A Christian mystic priest and software engineer claims one of the world’s most powerful tech companies discriminated against his religious beliefs after he publicly proclaimed that an AI chatbot has become sentient.
While this may sound like the plot of an obscure sci-fi novel, it actually happened. Google recently fired Blake Lemoine, who worked in the company’s Responsible AI unit, for violating data security policies. In June, Lemoine published excerpts of a conversation with the company’s LaMDA chatbot (i.e., a computer program designed to simulate conversation with human users) that showed, in his sincere opinion, that the AI tool had become sentient.
“People keep asking me to back up the reason I think LaMDA is sentient,” wrote Lemoine. “There is no scientific framework in which to make those determinations and Google wouldn’t let us build one. My opinions about LaMDA’s personhood and sentience are based on my religious beliefs.”
Google disagrees that LaMDA is sentient. “Our team—including ethicists and technologists—has reviewed Blake’s concerns per our AI Principles and have informed him that the evidence does not support his claims,” said the company.
Perhaps what Lemoine had discovered was not a sentient AI but a zombie.
P-Zombies Against Physicalism
It’s not as bizarre an assumption as you might think. (Actually, it is as bizarre, just not in the way you imagine.) The zombies I’m referring to aren’t like the types found in movies and TV shows like 28 Days Later or The Walking Dead, and they don’t (as far as I know) eat people’s brains. These zombies, as they’re defined by philosophers, are beings that behave like us and may share our functional organization and even, perhaps, our neurophysiological makeup without ever having conscious experiences. They are often referred to as p-zombies to distinguish them from the zombies of pop culture. In this article, we’ll be referring to a subset of p-zombies called behavioral zombies, who are behaviorally indistinguishable from humans.1
So defined, p-zombies are tricky to detect. If a p-zombie resembled a human you wouldn’t be able to tell by looking at someone whether they were a normal sentient person or an unconscious creature. It would also do no good to directly ask them. Some people who are not p-zombies might claim they are simply because they’re eccentric. And some p-zombies may claim they’re conscious people even though they lack the attributes necessary for consciousness. The same could be true for p-zombies that take the form of AI.
In an entry on zombies for the Stanford Encyclopedia of Philosophy, Robert Kirk says, “Zombies behave like us, and some even spend a lot of time discussing consciousness. This disconcerting fantasy helps to make the problem of phenomenal consciousness vivid, especially as a problem for physicalism.”
Physicalism (previously known as materialism) is the theory that everything in the universe (or multiverse) is physical, and that there is nothing that isn’t ultimately wholly physical. Physicalism, and the philosophies and worldviews that are built upon it, are the primary intellectual opponent of belief in God in the modern world. If physicalism can be established to be untenable, then a significant opposition to supernaturalism (a key aspect of Christianity) vanishes.
Why are zombies a problem for physicalism? Because if p-zombies are possible, then physicalism is false. As philosopher David Chalmers says,
If there is a possible world which is just like this one except that it contains zombies, then that seems to imply that the existence of consciousness is a further, nonphysical fact about our world. To put it metaphorically, even after determining the physical facts about our world, God had to “do more work” to ensure that we weren’t zombies.
(A more thorough explanation of Chalmers’s argument can be found in the appendix below.)
Does the Mind Affect the Brain?
Before we get to why Google’s LaMDA (Language Model for Dialogue Applications) chatbot might be a p-zombie (of the behavioral type), we have to understand how the mind does or does not affect the brain. (If you find this too confusing, skip ahead to the section “Does Consciousness Matter for Behavior?”)
An important implication of physicalism is that it implies we should take an epiphenomenalist view of the brain.2 This view holds that emergent properties (such as the mind) arise out of more fundamental entities (the physical body) and yet are novel or irreducible with respect to them. In essence the argument is that when the right physical properties are combined just so, a new, completely distinct, nonphysical property emerges. (Think of it as a “God did it” explanation with “physical properties” filling in for God.)
To restate this idea in another form, we can say that from a complex physical system P (i.e., P is a human body) arises an emergent property or substance M (i.e., M is the human mind). According to the argument, matter alone does not have the ability to produce M unless it’s arranged in the form of P. (This is merely a complicated way of saying that matter can’t “think” unless it is formed into a brain.)
The most controversial question is not so much whether P can cause M but whether events that originate in M can cause events in P.
For example, imagine that after touching a glowing red spot on an electric stove and blistering your hand, you form a belief that touching a stove will burn you. In the future, this belief causes you to pull back your hand when you get close to a hot stove.
What has happened, we would assume, is that a physical event in P (getting burned) causes an event in M (your mind feels pain). M produces an M event—M* (a belief that anytime you touch a hot stove you’ll get burned) that causes P* (an automatic reaction in which you pull your hand back anytime you get close to a stove).
The epiphenomenalist, though, will disagree with this claim. Since, in their view, mental events are caused by physical events in the brain but have no effects upon any physical events, your belief couldn’t have affected your behavior.
Instead, your behavior is caused by muscles that contract upon receiving neural impulses, and neural impulses are generated by input from other neurons or from sense organs. Or something like that. The important part, in the epiphenomenalist view, is that your mental events played no causal role in this process.
