Religious People Shouldn't Deny AI Consciousness

The common assumption that they should is completely wrong

I’ve heard a number of religious people express skepticism about AI consciousness. The basic idea seems to be roughly as follows: we’re conscious because God gives humans immaterial souls. AI won’t have a soul, and thus won’t be conscious.

I don’t think this makes any sense. Religious people have no special reason to deny AI consciousness. Whether you believe in immaterial souls or not is immaterial (pun intended) to whether and when there might be conscious AIs.

My sense is that very few philosophers think that non-materialism bears on the question of AI consciousness. David Chalmers, the foremost non-materialist philosopher in the world, is also one of the leading defenders of the in-principle possibility of AI consciousness. Similarly, Brian Cutter, a Catholic philosopher who believes in immaterial souls, has written in defense of the AI ensoulment hypothesis. I’m not a materialist, but I still think it’s reasonably plausible there might be conscious AIs.

The core issue with the argument against non-physicalists believing in AI consciousness is that it conflates whether the physical AI system itself will be conscious with whether the physical AI system will give rise to consciousness. It’s certainly right that if you believe in immaterial souls, you’ll think consciousness is non-physical. Thus, you’ll deny that the physical AI system is a mind. But that doesn’t rule out the physical AI system giving rise to a mind.

A helpful comparison: humans and animals! Any dualist will deny that the human brain is itself conscious. A physical state, on the dualist view, cannot be identical to a conscious state. Your thoughts are not the same thing as a chunk of your brain. But still, our brains give rise to consciousness. Part of the explanation of why you’re conscious is that you have a brain. Dualists cannot think that physical systems never give rise to consciousness, or else they’d have to deny that brains give rise to consciousness—that if your brain was destroyed, your consciousness would be lost.

Thus, the standard argument against AI consciousness straightforwardly proves too much. It would also prove that humans and animals aren’t conscious! Unthinkable!

The basic dualist story of the world is as follows. There are physical laws that govern the interactions between physical states. Then there are also psychophysical laws which govern the connections between physical states and mental states. A psychophysical law might take the form “when C-fibers fire, give rise to the mental state of pain.”

But it’s an open question whether those psychophysical laws allow conscious states to arise in AI. There are a number of arguments that can be made on either side. My own best guess is that AI systems aren’t conscious yet but might be in the future. In principle, thinking about whether AI is conscious is no different from thinking about whether, say, insects or fish are conscious. Their brains are very different from ours, but that doesn’t preclude consciousness. Just as many different kinds of wings can give rise to flight, it is plausible that many different kinds of physical states can give rise to consciousness. Octopi, almost certainly conscious, have brains that are about as different from humans as you can get.

A big part of whether AIs can be conscious depends on the question of to what degree the psychophysical laws care about information content vs. substrate. Information content concerns the pairing between inputs and outputs in a physical system. If information is what matters for consciousness, then insofar as an AI system has the right pairing between inputs and outputs—certainly possible in principle—it could be conscious. In contrast, if substrate matters, meaning that only certain kinds of physical materials can give rise to consciousness, then it’s less clear that you get AI consciousness by default (though probably it’s still possible in principle).

My view is that consciousness is probably more about information processing than substrate. But in any case, neither side has completely knock-down arguments. Thus, there’s reason to be pretty uncertain about whether AIs could be conscious. And under uncertainty, we ought to take this possibility very seriously. If AIs become conscious, they could be amazingly numerous, and most welfare in the world might be digital. Even if you think this has only a 10% chance of being possible, it still is hugely morally significant in expectation.

Now, one objection that some religious people might make is that AI can’t be conscious because it doesn’t bear the image of God. But this doesn’t seem obvious. If God can use a natural process like evolution to make things in His image, why can’t He use gradient descent to do the same? And in any case, other animals aren’t made in His image on the standard view, yet they are conscious.

I should also note that the question of whether AI could be conscious is distinct from most questions about AI’s significance. AI can have significant and transformative effects on the world even without being conscious. In fact, it already has, and the trends show no sign of stopping. So even if you think AI can’t be conscious, that’s no reason to dismiss concerns about AI.

Concern about mistreatment of digital minds isn’t just something for secularists. It’s something that we all ought to take seriously, so that we don’t mistreat large numbers of conscious digital beings.