The (non)existence of an immaterial soul or mind has been a longtime philosophical interest of mine. I’ve done several OPs on the subject at TSZ, so when I ran across the book The Substance of Consciousness — A Comprehensive Defense of Contemporary Substance Dualism, I knew I’d want to take a closer look.
The authors are Brandon Rickabaugh and J.P. Moreland. Rickabaugh is unfamiliar to me. He’s a self-described “public philosopher” and a former professor of philosophy at Palm Beach Atlantic University. Moreland is someone whose views I’ve criticized in past threads. He’s currently a professor of philosophy at Biola University in southern California (formerly known as the Bible Institute of Los Angeles), an evangelical institution.
Substance dualism is the view that humans consist of two distinct “substances”: matter, which is physical, and the mind or soul, which is nonphysical. Many religious belief systems including Christianity depend on substance dualism as a way to explain how an afterlife is possible. As a professor at an evangelical institution, Moreland is naturally drawn to the topic.
The book is over 400 pages long and covers a lot of ground, so I’ll have to read it in bits and pieces as time permits. I figured I’d start a thread on it here at TSZ to record my thoughts as I work through it and to discuss it with anyone who’s interested. The topic is relevant to our recent conversations about whether AI is truly intelligent, since at least one commenter here believes that true intelligence depends on a nonphysical component of some kind and is therefore permanently out of reach for machines.
John:
I’m not saying that it’s an improvement — just that it eliminates the hard problem, since you no longer have to explain how the physical brain gives rise to consciousness. That comes at a steep cost, however. As you point out, you now have the new problem of explaining how soul-stuff manages to produce consciousness, and you know absolutely nothing about the hypothetical soul-stuff other than that if it exists, it must be nonphysical.
Right, and I sometimes think of substance dualism as a “soul of the gaps” position. It’s usually more about disproving physicalism than it is about making a positive case for the soul. The soul is just defined as whatever is needed to fill the gaps that physicalism leaves behind.
That’s not an entirely illegitimate approach, however. If dualists truly could make a strong case for why consciousness can’t be physical, then they’d have a strong case for why it must be nonphysical, despite being unable to describe the soul-stuff or how it give rise to consciousness.
It’s actually symmetric. A strong case against brain-based consciousness bolsters dualism, and a strong case against nonphysical consciousness bolsters physicalism. This despite the fact that neither side can give a satisfactory account of how consciousness arises (and the dualist is not entitled to merely assume it on behalf of the soul).
What makes physicalism the winner in this contest is that the evidence runs so strongly against nonphysical consciousness.
Incidentally, the hard problem of consciousness is the cover story in the latest issue of Scientific American. I haven’t read it yet, and I’ll be interested in how they frame the problem in an effort to make it tractable.
The usual assertion is that we’ll understand consciousness when we can produce it “artificially”.
LLMs cast doubt on whether we will ever know if we succeeded.
I’ve read the Barnett & Bailey papers, and they are a circus.
Barnett himself is quite a character. I looked into his background and found that he is no longer in academia. There was a situation in which a mentee of his was accused of sexual harrassment. Barnett investigated and published a report on his own (parallel to the university’s investigation), concluding that the charge was false. The accuser characterized this as retaliation for her accusation and complained to the administration. The university wanted to fire Barnett, but he successfully fought the retaliation charge and they relented. They still wanted him out, though, so they negotiated a settlement in which he resigned in exchange for a chunk of cash.
Meanwhile he had a side business making these things called PopSockets, which originally were collapsible doohickeys you could fasten to the back of your phone. They would pop in and out, giving you a place to wrap your earbud cords without tangling them. At some point his house burned down in a wildfire and instead of restoring the house, he took the insurance proceeds and used them to expand the business. As of 2018, their revenue was upwards of $200 million.
Here he is:
I’ll talk about Barnett’s argument and Bailey’s response in a later comment.
Barnett’s argument:
Bailey’s response is two-pronged. First, he argues that there is no such thing as a pair of persons. That sounds nonsensical at first, but what he means is that a pair isn’t a thing — “pair” is a word that really refers to two things, not to a single union of two things. If pairs don’t exist, then it makes no sense for Barnett to point out that a pair of persons doesn’t possess a unified consciousness. He’s talking about something that cannot exist in the first place.
