Reinforcement learning might be the best path to advanced AI, because it’s the same algorithm that brains usetechnologyreview.com
- Brains acquire skills through reinforcement learning -- like AI, but with dopamine instead of reward functions.
- New research from DeepMind shows just how closely reinforcement-learning AI and brains are connected.
- They’re attempting to explain the brain’s dopamine system through the lens of a new algorithm, which predicts rewards as a distribution (as opposed to a number).
- After a test with mice, they found evidence that the brain does indeed use distributional reward predictions to improve its learning algorithm.
- The fact that brains use it validates reinforcement learning as a promising path to advanced AI, and opens up new avenues for neuroscientific research.