Sysadmin....................
Teach your computer to play games then sit back and watch it try to beat your high score.
Mr. Brown’s delves into the ‘Datacentre Operating System’ a contender for running containers as well as big data applications.
OpenAI, the Silicon Valley non-profit dedicated to building ‘safe’ AI (and which has backing from several wellknown billionaires) recently released Universe ( https://universe.openai.com), labelled as ‘a software platform for measuring and training an AI’s general intelligence across the world’s supply of games, websites and other applications.’ The software allows an AI agent to use a computer in a similar manner to a human (via keyboard, mouse and screen). Using VNC, Universe presents a program to an agent via a remote desktop. The agents get no special access to any program internals or APIs etc. The project website announced that many well-known games development studios have granted free access to agents using Universe to play their games and it’s hoped that many more will follow suit.
Of course, games are not the only programs that could be accessed via this route, but they are a good model for testing reinforcement learning type agents (they have a defined, measurable reward for example, such as gaining a high score). OpenAI also provide a project called Gym which is a toolkit for these kinds of agents.
The goal of the project is to help develop a single AI agent that can use its experience in Universe-provided environments to quickly master unfamiliar ones. This would be a step towards the goal of a machine demonstrating general intelligence. This type of all round, problem-solving ability remains out of reach for systems at the moment although there are many ongoing efforts in this area. Universe is available to download from GitHub ( https://github.com/openai/universe). The games themselves generally run in Docker containers and Universe is largely written in Python. Around one thousand Flash-based games are available with Universe, as well as several from Atari (from the 2600 era) and a lot of example code for writing agents etc. The OpenAI team would like contributions from the community and having had an initial play around with it, the project is tremendous fun and will hopefully feature in a future episode of the magazine (assuming agents haven’t replaced us all by then).