What Is Computational Creativity and How Do Machines Generate New Ideas?

Computational creativity is the use of machines to generate concepts that are both novel and useful, rather than merely to classify, predict, or optimize within fixed rules. In the approach described by Imagination Engines, idea generation comes from chaotized neural networks that brainstorm continuously, while separate monitoring networks detect and reinforce the outputs that show novelty, utility, or value. This article explains that mechanism, how it differs from artificial consciousness and artificial sentience, and where the practical limits lie.

The core mechanism: chaos-driven brainstorming

Imagination Engines traces its work back more than four decades to artificial neural networks designed to hold "chaos-driven brainstorming sessions." The key design choice is to keep the generating network in a chaotic, restless state instead of letting it settle into a stable pattern.

• Input: an initiating concept or seed, represented as a chain of linked concepts.
• Action: the chaotized network produces a continuous stream of candidate notions and action plans, each built from interrelated memories of a linguistic, visual, or auditory kind.
• Expected result: a large, varied pool of candidate ideas rather than a single answer.

The company describes this ongoing stream of notions as equivalent to the brain's stream of consciousness. The point is volume and variety: a chaotic net does not converge early, so it keeps proposing combinations that a rule-bound system would never enumerate.

How useful ideas are separated from noise

A stream of ideas is only valuable if something filters it. Imagination Engines describes monitoring networks that sense significance in certain generated ideas and selectively reinforce those concepts judged to have novelty, utility, or value. Reinforcement then shapes what the generating network produces next, so the system drifts toward ideas that pass the filter.

A concrete way to picture this, for example when a team needs new product concepts: the generating net emits many loosely related combinations, and the monitoring net keeps the ones that are both unfamiliar and plausibly workable. The filter is the practical heart of the system. Without it, computational creativity produces novelty without usefulness.

How it differs from artificial consciousness and artificial sentience

These terms are often conflated, but the source material draws a clear line.

Imagination Engines itself calls its early work "artificial" or "proto" consciousness and explicitly asks whether it is the kind of consciousness humans experience or only a first-order approximation. The sentience work goes further, modeling how whole neural nets interconnect to produce anticipatory responses to synthetic thought, such as "A, then B and C will happen," and how strongly significant stimuli can freeze cognition into a single feeling. In short: creativity is about producing ideas; consciousness is about a stream of ideas with a felt quality; sentience is about feeling and emotion. A system can be creative without being either of the other two.

Practical limits and how to evaluate output

The source material is a research announcement, not a product specification, so treat the following as evaluation criteria rather than guarantees.

• Novelty is relative. What counts as new depends on the existing pool of ideas; a concept can be novel to the system and stale to the domain.
• Utility needs an external judge. The monitoring net's sense of value is only as good as what it was trained to recognize. Human review remains necessary before acting on output.
• Chaos is not control. A restless generator produces volume, not reliability; expect many weak candidates per strong one.
• Claims of consciousness are contested. The company's own framing invites the question of whether its systems approximate human consciousness; readers should separate the engineering result from the interpretive claim.

To use this approach in practice, define the seed concepts, decide in advance what "novel and useful" means for your domain, and keep a human filter on the reinforced output. That combination is what turns a stream of machine-generated notions into computational creativity you can actually act on.