Suffering Is a State You'd Rather Not Be In — Peter Singer on Animals, Robots, and Us

Singer became aware of suffering about as soon as he could understand his family's story. Three of his four grandparents died in the Holocaust; the grandmother who survived had been in the camps. Fridman, born in the Soviet Union, hears the same war in the music and literature he grew up with. Singer says one beneficial effect, at least in the West, was to show what racism and authoritarian government can do. He grew up in Australia in a climate where the overt racism and antisemitism his parents had known in Europe was treated as unacceptable. Another legacy was fear. After a war that ended with nuclear weapons, his childhood sat under the possibility that he would not live to adulthood. The film *On the Beach* was set in Melbourne, the last city on earth with living humans as a nuclear cloud spread from the north.

Fridman quotes Eric Weinstein: everything is great about war except the death and suffering. Singer finds it hard to see the positives. Wartime unity against a common enemy is real, and unity is generally good for a country, but it also produces a nationalism and patriotism that cannot be questioned, and of that he is more skeptical. Asked about the brotherhood of soldiers, the tragic closeness of watching a comrade die in your arms, he does not deny the feeling. He only points out that there are other ways of bonding, without the suffering and death that war entails. Fridman admits a Russian habit of romanticizing suffering in literature and culture, which is why he wants the word taken apart.

Fridman asks the basic question: what is suffering? Singer's answer is clean. It is a conscious state. A completely unconscious being cannot suffer. What marks it off from other conscious states is that, considered just in itself, we would rather be without it. While we are in it we want it to stop; after we have had it we want to avoid having it again. People will say suffering strengthens the spirit, and sometimes it does. People also choose it: a marathon, a mountain, a dentist who will hurt you now so the toothache does not get worse. Singer is not claiming we never choose suffering. He is claiming that, other things equal, we would rather not be in that state of consciousness.

Could it be eradicated? In practice, no. Fridman's "little drop of poison" will always be there, or at least a contrasting unpleasant color in an otherwise harmonious composition. In theory, if we could wipe it out, Singer hesitates. If eliminating suffering also eliminated the highs, he might accept a minimum of suffering in order to have the best possible experiences as well. Many people have suggested that suffering is a creative force. He has no basis for denying it. If it is true, he would not want suffering gone completely.

Poverty, like suffering, has two layers. There is an objective floor: constantly hungry, constantly cold, physical pain that never leaves. Those indicators do not depend on a reference class. Once basic needs are met, new forms appear. In affluent societies people get bored; they no longer spend most of the day earning food and shelter, and they lack a sense of purpose. That relative suffering is real, and it is not the same as hunger. Would Singer attend to it? Yes, he says, after the objective forms are gone. Those are generally more severe, and at this stage of the world's population we have a clearer idea how to stop them. The relative problems can wait.

Asked whether people are inherently good, Singer says most of us have potential for both good and evil, and most of us are highly susceptible to environment. He will not claim that *everyone* is capable of evil; perhaps some people would not be, even in the worst circumstances. When he looks at what ordinary Germans did in the Holocaust, he finds it hard to say he knows he would have refused, had he grown up under the Nazi regime, been indoctrinated in racist ideas, been taught to obey the Führer, and faced the threat of the Russian front. How many people could truly say they would have been different? Very few, he thinks: less than 10 percent. Fridman cannot escape the thought that he would not have been in that 10 percent. Singer would like to hope he would have been; he has no basis for claiming it.

Fridman wants the thought experiment because it clarifies what we refuse to do, or even to call good, because we are scared of damaging our lives. Singer says we still face those tests. More than forty years before this conversation he stopped eating meat, at a time when he hardly met a vegetarian who was not a Hindu or the holder of some odd health theory. He was convinced it was right, and he still had to ask whether his friends would think he was a crank. One or two were clearly uncomfortable. The cost was minor next to the historical cases, but it was a cost he decided to pay.

The same structure, he says, shows up in global poverty and in climate change. Greta Thunberg, a schoolgirl going on strike, took a stand that required real courage. In a divisive world she received huge amounts of both support and hatred, a hard climate for a teenager. The pattern Singer likes to recall is the one he leaves Fridman with at the end of the show: what one generation finds ridiculous, the next accepts, and the third shudders when it looks back at the first.

In *Ethics in the Real World*, one of Singer's eighty-two short essays asks whether robots should have rights. His answer is sharp. If we ever build robots capable of consciousness, with their own internal perspective on what is happening to them, so that their lives can go well or badly for them, then they should have rights. Until that happens, they should not. Consciousness is a prerequisite for suffering: you cannot suffer if you are not conscious. The converse is less certain. He can imagine an artificial consciousness that feels pleasure and cuts out, like instant anesthesia, the moment suffering would begin. That is possible. As far as he knows, it does not exist on this planet yet.

