DABUS’s output gets progressively more surreal as cognitive flow is reduced. Credit:Imagination Engines
DABUS is a new type of AI that paves the way for smarter machines. Nothing new there – but this machine-learning system is different. DABUS, short for “device for the autonomous bootstrapping of unified Sentience”, is deliberately created to be mentally unstable.
Since the 90s, computer scientist Stephen Thaler has been injecting noise into a special neural network to generate novel ideas. Thaler, CEO of Missouri-based Imagination Engines, calls this module an Imagitron. The stream of creativity is assessed by a second network called a Perceptron, which provides feedback to improve ideas. This AI approach of neural networks playing off each other has been adopted by Facebook, Google and others – under the name “generative adversarial networks” – as a method of creating images that look authentic to human observers.
Artificial Intelligence where neural nets play against each other and improve enough to generate something new. Rob Miles explains GANs.
Now Thaler has introduced a technique to assess ideas according to how they resonate with existing knowledge – the AI equivalent of art or music that triggers happy or unhappy associations. Another process boosts the slow and tentative rhythm characteristic of creative activity in a neural network. The system swings between extremes of unimaginative plodding and novel thinking. It can also exceed the bounds of sanity.