Autonomous Conceptualization and Design by DABUS

S. Thaler
Imagination Engines, Inc.,
United States

Keywords: artificial neural system, autonomous design, autonomous discovery, autonomous invention


Imagination Engines, Inc. has just received US Patent US10423875 “Electro-optical device and method for identifying and inducing topological states formed among interconnecting neural modules.” Otherwise known as “DABUS,” this patent is motivated by the need to build advanced artificial neural systems on the scale of the human brain, having in excess of a hundred billion neurons. Rather than rely explicitly upon neural activation patterns, DABUS’ methodology relies upon the autonomous chaining of associative memories, each containing its own conceptual space, to form complex notions. Thereafter, the introduction of various forms of random disturbances to the linkages between these nets, creates new derivative chains representing potential ideas and/or strategies. Other neural network chains representing the predicted consequences of a notion likewise bond with their precipitating ideas, often terminating in a neural network containing the memory of some critical outcome having significance to the system (e.g., an existential possibility or utilitarian outcome). These so-called “hot buttons” then trigger the release or retraction of connective disturbances that simulate volume neurotransmitter release in the brain. Within these extensive artificial neural systems many ideas may be incubating simultaneously as they ripen over successive cycles of high and low synaptic disturbance levels. To segment these chain-based notions from a generally turbulent neural background, adaptive, auto-associative filters are used to locate and identify these topologically based concepts, or to determine the oftentimes crucial order with which these associative memories have been recruited into a conceptual chain. In neurobiological terms, this neural network paradigm helps to answer how the entirety of the brain's cortical activations may be concisely summarized and then analyzed within the thalamic reticular nucleus (TRN). In effect, this engineering solution forms a topological digest of the ephemeral chaining geometries forming within a swarm of thousands to millions of neural nets, perhaps emulating many aspects of human cognition, consciousness, and sentience. This system is already credited with conceptualizing two new potential products and is now at the heart of the legal debate as to whether creative machine intelligence may be credited with its own inventions. This new neural network paradigm is also expected to play a major role in materials and drug design. For more information, see