Welfare Biology and AI: The Psychopath, the Nematode, and the Arahant
Pain without a sufferer: a speculative model of invertebrate experience, drawn from no-self psychopathy and Buddhist phenomenology, and what it predicts about welfare ranges.
This is Part 4 of a five-part sequence on welfare ecology. Part 1 introduces the ethical premises. Part 2 covers the empirical landscape. Part 3 covers interventions. Part 5 covers AI.
A key uncertainty so far has been that of the welfare range of nematodes. More is known about the welfare ranges of some arthropods. But what does that mean – how do we empathize with bugs? If nematodes with 300 neurons can suffer, is that suffering intense or barely a flicker? If mites with 2,750 neurons experience something, how does it compare to what a human or a dog experiences? Is it like our pain minus most of the intensity, or is it something qualitatively different – a kind of experience that doesn’t map neatly onto the human pain spectrum at all?
I want to propose what I’ll call the no-self model – speculative but grounded in phenomenological reports from humans with psychopathy and advanced Buddhist meditators – that may help us think about this question.
The Reinforcement Learning Argument
Let me start with the case for invertebrate pain that doesn’t rely on any phenomenological model.
Classical conditioning in nematodes. C. elegans, the model nematode with some 300 neurons, demonstrates classical conditioning: It can learn to associate a previously neutral stimulus (e.g., a particular chemical) with a noxious one (e.g., high salt concentration), and subsequently avoid the neutral stimulus. This is the same learning mechanism Pavlov demonstrated in dogs.
Classical conditioning requires, at minimum, that the organism has:
A state that is worsened by noxious stimuli (something functionally analogous to “bad”).
A mechanism for associating neutral cues with that bad state.
A motivational system that drives avoidance of the associated cue.
That is, nematodes are doing reinforcement learning, and reinforcement learning requires a reward signal. The question is whether the negative reward signal constitutes suffering.
Sensitization. Nematodes also show sensitization: Repeated exposure to noxious stimuli makes them more responsive, not less. This is the opposite of habituation and suggests something analogous to hyperalgesia in humans – the nociceptive system amplifying its own signal. If the signal were merely mechanical, we might expect habituation (wear and tear on sensors); sensitization implies that the system is increasing its attention to the aversive stimulus. That’s a functional signature of pain, not mere damage detection.
The neuron count problem. But 300 neurons is not a lot. The human brain has ~86 billion neurons, most of which are not directly involved in pain processing, but the pain matrix (anterior cingulate cortex, insula, somatosensory cortex, prefrontal cortex) involves millions of neurons. The idea that 300 neurons could produce anything comparable seems implausible on its face.
This is where the psychopathy connection comes in.
No-Self Psychopathy: Pain Without a Sufferer
M.E. Thomas, a diagnosed psychopath and author of Confessions of a Sociopath, identifies with the description of her psychopathy as her having a “small self” – so small as to be virtually absent. As I discussed in “Is Enlightenment Controlled Psychosis?”, this no-self psychopathy has a distinctive phenomenology:
Insults don’t land, because there is no one there to be insulted. Punishments don’t land, because they are threats to who exactly? There’s no one there. Pain and all the emotions that build upon the nociceptive system signal threats, but they are meaningless signals if there is no one there to be threatened.
This is important. Humans with no-self psychopathy experience pain in the moment, but the whole self-referential apparatus that most people have built on top of it is missing. Just pain, instead of a high-dimensional experience of suffering that is intertwined with self/identity.
Consider how two people might react to being told that they’re stupid:
Person 1:
She thinks I’m stupid.
If I’m stupid, I’ve probably said a number of shamefully wrong things in front of others.
Why didn’t they tell me? Maybe they are so used to me being stupid that they don’t bother anymore.
To them I’m like that unself-aware comic relief guy that TV shows add for the extra slapstick humor.
I’m the butt of everyone’s jokes, and I never noticed.
They’re all laughing or eyerolling about me behind my back.
Maybe they still tolerate me because my body type is conventionally attractive.
I better fast more or my body type might change and they’ll finally get rid of me and I’ll die alone on the streets.
I’ll be helpless out alone on the streets. Random strangers will rape me.
What if I’m homeless and then all my identity documents get stolen.
I can’t go to the police or they’ll think I’m stupid too.
I can never be seen by any of my friends again. If they never meet me again, I can’t say any more stupid things to them, and they’ll continue to tolerate me.
I’ll be all alone, but at least I can’t do any more harm.
Oh god, what if I do harm by being so stupid.
I’m so close to completing my psychology PhD, but then what? I can’t possibly work with people afterwards or I might do harm by being so stupid.
I have to sabotage my PhD so I’ll have an excuse to remain a student forever.
Was it arrogant of me to think that I could ever work with people?
Maybe I was arrogant too on top of being stupid.
I’m just like Trump, and all my friends hate Trump.
Why haven’t they all rejected me yet? I must’ve been fooling them!
I’m a fraud on top of being arrogant and stupid!
I should just kill myself right now. I’m evil. I shouldn’t be allowed to exist.
It’s the only responsible thing I can do. I need to save the world from myself.
