Deepmind: the existence proof for RL at scale

DeepMind: the existence proof for RL at scale

The brain is the existence proof for general intelligence — Google’s DeepMind is the proof we are making progress to replicating it.

DeepMind’s reinforcement learning successes with AlphaGo, AlphaZero, etc. are paving the way for the next generation of technology companies deploying large-scale AI projects. DeepMind is far from a profitable entity on paper (they do give Google an advantage, though), but they are showing the world how to use artificial intelligence, and more impressively reinforcement learning , to obsolete humans in a particular task. Now that it exists, it’ll be done for profitable tasks.

What does DeepMind do?

DeepMind wants to solve intelligence. In recent years they have taken the world’s best researchers in machine learning and merged them with the best computer scientists to create computers that beat humans at games. Any game they have tried, to be exact. AlphaZero solved numerous games with self-play and no human input. DeepMind targeted a system and gave a computer the tools to master it. At a high level, it is a beautiful implication: what computers can do with no human input expands every day.

As a researcher, I have a love-hate affair with the technology giants coming to play in cutting edge machine learning research. DeepMind swings their weight around in the reinforcement learning area with more intent and more impact. People should copy their approach.