence research , " Frans said. He developed an algorithm to help a virtual robot learn how to use its leg system effectively, knowing which movement to use for any action properly.

In the tests, this algorithm helps two virtual robots with 2 and 4 legs do many things, including moving in the maze, walking, crawling quickly and accurately. Research has been submitted to ICLR, the leading conference on machine learning."Kevin's research provides a new approach to an old problem, some of which are more successful than anything that has been tried before , " Schulman said.

But Frans also told me that I didn't rely on the family for this AI research project, although he was not the only "technology man" in the house. Frans's father is currently working in silicon chip design industry at Xilinx.

Olga Russakovsky, professor of machine vision research, focuses on technology and how to enable image-based problem analysis machines, saying that it is possible to research and publish scientific reports on machien learning. when I was very young it was a strange thing. In general, students are not interested in artificial intelligence, they need to build excitement and be taught for a long time.

Technology limitations are also a big obstacle: Frans' home computer is not strong enough to handle big ideas, so Frans uses bank cards, opening an account using cloud-service. Google computing to test the code that I designed. I recommend that other girls who want to learn about machine learning should try it."The best thing to do is step up and actually try it, do it yourself with your own hands," Frans said.

I myself want to be able to help younger generations to study artificial intelligence. Perhaps it will start from his 7-year-old brother."I think she likes the code. Maybe when she's a little older, I can help."