A Deep Learning Alternative Can Help AI Agents Gameplay the Real World
1 min read
Summary
UK software firm Verse AI has developed a machine-learning (ML) system that it claims can master simple video games more efficiently using less data than current neural networks.
Known as Axiom, the system differs from current artificial intelligence (AI) technology as it incorporates prior knowledge about physical interactions, subsequently modelling game behaviour based on active inference algorithms.
The approach, inspired by the free energy principle, was developed by neuroscientist Karl Friston and is intended to build AI agents that emulate human cognition.
While the conventional approach to learning to play games is to use deep reinforcement learning, this requires a lot of experimentation to succeed.
Axiom, however, has shown it can excel at various simplified versions of popular games using fewer examples and less computational power.