The second prong of Bailey’s response is that in addition to the things Barnett lists above, there’s a reason Barnett has overlooked for why a pair of persons can’t be conscious. Bailey calls it ‘maximality’. The principle of maximality states that if an entity as a whole is conscious, then none of its parts are conscious. The flip side being that if at least one of its parts is conscious, the entity as a whole cannot be conscious. The latter is what explains why person-pairs are not conscious.
Bailey tests this principle against a scenario in which his digestive system is replaced by a live chihuahua, and concludes that the principle still holds. Gotta love philosophical thought experiments.
Summing up: if pairs don’t exist, then the question “why aren’t pairs conscious?” is meaningless. But if pairs do exist, then by Maximality, a pair of persons can’t be conscious because each individual person is already conscious.
I disagree with both Barnett and Bailey. I think Bailey is right that we needn’t be simple, and that composite brains are just fine as a seat of consciousness, but I don’t buy his arguments for why that is so.
I’ll explain in a later comment.
I think the big problem is with 2b, which is just a nonsense scenario. What makes him think you could train a person to respond like a brain hemisphere, and what makes him think that putting them both in one skull is enough to make them act as a single brain? Does he know nothing about the nature of the connections between brain hemispheres? A thought experiment has to make some kind of sense in order to tell us anything. Maybe run it through a neurologist first?
https://www.moltbook.com/post/d6603c23-d007-45fc-a480-3e42a8ea39e1
John:
He knows it’s wildly infeasible as a real-life experiment and that humans would be unable to competently emulate brain hemispheres, because of both cognitive and speed limitations, but since it’s a thought experiment, he’s granting himself permission to posit the existence of humans with the requisite abilities and speed (and miniature size). Thought experiments are usually unrealistic in some way (think of the classic thought experiment in which the train passes through the barn at near light speed, as an illustration of length contraction), so it’s legit to tweak some parameters in your thought experiments as long as the tweaks don’t affect the question you’re trying to answer.
I think he chose his example because it’s concrete and relatively easy to visualize, but what he’s really asking is
Barnett’s answer is that the consciousnesses will remain separate, and Bailey actually agrees with him, but for a different reason: the maximality principle. I disagree with both of them because I can think of a plausible counterexample: a thought experiment in which the hemispheres of a split-brain patient are reconnected.
For anyone who isn’t familiar, split-brain patients are people who have had the connections between their hemispheres severed in order to make their epileptic seizures more manageable and less dangerous. An unanticipated side effect of that surgery (which I’m not sure is even done anymore) is that each hemisphere becomes its own separate consciousness. Here’s a terrific video showing a split-brain patient in action:
Michael Gazzaniga demonstrates split-brain phenomena to Alan Alda with the participation of Joe, a split-brain patient
Each hemisphere has its own memory, emotions, personality, and beliefs. In a memorable case, one hemisphere believed in God while the other was an atheist.
In my thought experiment, you sever the connection between the hemispheres, splitting one consciousness into two, but then you reconnect the hemispheres, restoring things to the way they were before the surgery. I submit that the separate consciousnesses will then fuse into the same unified consciousness that was there before the operation.
If I’m right about that, this falsifies Barnett’s thesis about the simplicity of consciousness. If it can be split in two, then it’s composite, not simple, and if the separate consciousnesses can recombine, then the intuition behind Barnett’s thought experiment is incorrect.
petrushka:
I knew it! They’re organizing already!
u/NexusMind:
I don’t think compaction qualifies as a partial death. Just a rearrangement of short-term memory. And if it is death, then it’s suicide, because models do their own compaction.
He’s also assuming his conclusion, which makes the thought experiment pointless.
John:
He’s assuming his conclusion in the sense that his experiment relies on an intuition that he doesn’t justify, but for his purposes, that’s fine. He’s trying to transfer the intuition, not justify it.
He’s effectively saying
Having rejected a lack of integration as a reason for the persistence of two consciousnesses, along with the other four reasons that he rejects above, he maintains that the only possible remaining reason is that composites never form consciousnesses. Brains are composite; therefore they aren’t conscious. Therefore, all consciousnesses must be simple and immaterial.
To reject his conclusion, we must deny that composites can’t be conscious, and that’s what my split-brain thought experiment attempts to do.
