How did a San Francisco programmer achieve 12 unrealistic goals?

Фото: Depositphotos

When 24-year-old programmer from San Francisco, Max Deutsch decided to defeat the world chess champion, he absolutely did not understand this game. This did not stop him and did not even cause much concern. He had a whole month to prepare for a meeting with the most powerful chess player in the world. "Lenta.ru report»I found out what can be achieved during this time.

A year ago, Max Deutsch made a list of the skills he wants to master. He wanted to know how to draw portraits and read freestyle without preparation; how to improvise on an electric guitar and collect a Rubik's cube in 20 seconds; how to backflip off and how to develop artificial intelligence for automatic driving.

It takes months or even years to learn such things, but Max took exactly 30 days for each skill. And incomplete: he was going to learn only after work and on the road. If you don’t donate to sleep and chat with friends, then a couple of hours a day were left for the hobby, not more. He thought that was enough.

The victory over the chess champion was the most ambitious point of the plan. Max left her for last, and for a start he started doing something simpler.

Task list

November 2016 of the year Max dedicated to memory training. He decided to learn how to memorize the order of playing cards in the deck in 2 minutes. The winner of the International Memory Championship, Alexander Mullen, does it in 16 seconds, but for an amateur and 2, minutes is an excellent result. Max remembered the order in 10 once longer.

In December, Max learned to draw. He armed himself with video courses for beginning artists and, after 14,5 hours, painted the face of British illusionist Derren Brown on paper. Then I practiced a bit in celebrity photos and on 15 the day I took up a self-portrait. The result exceeded expectations. It's hard to believe that just a few days ago, this man did not draw any better than a child.

In May, Max developed artificial intelligence for driving. He wrote a program that analyzes video from the camera on the windshield of the car and decides where to turn the steering wheel and which pedals to press. General ideas spied in scientific work nVidia, pattern recognition piled on the finished library OpenCV, and the neural network has implemented technology-based Tensorflowwhich developed in Google. A full-fledged program for driving a car is much more difficult, but Max did not expect that he would receive the finished product - he just wanted to understand how such systems are arranged. And achieved this goal.

One by one, Max struck out almost all the points from the plan. He accomplished everything he planned. Only the match with the champion is left.

Max knew the rules of chess from childhood, but he played on strength a couple of times a year. His ELO rating did not exceed 1100 - this is the beginner level. To be a grandmaster, you need at least 2500. Chess players with a rating of at least 2700 have chances for a world championship. And only 2800 players on the planet overcame the 4 bar. The rating of the strongest chess player on the planet, 27-year-old Norwegian Magnus Carlsen, is 2837. Max wanted to defeat him.

Magnus is a rare talent. He was only 13 years old when he stood against Garry Kasparov, who at the time was considered the number one chess player. After 10 years, Magnus became the world champion, first in classical chess, and a year later in rapid and blitz. Before him, no one got all the 3 crowns at once.

Max did not expect that chess celebrity will descend to an amateur like him. At first, he was going to play not with the living Magnus, but with a computer program that mimics his style. But the idea interested The Wall Street Journal. Journalists offered to help with the organization of this match. Max, of course, was not against, but the champion did not refuse. The party was appointed on the ninth of November.

Professionals spend on training thousands of hours, and Max had a little more than a month. For such a period it is impossible to become a strong chess player, at least in the usual sense of the word. But Max hoped that there was a workaround. “If I can't learn to play chess humanly, I have to do it like a computer,” he decided.

Chess programs use the fact that the machine works much faster than the human brain. She calculates all possible variants of the party’s development several steps ahead, building a branched decision tree, evaluates each branch, and then chooses a move that leads to the most advantageous positions in the future.

A person is not able to keep all the moves in his head, as a computer does. But in the algorithm, which evaluates the perspective course with decent accuracy without analyzing the entire decision tree, nothing is impossible. According to Max's estimates, to conduct such calculations in your mind, you need to memorize a little more than 12 of thousands of parameters. If you own the mnemonic techniques - it is quite possible task.

Scramble

Max created a neural network and “fed” her the recordings of thousands of parties in which Magnus Carlsen participated. He hoped that the network would learn to distinguish promising moves and give it the parameters that are needed for mental calculations. At first, the accuracy of the algorithm steadily increased. This continued until she reached 70%. Overcoming this ceiling did not work. Max added data for training, but this only slowed down the data processing. Computational power was sorely lacking.

Time was running out, and Max still had no numbers to memorize. When he flew to Hamburg to meet with Magnus, the computer continued to process the data. “I didn’t expect my computer to take so long to find this algorithm,” he confessed to the correspondent before the game. The Wall Street Journal. “We are talking now, and the car is still puffing over it.”

It was too late to refuse the game. Without the algorithm, Max had to play only on his own.

Magnus arrived. It would seem that the champion has nothing to fear, but he does not relax. “There have been surprises in the past,” he explains.

But luck could not be eternal. On the 12 move, Max moves the queen, not noticing that he is hitting a knight. This is a fatal mistake. Then his position only gets worse. On 39, the game ends with a champion victory.

When it was all over, Magnus smiled for the first time in the game and extended his hand to the loser. “The first ten moves were good,” he praised. When Max admitted that he never completed his algorithm, the chess player was a little disappointed: “Maybe next time?”.

Learning neural network ended only a week. Max immediately checked the algorithm on the failure of the queen, who cost him the party. "Bad move", - belatedly issued a neural network. If she was ready a little earlier, mistakes could have been avoided.

The annual plan is fulfilled, but Max does not intend to tie up with self-study. In August, he opened an educational startup and is now thinking about repeating the experiment - you just need to invent 12 new tasks. Maybe next year?