DeepMind has opened free access to a virtual machine learning environment

Recently, representatives of the DeepMind division (now part of the Alphabet holding) reported on providing free access to developers to the source code of the DeepMind Lab platform. This is a Quake III based machine learning service that is designed for artificial intelligence training. Namely - learn how to solve problems in three-dimensional space without human intervention. The platform is based on the Quake III Arena game engine.

Inside the game world, the AI ​​gets the form of a sphere and the ability to fly while studying the surrounding space. The goal that developers have set themselves is to teach the weak form of AI to “understand” what is happening and respond to various situations occurring in the virtual world. "Character" can perform a number of actions, move through the maze, explore the immediate environment.

“We are trying to develop various forms of AI, capable of performing a number of tasks from the usual study of the game world to the accomplishment of any actions with an analysis of their consequences,” said Shane Legge, chief research associate of DeepMind.
')


Experts hope that AI can learn by trial and error. Games in this case - almost perfect. For example, earlier in DeepMind they used (and are using now) the Atari game console in order to teach the neural network to perform the sequential actions necessary for the game.

But the open three-dimensional world that can be changed represents a much more promising environment for learning AI than the flat world of graphically simple toys for Atari. AI in the three-dimensional world has clear tasks that are consistently changed in such a way that the experience gained in solving each previous task turns out to be useful for AI in the course of solving the next one.

The advantage of a three-dimensional environment is that it can be used to train computer systems to respond to various problems that a robot can expect in the real world. With this simulator, industrial robots are trained without problems. And in some cases it is easier to work with a virtual environment than to teach such systems “manually”.

Moreover, most modern neural networks are designed to solve one specific problem (image processing, for example). The developers of the new platform promise that it will help create a universal form of AI capable of solving a large number of tasks. And the help of people in this case, the computer system is not needed. The generation of the environment for the neural network occurs every time in a random order.


According to the developers of the platform, it helps the AI ​​learn in much the same way as children learn. “How did you or I studied the world in childhood,” one of the DeepMind employees cited an example. “The community of machine learning professionals has always been very open. We publish about 100 articles per year, in addition, we have opened the source code for many of our projects. ”

Now Google DeepMind has opened the source code of DeepMind Lab, uploaded it to GitHub . Thanks to this, anyone can download the platform code and modify it to fit their needs. Representatives of the project declare that new experts can create new game levels independently by uploading their own projects to GitHub. It can help the whole community work towards achieving their goals faster and more efficiently.

This project is not the only one for DeepMind. Last month, his representatives signed a cooperation agreement with Activision Blizzard Inc. The goal is to turn Starcraft 2’s environment into a testbed for artificial intelligence. Perhaps in the near future other game developers will join this project. By the way, AI in the gaming environment does not get any advantage over the opponent, using only visual information for promotion, just like a person.

In practice, this means that Google AI will need to anticipate what the enemy is doing at any particular time in order to adequately respond to the actions of the “enemy”. In addition, you will need to quickly respond to what went off plan. All this will allow to test the next level of capabilities of artificial intelligence. “In the end, we want to use these abilities to solve global problems,” said Demis Hassabis, founder of Deepmind (Google bought it in 2014, and now, based on the achievements of the purchased company, AI is being developed).

Specialists associated with AI, expressed cautious approval of the project. “It is very good that they provide a large number of environment types,” said Ilya Sutskevar, co-founder of OpenAI. “The more a system will encounter with more types of environments, the faster it will evolve,” he continued. Indeed, the three-dimensional learning environment of AI contains over 1000 levels and types of environments.

Zoubin Gahrahmani, a professor from Cambridge, believes that DeepMind Lab and other platforms to enhance the development of artificial intelligence contribute to progress, allowing researchers to develop the environment. At the same time, projects like this are quite transparent. He also noted that it takes a lot less time for a person to achieve a certain level of the game than a computer. Therefore, the professor expresses doubt that the AI, its weak form, will be difficult to at least bring to the level of a person in terms of learning speed.