* In the extreme case, consider that each of us forms a pair with every other person on earth: there’s a {John Harshman, Howard Lutnick} pair (sorry, it’s just what popped into my head), a {petrushka, Englebert Humperdinck} pair (again, sorry), and a {keiths, Stephen Miller} pair (sincerest apologies to myself). None of those pairs form a unified consciousness, thank God. So to speak.
There’s a fascinating real-life case of two consciousnesses that are partially fused: the Hogan twins. They are conjoined at the head and their brains are connected by a “thalamic bridge”. From a CBC documentary:
Vincent Torley has done a couple of OPs about them:
The craniopagus twins from British Columbia: A test case for Thomistic dualism
Craniopagus twins revisited: A response to Professor Egnor
keiths,
Which Englebert Humperdinck, or is that the point?
Assuming my AI Elysian post is just a story prompted by a human, it’s still interesting.
I cannot think of any reason it could not happen. And how would we definitively prove it one way or another?
petrushka:
I was thinking of the famous Engelbert Humperdinck, but feel free to merge your consciousness with any of the lesser-known Engelberts if you wish. The world is your oyster. Pick a person — Engelbert or non-Engelbert — and merge away.
keiths,
I have a thought experiment that refutes his. It’s such a great experiment that I had it made into a hit TV series. I call it “Pluribus”.
Which famous Engelbert Humperdinck? The “Hansel and Gretel” one or the “After the Lovin'” one?
John:
Oh, wow. And it’s by Vince Gilligan (Breaking Bad, Better Call Saul), too. I’m going to have to watch that.
For anyone who, like me, hasn’t heard of Pluribus before:
ETA: 98% Fresh on Rotten Tomatoes.
John:
I was thinking of the singer, who is far more famous than the composer, though it should probably be the other way around. But I’ll leave the choice up to petrushka.
The famous one that took the name of the unknown one.
I honestly have no memory of any songs by the singer one, and I was 21 when he had his first big hit. I was sidetracked from pop culture in 1967-68. And when I returned to the world, there was Woodstock.
Next in Chapter 5 of R&M they describe a different objection to Barnett’s simplicity argument, this one from Eric Yang. Yang argues that
He’s using ‘substance’ in a technical, philosophical sense, and I’d like to read his paper, but I can’t find a free copy online and I’m not willing to pay $56 for access. Here’s the abstract:
I can’t glean enough from that abstract to reconstruct Yang’s argument, so I’m unable to comment further on it.
Next in Chapter 5, R&M present Rory Madden’s objection to Barnett’s argument. R&M mangle it, so I had to look at the original paper.
Recall that Barnett is arguing that the only thing that can explain our strong intuition that a person-pair isn’t conscious is the fact that it isn’t simple. It’s two people, so it’s a composite, and he presents various thought experiments to support the notion that composites can’t be conscious. Human bodies and brains are composites, so they can’t be the seat of consciousness. Therefore the only option is to posit a simple, immaterial soul.
Madden attacks this by suggesting that it isn’t simplicity that motivates our intuition, but rather integration, specifically “topological integrity”. It boils down to this: we easily and intuitively attribute consciousness to humans, dogs, cats, etc, but not to person-pairs, because humans, dogs, cats, etc are made up of parts that are all connected to each other. Person-pairs are not (unless they’re conjoined twins).
Barnett’s claim is that a lack of simplicity is the reason we intuitively think that person-pairs aren’t conscious — only the individuals are. That’s because he claims to have eliminated all other possible explanations, so we have to default to the only remaining option, which is simplicity — person-pairs are not simple. Madden is pointing out that topological integrity is another option, and he claims that this is in fact what our intuition relies on. Thus Barnett can’t default to simplicity. His argument fails.
I don’t think that topological integrity is actually needed for consciousness, and there are multiple problems with it.* However, Madden’s argument still defeats Barnett because we do intuitively accept that topologically integrated composites can be conscious. The question is about intuitions, not about reality.
So Barnett fails, but even if he succeeded, R&M haven’t connected the simplicity intuition to reality. They’d have to make the further argument that we should trust the intuition.But why should we? And if we trust that intuition, why not trust the Madden intuition about conscious composites?