Person 2:
She thinks I’m stupid.
Person 2 has a much more peaceful internal experience, and while it’s entirely possible to have a self experience that is not self-punishing and catastrophizing, a sure-fire way to produce that state is to not have a self in the first place.
Perks of having a self. But a pervasive feature of human culture, and probably that of other mammals too, is that the use of a self is strongly encouraged. Parents (at least Winnicott’s good enough parents) instinctually teach selfhood through contingent marked mirroring, i.e. the practice whereby a parent recognizes what a child feels (contingent), assesses it using their adult sense and sensibilities (marked), and mirrors this marked version back to the child like, “Oh, I see that you’re cold but it’s okay, we’re almost home.”
This process teaches the child that in our culture it’s understood that the child has feelings and can communicate them, that the parent has feelings that are distinct from the child’s feelings, and that inanimate objects don’t have feelings. Implicit in that is an assignment of primary responsibility of certain feelings to certain people, an important convention that underpins our legal systems. Deterministically speaking, it would’ve been fair to agree to blame all transgressions on our earliest ancestors, but we’ve found that arrangement to not be culturally adaptive.
But selves are not only useful to make social conventions more intuitive. Selves also facilitate learning!
M.E. Thomas has coined the term “psychopath stupid” for the way that she avoids owning knives lest she cut herself and the way she keeps missing flights. She has a high pain tolerance and low proneness to fear, so the direct feedback she gets from either of those is mild. Crucially, there is no hint of self-referential catastrophizing.
Imagine a friend tells you on the phone, “Ugh, I missed my flight and had to book a new one, and now I’m missing out on some of my vacation.” Then a year later the friend tells you, “I’m flying to [vacation destination] on Friday.” It takes special friendship skills to remember that the friend might miss the flight again and to remember to remind them on Thursday evening. I imagine the distance that we feel to the experiences of a friend or acquaintance is illustrative of the distance someone without a self feels to “their own” (as it were) experiences.
M.E. Thomas has some insightful videos on this topic, is writing a book about it, and often argues in favor of selves. She has trained selfhood over the past decade and loves it. She also argues that we need to keep up the practice or selfhood slips away, like a muscle if it’s not trained. This may be more true the later in life it’s learned.
But note that M.E. Thomas’s self experience is (or was) unusually absent. In my interview with Daniel Ingram, I got the two people together who I know who have the least selfhood going on. Everyone else I know (all my other psychopathic friends included) either has many noncohesive fragments of a self whizzing around or even has an unusual rigidity to their self experience that is indicative of extensive repressed parts. Tiffany talks in her interviews about episodes when she dissociated from all her alters and temporarily entered a state that might be similar.
Mentalization-based treatment (MBT) could be seen as an intervention that is aimed at practicing or strengthening a sense of self. As the name suggests, it trains an ability called mentalization that is the antidote to the three types of non-mentalizing: When you mistake your own emotions/attitudes/etc. as dwelling in another, when you mistake your own assumptions about the emotions/attitudes/etc. of another for their actual emotions/attitudes/etc., and when you mistake fictional emotions/attitudes/etc. for real ones.
There are conventional as well as objective aspects to good mentalization, and the conventional ones overlap greatly with selfhood.
The Buddhist Parallels
If good-enough parenting and MBT are ways to learn selfhood, then Buddhist trainings to attain anatta/no-self are methods to unlearn selfhood – MBT as reverse Buddhism.
Daniel Ingram, author of Mastering the Core Teachings of the Buddha, describes the meditative stage of Equanimity (the 11th of 16 ñanas, or “knowledges”) in terms strikingly similar to no-self psychopathy:
It’s open, expansive, flowy, very natural, very clear, very inclusive, very equanimous, and it can feel very nice but in a less dramatic way than the A&P [Arising and Passing Away] which can feel very rapturous or even orgasmic or something very blissful for some. … There’s not this contraction into guilt or fear or whatever. The sense of self seems very thin or kind of fluid or liquid or transient, like why would you care about those things? Why would you be worried with remorse or scared or something. So there is this funny sort of Equanimity-ish quality.
Advanced meditators who reach this stage – and later stages like Fruition – report experiencing sensory inputs, including painful ones, without the self-referential overlay that normally makes them terrible. They note the pain and move on. The pain is still registered but doesn’t grab them.
As I argued in “Is Enlightenment Controlled Psychosis?”, insight practice is, in a sense, a structured curriculum for unlearning the three things infants learn in their first years: interpreting sense inputs (impermanence), usefully assigning valence to states (dissatisfactoriness), and sorting the states into self and other (no-self). Advanced meditators achieve a state where all three are unlearned, be it just momentarily, and the result is a phenomenology remarkably similar to what people with no-self psychopathy report.
The Buddhist term for the self-referential component of suffering is upādāna – “clinging” or “grasping.” The claim of the Buddhist tradition is that suffering = pain × clinging. Without clinging (without a self anything could cling to), there is pain but not suffering in the full sense.
I’m making a distinction between:
Nociception. Nerve signals.