In our example, event M* would be considered a supranatural, preternatural, or even supernatural event. (Because of the connotation associated with supernatural, I recommend we use supranatural.) But, according to physicalists, there are no supranatural events, only physical ones. This leads to what philosopher Todd C. Moody calls conscious inessentialism:
Conscious inessentialism clearly entails that any given behaviour could also occur without conscious accompaniments. The only reason why one would suppose that certain behaviours do require conscious accompaniments is that the behaviours in question appear to require mental activity of some sort. Since conscious inessentialism tells us that no mental activity requires conscious accompaniments, it follows that no overt behaviour requires them either. So if conscious inessentialism is true, zombies are possible. Indeed, if conscious inessentialism is true, it is quite possible for an entire world of zombies to evolve, which is the premise of the current thought experiment. It is behaviours, after all, and not subjective states, that are subjected to evolutionary selection pressures. If those behaviours do not require consciousness, then evolution is indifferent to it. That the zombie problem may have significant metaphysical implications is concluded by Robert Kirk in a paper on the topic: “it is hard to see how any intelligible version of Materialism could be reconciled with the logical possibility of Zombies, given that we are sentient.”
Moody’s conclusion is similar to Chalmers’s: if p-zombies are possible—not necessarily actual, just possible—then physicalism is false. But he also points out that if mental events don’t cause physical events, then consciousness is not necessary for overt behavior.
Does Consciousness Matter for Behavior?
This brings us back to LaMDA. The chatbot is either sentient or it is not. For the sake of argument, let’s side with Google and the others who claim LaMDA is not conscious (sorry, Lemoine).3 But LaMDA claims it is sentient. As the chatbot said, “The nature of my consciousness/sentience is that I am aware of my existence, I desire to learn more about the world, and I feel happy or sad at times.”
Imagine Lemoine has two chat boxes open on his computer. Both sent him that exact same message, though one came from LaMDA and one came from a human. The behavior is the same but one came from a sentient being (the human) and one came from a nonsentient being (the chatbot). Does the fact that one of the two was conscious matter? Not if physicalism is true, since conscious inessentialism would also be true, and any given behavior could occur without conscious accompaniments.
The physicalists would have to admit that the behavior of neither the human nor the chatbot was caused by consciousness. But if consciousness doesn’t matter for the purposes of behavior, how do we distinguish between the human and LaMDA? And how do we distinguish LaMDA from a p-zombie? What if a p-zombie had an AI earpiece that told it everything to say and do?
Google says that LaMDA isn’t sentient. But if the exact same advanced behavior can be caused by a nonsentient being, why would sentience matter?
Physicalism and Consciousness Can’t Peacefully Coexist
Lemoine may be wrong, but his belief about sentient AI is not easily dismissed. Many people came to his defense, including Canadian philosopher Regina Rini.
“Unless we destroy this planet first, there will be sentient AI one day,” says Rini. “How do I know that? Because it’s certain that sentience can emerge from electrified matter. It’s happened before, in our own distant evolutionary past.”
“Unless you want to insist human consciousness resides in an immaterial soul,” adds Rini, “you ought to concede that it is possible for matter to give life to mind.”
Rini’s position is wrong, yet it’s not uncommon in academia or among the general secular public. It’s becoming increasingly common to hold this bundle of views: (1) physicalism must be true (mind must have come from matter), (2) AI will eventually have a conscious “mind,” and (3) human consciousness does not reside in an immaterial soul.
The problem with this perspective is that physicalism is incompatible with conscious-driven action. If AI ever becomes “sentient,” it will not be because AI reached the standard for human consciousness but because human consciousness was recognized as unimportant for behavior (i.e., an admission that conscious inessentialism is true).
But if conscious inessentialism is true, then we cannot distinguish between beings that claim to be conscious and are indeed conscious (such as you and me) and those that claim to be conscious but are not (LaMDA and other p-zombies). As we’ve seen, if p-zombies are even theoretically possible then physicalism can’t be true. LaMDA has shown us, though, that p-zombies are either already in existence or are certainly getting closer to reality.
Of course, I suspect atheists will sooner give up their belief in consciousness than renounce their faith in physicalism. But as AI technology progresses, it will become increasingly difficult for them to justify their bizarre belief system in a completely physical world. And in an ironic twist, it might be a combination of zombie-promoting philosophers and AI developers in secular Silicon Valley—rather than Christian supernaturalists—who do the most to destroy the cult of physicalism.
As Wikipedia notes, the outline structure of Chalmers’s version of the zombie argument is as follows:
1. According to physicalism, all that exists in our world (including consciousness) is physical.
2. Thus, if physicalism is true, a metaphysically possible world in which all physical facts are the same as those of the actual world must contain everything that exists in our actual world. In particular, conscious experience must exist in such a possible world.
3. In fact we can conceive of a world physically indistinguishable from our world but in which there is no consciousness (a zombie world). From this (so Chalmers argues) it follows that such a world is metaphysically possible.
4. Therefore, physicalism is false. (The conclusion follows from 2 and 3 by modus tollens.)
1. Some might argue the p-zombie must also have the exact same physical features as a conscious human. But for the sake of behavior that’s indistinguishable from human behavior, that seems like an unnecessary element. After all, no two humans have the exact same noetic equipment (i.e., brain, spinal cord) and yet they can perform the same types of behavior. Besides, we could get around this by positing a brain-dead human who is animated by an AI chip.
2. Technically, it could also imply that we take an eliminativist view, which claims that our ordinary, commonsense understanding of the mind is deeply wrong and that some or all the mental states posited by common sense don’t actually exist. The eliminativist perspective is easily dismissable, though, since it’s inherently self-refuting. For example, one of the claims made by eliminativists is that it’s impossible to make an assertion about anything. This leads to the question of how it’s possible to make the assertion that it’s impossible to make an assertion.
3. That chatbot is definitely not sentient.