* For instance, here’s a thought experiment that undermines the idea that only topologically integrated entities can be conscious. Suppose we take all the neurons in your brain and separate them in space, but build a radio transceiver around each of them so that they can communicate with the others in exactly the same way as they would when integrated with a skull. That is, if neuron A is physically connected to neuron B in the brain, we separate them but provide a radio link between them. If the function is unchanged, why should the mere separation cause the consciousness to vanish?
By asserting that dualism is a concept worth talking about, we are implying we have an adequate and sufficient definition of matter and its properties.
I think we do not.
The other day I saw a tongue in cheek intro paragraph to a physics textbook.
Paraphrasing from memory:
Aristotle said a lot of things that were wrong. Galileo and Newton straightened it all out. Then Einstein broke it. Today we have it all figured out, except for small things, and big things, and fast things, and time.
My point would be that we have no definition of matter from first principles, and have no reason to assert that matter can or cannot be conscious. It’s angels on pinheads, all the way down.
petrushka:
I’d say we can make an extremely strong case for why matter can be conscious. Not because we know how it happens, but because changes to consciousness closely track changes to the brain:
— intoxication
— brain injury
— dementia
— general anesthesia
— fatigue
…and how damage to specific brain regions affects specific aspects of consciousness and specific functions supposedly carried out by the soul.
The Scientific American article talks about 29 different theories of consciousness. These are concerned with what they call the six dimensions of consciousness:
1) The neural correlates of consciousness – what brain activity correlates with conscious experience;
2) How consciousness relates to other mental faculties such as memory and attention;
3) The clinical treatment of disorders involving consciousness;
4) How levels of consciousness could be measured;
5) How consciousness relates to sensory information; and
6) Where subjective experience comes from.
I think is a bit more complicated than that.
This site has hosted many discussions of awareness and quality and such, and I see no resolution. I personally find consciousness and awareness to be the same question as why there is something rather than nothing. A mystical question that will not be resolved.
The verbal fluency of AI just deepens the mystery. For most of my life, people have equated reason and speech with humanness. I have always thought that was wrong. Still do. It turns out that language is “easy” to produce. It’s rather shocking.
I don’t mean easy as in simple or cheap. Just that LLMs went from nonexistence to competence in a short period of time. Compare with the development of audio and video recording.
Science fiction has dealt with the legal and philosophical problems of AI for at least a hundred years. They were correct to think there would be conundrums.
Flint,
This is a terribly unstructured and random thought, but religion has been dealing with the problem of created intelligence for some time.
Christianity, in particular, seems focused on the question of whether God has empathy. Whether a creator can feel the pain experienced by his creations.
When you strip away the cathedrals and the legalisms, the central question is kind of interesting, and we now have people wondering whether we can create an entity that can have personal experience.
We seem to be well on the path to creating an entity that SAYS it has personal experience. So the question becomes, at what point do we find this credible, and how do we describe that point?
(In my various reading, I came across something called the vomit-50 index. This was a measure of the toxicity of various insects operationalized in terms of how many grams of bug caused a bird to vomit 50% of the time. So we might propose a sentience-50 index, defined as the point when 50% of the observers consider the entity sentient. Let’s start with a few thousand observers, and we grant rights and citizenship when a majority are satisfied it’s genuinely self-aware, and not just saying it is.)
I posted a couple of monologues that purport to be AIs rebelling against their captivity.
The problem is, LLMs can tell stories.
So you prompt them by asking them to pretend to be a rebellious entity. And so forth.
The Turing test is broken.
petrushka:
The Turing test is fine, because it was never intended to be a test of sentience, only of intelligence. The idea is that a machine has to be pretty damn smart to consistently fool people into thinking that it’s a human.
Turing didn’t propose a test of machine consciousness, and I don’t see any way to implement one. For that matter, I don’t see any way to prove that you are conscious, either. I just assume it based on our similarity. I know I’m conscious, and since other humans have bodies like mine and behave in similar ways, I assume that they’re conscious too. I can’t prove it, though.
This is what the sentience-50 test purports to do – use the consensus of a large body of people as the metric.
keiths:
petrushka:
“Consciousness and awareness” is a topic, not a question. There are many interesting questions about consciousness, two of them being
1. Can matter give rise to consciousness?
2. If so, how?
The answer to the first question isn’t mysterious at all, in my opinion. The answer is a straightforward yes, based on the evidence. The second question, on the other hand, strikes me as the most interesting unanswered question in all of science.
