Consciousness Creep

Most people assume that if consciousness occurs in a human invention, it will announce itself. A more sinister possibility is that some machines are already self-aware, but no one thinks to test for it, or even knows how. Machine consciousness may not look like human consciousness. In particular, it may not choose to show itself, so society must develop ways of checking. Artificial intelligence experts are creating machines with neural networks like a human brain and with deep-learning systems that can develop abilities through experience. It is important, ethically, to avoid taking advantage of sentient beings, even those that humans create.

“Could our machines have become self-aware without our even knowing it? The huge obstacle to addressing such questions is that no one is really sure what consciousness is, let alone whether we’d know it if we saw it.”

One testing option is ConsScale, which uses a checklist to capture physical features, mental and emotional attributes such as self-recognition, and the ability to lie. It then rates consciousness on a scale from one to twelve, reflecting biological and consciousness states from “dead to superhuman.” However, the ConsScale makes the unpopular assumption that everything exists on a continuum of consciousness. The Integrated Information Theory (IIT) may be a better alternative. According to IIT, any system could be conscious, as long as it has some kind of neural network. Its algorithm quantifies the degree of “network interconnectedness”, or Φ, as the degree to which the machine shares information across its network. Consciousness is possible in any system scoring above zero. However, Φ is currently difficult and time-consuming to calculate, so people generally consider the quality of interconnectedness of a system rather than quantifying Φ. Even so, IIT’s inventors have succeeded in detecting the difference between the brainwaves of a conscious human and one who is deeply asleep or under a general anesthetic.

“For want of recognizing what we have brought into the world, we could be guilty of…the creation of sentient beings for virtual enslavement.”

Tests also need to be able to detect a “zombie” – that is, an unconscious system that can mimic conscious behavior. Self-aware systems demonstrate more complex behaviors because they respond to environmental feedback. Programmers need to equip unconscious systems for all possible outcomes if they are to compete, which they can’t do if resources are limited. So it’s possible to argue that a robot that shows human abilities is conscious when subject to the same physical constraints as a human.