Brain-Like AI Can Now Predict What Happens Next, Using Far Less Power
Brain-inspired AI that predicts what happens next could make robots smarter and cheaper.
A researcher has designed a spiking neural network (SNN) that predicts the sequence of external world states, starting from the current world state. According to the paper's abstract, the SNN doesn't create a world dynamics model; instead it incorporates an SNN trained to predict the next world state and provides all the mechanisms needed to build the chain of predicted world states. Those mechanisms are entirely spiking — implemented as spiking neuron ensembles. The article describes this neuronal structure and tests its operation on a classic reinforcement learning benchmark: ATARI ping-pong.
- The AI predicts a chain of future moments, one after another, instead of just reacting to the present.
- It only 'fires' when needed, like brain cells, which could slash the huge electricity bills of modern AI.
- Testing stopped at Atari ping-pong, so there are no robots, cars, or real products using it yet.
Why It Matters
Could mean AI that plans ahead on low power — cheaper robots and gadgets that don't need giant data centers.