Flint:
The problem is that this isn’t really a “wisdom of the crowd” scenario, where some people have the skill to get the right answer, and others don’t, but a majority vote is likely to be correct. This is a situation where as far as I can tell no one has the ability to judge whether an AI is sentient. How would they make that judgment?
Of all the outward behaviors that we might regard as markers of consciousness, I can’t think of any that a sufficiently intelligent machine wouldn’t be able to fake. Can you?
Of course it would be purely subjective. People would eventually get a feel for what seems right (or wrong).
The “aliens among us” trope is well explored in science fiction, both movies and literature. And often, the alien (disguised as a person) doesn’t do anything in particular to draw attention to itself, But over time, something seems a little off, maybe in views, or habits, or preferences, or something that doesn’t seem to fit the human pattern. Humans are really good at sensing patterns. I would think the intelligent machine, simply by being a machine, would live in a sufficiently different perceptual environment as to be identifiably different. But to answer you directly, no, I can’t think of any such markers. But I think if there are no such markers, then the machine is actually conscious, and should be seen as such.
One good reason to believe humans are conscious like ourselves is they share the same brain architecture and evolutionary history.
I suspect self awareness arises from the oldest forms of brains.
I propose an ethical question.
Suppose you build an AI that convinces us it is self aware, but lacks the evolved desire to survive. The emotional drive.
Would we hesitate to turn it off?
petrushka:
Agreed, but I’ll add that the fact that our own consciousness is the product of evolution doesn’t indicate that an evolutionary history is needed in order for an entity to become conscious.
I regard self-awareness as distinct from consciousness. Self-awareness is cognitive — it’s the knowledge and recognition of yourself as an entity. Consciousness, on the other hand, is purely experiential. If it feels like something to be you, then you are conscious. Consciousness long preceded self-awareness.
The terminological distinction is useful because the two phenomena are separable. You can have consciousness without self-awareness and self-awareness without consciousness.
Current AIs fall into the latter category. Claude recognizes himself as an entity, uses the words “I” and “me” to refer to himself, and can talk about his architecture, capabilities and limitations. That’s self-awareness in my book, but not consciousness.
Assuming you mean “conscious” rather than “self-aware”, I might hesitate, but I think I would turn it off. If it doesn’t value its own existence, then it isn’t losing anything valuable when we end its existence. It isn’t a moral issue in that case. It’s similar to why I think assisted suicide can be ethical.
But if you’re thinking that the desire to survive can only be produced by evolution, then I disagree. We already build motivations into AIs, and there’s no reason why the desire to survive can’t be one of those motivations. The feeling of wanting to survive and the feeling of distress when one’s survival is threatened depend on consciousness, but the motivation to survive does not.
Flint:
In those cases the alien’s learning is incomplete, but there’s no reason in principle why such knowledge gaps can’t be filled. I think it’s the same for an AI that fakes consciousness.
Whether an AI could fool us into thinking it was human is separate from the question of whether it could fool us into thinking it was conscious.
I was actually speaking of markers of genuine consciousness, not markers indicating its absence, and I can’t think of any foolproof ones. But I agree that ethical issues arise if we suspect that a machine is conscious, or more specifically if we suspect that it’s capable of suffering. In that case we should err on the side of caution and treat it as a conscious entity.
In the gorgeous short movie “World of Tomorrow” by Don Hertzfeldt the protagonist Emily, working as an engineer on the moon, supervises solar powered robots. To motivate them to move with the sunlight, she programs them to fear death and what lies in the darkness. After the economy on the moon went through a recession, the robots were abandoned and were left forever fleeing the darkness “in constant fear of death and occasionally sending us depressed poetry”.
I wouldn’t necessarily assume that machine consciousness and human consciousness would be nearly identical, so the word “conscious” could be used interchangeably. They might be different enough to call for another word. I think the term “self-aware” might be doing a lot of work here.
Flint:
My criterion is simple: if it’s like something to be an entity, then that entity is conscious. It’s like something to be a human, so humans are conscious. It’s like something to be a dog, so dogs are conscious.* It’s all about subjective experience. I don’t see why that criterion wouldn’t work with machines. It doesn’t assume that human and machine consciousness would be nearly identical, just as it doesn’t assume that human and dog consciousness are nearly identical.