Pain. A very rudimentary and evolutionarily old interpretation of nociception as undesirable.
Self. A newer, learned identification with some sensations.
Suffering₁. Pain × self, pain that is concerning because it concerns “oneself.”
Suffering₂. All the feedback effects between self and suffering.
Mentalization. High reflective functioning that allows you to introspect on these processes.
I argue that nociception is neutral, pain is bad, and the addition of a self makes it all worse, but I don’t know how much worse. I don’t think mentalization has an influence on the process except in that it can put the brakes on the catastrophizing processes of suffering₂.
A friend of mine suggests that without a self, it’s neutral whether 10 people suffer for a duration d or 1 person suffers for a duration of 10d. With a self, the longer suffering is disproportionately worse because, w.l.o.g., at time 5d, there is someone there who’s still affected by the suffering from 0 to 5d.
The Nematode Model
Many invertebrates (excluding at least bees and ants) don’t have much culture, so there’s no reason for them to have evolved selfhood. Their suffering will stem more purely from the direct experience than from any self-referential feedback effects.
Bees and ants are the interesting case. Eusocial Hymenoptera have role-based hierarchies (forager, nurse, guard, queen), social learning of those roles, and recognition behaviors – exactly the cultural ingredients the model predicts should produce something self-like. If the model is right, social insects should show signs of suffering₁ that solitary insects shouldn’t. Whether they actually do is an open empirical question, but it’s where I’d look first – and the place where the welfare ranges of arthropods most plausibly diverge from one another, with eusocial species at the wider end.
As an aside, that’s also why people without a self can appear so manipulative, controlling, or boundary-crossing: They have many of the same impulses to seek or shun things that we all do, but whereas the Median Mary has a sharp distinction where controlling herself is good and controlling others is bad, the Anomalous Anattarite can’t make that distinction and only seeks and shuns on behalf of “themselves” more often because that’s someone they find easiest to control. They’re much like an EA who goes into politics not to get rich (personally) but to produce riches (grow the pie for everyone).
While the lack of a self doesn’t eliminate the suffering – pain is still pain – it eliminates a lot of its complexity. It stands to reason that there’s no such thing as trauma for flies.
That’s also why I like to say that turtles make for great enlightened masters: They have, by nature, the no-self experience that Buddhist practitioners seek.
A nematode, by this model, has:
Nociception. Yes. The neural hardware is there and functionally active.
Pain. Probably. The evidence from classic conditioning suggests the signal carries intrinsic “badness.”
Suffering₁. Probably not. It seems like an overly strong assumption to think that the conditioned behavior is mediated by a self.
Suffering₂. Almost certainly not. This is a stronger version of suffering₁.
This gives a principled story about how welfare ranges might cash out across taxa. Rethink Priorities’ welfare-range estimates aggregate 47 hedonic and 35 cognitive proxies; they don’t lean on neuron counts as a primary input, though Vasco Grilo has noted that the results end up mildly correlated with neuron counts anyway. The no-self model adds an orthogonal cut: it predicts a narrower between-taxa gap on the pain layer – which seems to be broadly comparable across all probably-sentient animals – and a wider gap on the suffering layers, which depend on selfhood, are absent in nematodes, and are plausibly present in eusocial Hymenoptera. If the model is right, RP’s aggregated ranges may bundle two things that come apart architecturally: the presence and intensity of the pain layer, and the presence of the suffering layers built on top of it.
How This Connects to the Sequence
The no-self model of invertebrate experience doesn’t change the practical recommendations from part 3. Reducing populations of likely-suffering organisms through NPP reduction remains the cleanest intervention regardless of whether their suffering is “full human-like” or “attenuated no-self.”
But it does sharpen the way the Welfare Footprint Institute Pain-Track categories that part 3 leans on – annoying, hurtful, disabling, excruciating – should be read across taxa. The higher categories are typically defined in terms that include mental withdrawal, inability to attend to anything else, and despair – features easier to imagine in animals with selfhood than in animals without. Hurtful and below might be available to organisms that have pain but no self. If that’s right, the WFI scale plausibly compresses for organisms below the selfhood threshold: their worst pain might still be very bad, but it would be bad in a different way than the worst pain of a self-having organism, even at the same nociceptive intensity. I’d want WFI-style estimates to come paired with a claim about which categories the organism is architecturally capable of reaching.
It also affects how we should allocate research attention:
For nematodes: The case for suffering, as I use it in this article, is weak, but the question remains whether the pain needs to be more intense to compensate: Some forms of learning are impaired for people with a no-self experience, but perhaps it proved evolutionarily adaptive to enhance the intensity of the pain until learning does happen rapidly enough. There is also the related argument, sometimes attributed to Richard Dawkins, that simpler minds might need stronger signals to learn.
For arthropods: The key question is whether they show signs of selfhood – flexible, learned roles in hives, pack-like organization, humiliation, etc. Bees might have wider welfare ranges than flies as a result.
In part 5, I’ll turn to the biggest variable of all: How does artificial superintelligence change the welfare ecology landscape? The answer involves Dyson swarms, digital suffering, and the most important value-loading problem in history.