Stibel proposes six predictions so his theory can be tested. Keeping sensory input constant while changing other variables, such as delay or conflict, will reveal whether competing demands spark more conscious engagement. Deliberately inducing failures of comparability, for example through dual-task loads, tests how conscious engagement increases or decreases relative to performance. If an AI agent hasn’t already been trained to do this, it’s unlikely to succeed on its own. Checking whether attention, memory encoding, and motor performance shift together or separately makes it possible to distinguish dynamic consciousness from mere parallel machinery. This segues into the next test, which involves disrupting the coordination of functions like memory and attention to see if consciousness was doing the work of holding them together. There should also be evidence of self-awareness, such as in some cephalopods, which may not have a centralized nervous system, but whose flexible behaviors suggest they’re aware of themselves. The sixth and final test asks whether AI can develop consciousness under pressure or whether it instead responds based on how it was programmed.

“Consciousness should be possible,” Stibel theorized, “But only in agents that must maintain themselves and the conditions their operations depend on, operate under real resource budgets, use a functional self-model, and act in closed loops where consequences change future viability.”

Boris Johnson can sleep easy. It doesn’t look like his sci-fi nightmare visions are about to pass through a portal and enter the realm of reality anytime soon.