Deepmind just published a mind-blowing new paper about reinforcement learning. One of our leading experts is focused on this topic, and we are using it for intelligent gripper design.
So what is reinforcement learning and why is this paper so important?
Reinforcement learning is a specific branch of machine learning that uses a reward system to teach new (virtual) behaviour. In this paper the reward system is simply the distance it can run. The longer it runs, the more that particular behaviour is promoted.
That sounds like Darwinism, but then in a virtual environment. And because machine learning is basically mathematics, it shows that we can create ‘virtual Darwinistic engines’, and that what we are as humans is just a mathematical consequence of our complex environment. Combine this with genetics and the advances of biomolecular deep learning and we have a new disruption coming up.
Maybe we can get rid of the long time constants associated with Darwinism and create an AI based evolution engine…