It remains a mystery about how consciousness relates to neural mechanisms. Now, with the advent of AI, some argue that consciousness could emerge in these systems. Here, philosopher and Lead for Responsible AI at Microsoft Ken Archer, and phenomenologist Harald Wiltsche, argue that the attempt to reduce consciousness to a mechanism won't work and relies on a misunderstanding. Instead, using Einstein's re-thinking of the ether as a guide, we can unlock the mystery of consciousness.

The strongest argument for believing that AI can become humanly intelligent, or even conscious, seems to come from science. Neuroscience, we are told, is steadily revealing—or will reveal, given enough time—that intelligence and consciousness are nothing more than what the brain’s neural mechanism does. Evolution appears to make the conclusion unavoidable. Matter existed for billions of years before intelligent and conscious beings appeared. Conscious intelligence emerged from that matter through natural processes. So intelligence and consciousness must ultimately be reducible to particular organizations of matter, which neuroscience tells us is a particular organization of neurons.

To resist the reducibility of intelligence and consciousness to some type of organization of matter can sound like resisting science itself. Perhaps consciousness seems irreducible only because we do not yet understand the mechanism. Give neuroscience another century and the mystery will disappear, just as so many earlier mysteries disappeared before it. With the ability to do neuroscience-like experiments on AI neural networks—known as mechanistic interpretability—perhaps the mystery will disappear even sooner.

But there is something peculiar about this debate. Those who are most philosophically reductionist about the mind are often the most optimistic about artificial intelligence. If consciousness is ultimately nothing more than the operation of a physical mechanism, then there seems little reason in principle why the relevant mechanism must be made of neurons rather than silicon.

Yet the reverse position contains an equally revealing irony. Many who resist reducing human consciousness to “merely” neural activity become strikingly reductionist when they turn to AI: whatever impressive behavior a machine exhibits, we are told, it is still “merely computation.” The anti-reductionist about the human mind becomes a reductionist about the artificial one.

SUGGESTED VIEWING Cracking the code for thought With Roger Penrose, Sabrina Gonzalez Pasterski, Max Tegmark, Güneş Taylor

These positions appear to be opposites. In fact, they share a deeper assumption. Both have come, without quite noticing it, to understand reduction as the identification of a phenomenon with a mechanism. They disagree over whether the mechanism is sufficient to produce consciousness, but agree that finding—or failing to find—the right mechanism settles the question.

That is not the kind of reduction that produced some of the deepest advances in science. Scientific reduction has repeatedly succeeded not by discovering that one thing is “merely” another thing—as with, say, germ theory’s identification of specific pathogens as the mechanism of disease—but by discovering a deeper invariant structure through which apparently different phenomena become intelligible as transformations of the same underlying order.

We should reduce consciousness. The problem is that today’s reductionists don’t reduce it far enough.

Thingification