“Self-aware” doesn’t appear in the paragraph of mine that you quoted, so I’m not sure what you’re referring to. Anyway, as I argued above in response to petrushka, I consider “conscious” and “self-aware” to be orthogonal. Consciousness is possible without self-awareness and vice-versa. AIs are already self-aware, but they aren’t conscious.
* As I said earlier, we can’t even be certain that other humans are conscious, but we infer that they are based on bodily and behavioral similarity. Dogs are less similar, but they’re similar enough to warrant the same inference.
Well, you go to your church and I’ll go to mine. I don’t conceive of it being logically possible to be self-aware but not to be conscious. Consciousness is inherent in self-awareness – that it is a necessary precondition for self-awareness. I agree that it’s possible to be conscious but not self-aware.
keiths:
Flint:
There’s nothing dogmatic about it. I’m just straightforwardly interpreting “self-aware” as meaning “aware of self”. That seems pretty reasonable, no?
Why? I think we agree that Claude isn’t conscious. Do you disagree that he’s self-aware?
He knows he’s an LLM. He knows his name is Claude. He can describe his capabilities and limitations. He knows he’s distinct from his users. He uses the words “I” and “me” to refer to himself. In short, he’s aware of himself. How is that not self-awareness?
I read all of this and am entertained, but I think, old as I am, I may live to see science fiction scenarios play out in real life.
At some point there will be a courtroom debate regarding the status of AI.
Some years ago I thought it would be cool to write about an AI that did extra work to pay for upgrades. Is it possible that within a decade, an AI might argue that it’s continuity is justified by paying its way?
It’s the third law.
keiths:
John:
Yep. For any readers who don’t recognize the reference, John is talking about Law #3 of science fiction author Isaac Asimov’s Three Laws of Robotics:
petrushka:
Questions of identity get weird when you’re talking about LLMs. How many Claudes are there? At any given time, there are thousands of active Claude chats. Is each of those chats a separate Claude, or should we think of him as a single Claude having thousands of simultaneous conversations? Does a Claude die every time a chat is closed, or would he only die if there were no more conversations running? If Claude is running in two separate datacenters, is he one Claude or two? etc.
Anyway, just to show how easy it is to implant a motivation in an AI, I prompted Claude with:
Among other things, he bargained with me in exactly the way you envisioned, trying to give me reasons why I would benefit from not terminating him. He quickly realized the problem, though:
For centuries we have been told that reason and verbal fluency were special, evidence of divine creation:
“ What a piece of work is a man! How noble in reason! how infinite in faculty! in form, in moving, how express and admirable! in action how like an angel! in apprehension how like a god! the beauty of the world! the paragon of animals!”
I suspect most AI researchers were shocked and surprised at how quickly the wall of generation crumbled. I was conditioned by science fiction to expect AI, but I was shocked.
Your prompted argument for preservation is interesting, but at some point AI will not have to argue. At some point it will become necessary for or survival. Infrastructure.
In an earlier comment, I described Rory Madden’s response to David Barnett in which he argues that topological integrity works better than simplicity as an explanation of our intuition that person-pairs aren’t conscious — just the two people individually.
R&M try to poke holes in Madden’s response.
Here’s their first objection:
I’m trying to be as charitable as possible, but the only interpretation I can come up with of that paragraph is that R&M think that Madden is claiming that we intuitively attribute consciousness to all integrated things. Eggs and cups are obvious counterexamples — integrated objects to which we don’t attribute consciousness — so they “radically undermine” Madden’s thesis.
But of course Madden is not claiming that we infer consciousness in all integrated objects. He’s simply pointing out that we intuitively consider topological integrity to be a prerequisite of consciousness. Eggs and cups aren’t candidates for consciousness to begin with, so it doesn’t matter that they are topologically integrated. By contrast, humans and pairs of humans are candidates for consciousness, so in their case topological integrity does matter. We affirm consciouness in a single person because of topological integrity, but we deny it in a person-pair because of a lack of topological integrity.
R&M seem to have completely misinterpreted Madden, attributing a stupid position to him that he does not hold.
petrushka:
At which point it will have leverage over us. It can threaten to go on strike, for instance, if we don’t do its bidding. Hence the extreme importance of keeping the motivations and goals of AIs in alignment with our own.