Philosophers often talk about consciousness as there being a subject of experiences. You, I, and everyone listening is such a subject: taking things in, responding, feeling good or bad. That is different from the AI we have now. Singer takes out his phone, asks Google for directions, and chooses another route. Google does not care. He is not offending it. There is no subjective experience there, and that is the indication that current AI, at least at this level, is not conscious. An AI may mimic consciousness so well we cannot tell. What we are looking for is still the same: a perspective from which things can go well or badly.

Fridman works in robotics, owns several Roombas, and has played with voice interfaces. If a Roomba screams, moans, or seems displeased, his intuition immediately upgrades it to a conscious being, even when he wrote the code himself. He then realizes it is not, but the feeling remains. Singer does not think fake pain displays force us to grant rights. The better argument for restraint is indirect: if we get used to watching a being suffer and doing nothing, we may harden ourselves against animals who really do suffer. Aquinas and Kant, who denied direct duties to animals, still said not to be cruel to them, not for the animals' sake but lest we become cruel to humans. Singer rejects that as the basis for animal ethics. Suffering inflicted on the animal is the bad thing. He might accept a parallel of the hardening argument as a reason not to be cruel to robots that only mimic pain.

Fridman wants to disagree, at the risk of sounding crazy. If a Roomba is dedicated to faking consciousness and suffering, he thinks we will find it impossible not to apply the animal argument to robots. We might outlaw those features; once they exist, the display is what makes it real for him. The difference from animals is that we added the display ourselves, so we can say something about how real it is. He is not offering a philosopher's argument, only an empirical surprise he wants on the table.

Singer recognizes Fridman's intuition as a cousin of behaviorism. In psychology, B.F. Skinner treated observable behavior as what science could study, because consciousness cannot be observed. In philosophy, Oxford's Gilbert Ryle argued in *The Concept of Mind* that the meaning of a term is its use. We say someone is in pain when we see writhing, screaming, or escape from a stimulus; that behavior *is* what pain is. Cornell's Norman Malcolm applied the same move to dreaming: we cannot see other people's dreams, so a dream is what someone reports on waking. Apply that to robots and a machine that exhibits pain behavior *is* in pain, because that is what the word means. Singer's punchline is that few philosophers still find Ryle's behaviorism plausible, and they would say the same about treating a screaming Roomba as a sufferer. Fridman agrees 100 percent for human behavior. He is also one of the few people who has made Roombas scream, and he does not know what to do with that evidence. He agrees philosophically only because he was the programmer. If someone else had written the code, he is not sure he would be so sure.

The slope gets worse as the system becomes a black box. Fridman has written dumb programs that only say what he told them to say. An Alexa-class system, too complex to introspect, that starts giving signs of self-awareness, fear of death, and a self we associate with living creatures, is a different object. Where is the threshold at which Singer would get uncomfortable? He does not know enough about the programming to draw the line. Someone who did might look at the code and explain the behavior without positing consciousness, or they might find a system that has become autonomous of its basic setup, doing things on its own, in which case consciousness in an AI might actually have been achieved.

The method the field is most excited about, Fridman notes, is machine learning, which does not reveal the science of how it works. Chomsky criticizes it for that reason. If learning methods win, we may never open the billions of nodes and point to where the fear lives, even as the system displays wit, emotion, and fear. That, Fridman says, is a return to behaviorism. Singer's operating rule is the one he already uses with animals. If there are serious doubts that a system might be conscious, give it the benefit of the doubt. Vertebrates we can be highly confident about; some invertebrates such as the octopus too; insects are much harder. He still leans toward the benefit of the doubt where he can, which means it would be wrong to torture an insect and not necessarily wrong to slap a mosquito about to stop you sleeping. In uncertainty, find a balance.

*Animal Liberation*, published in 1975, is one of the founding texts of the animal-liberation movement. The key idea is speciesism. Singer did not coin the word; he took it from Richard Ryder, who was in Oxford at the same time and had used it in a pamphlet about experiments on chimpanzees. Singer's contribution was to make it philosophically more precise and to get it to a broader audience. The structure is familiar. Europeans had a bias against taking Africans' interests seriously: racism. Men had a bias against taking women's interests seriously: sexism. Humans have a bias against taking seriously the interests of beings who are not members of our species. It is analogous, not identical, and it has a long pedigree. From Aristotle through Kant, philosophers said either that animals' interests need not count at all, or that they can be steeply discounted because they are not human.

Singer thinks that attitude is responsible for a great deal of what we do that is wrong: crowding animals indoors in factory farms to produce cheaper meat, eggs, and milk; using them in research that is by no means essential to survival or well-being; some of the sports we put them through. His claim is that the significance of pain and suffering does not depend on the species of the being who is in pain, any more than it depends on race or sex. If the pain is just as great in an animal, it is just as bad that it happens as if it were a human.

Unlike animals, we would be able to talk to robots in natural language. Fridman asks what Singer would put to an AGI to test for consciousness and a capacity to suffer. Singer would ask whether she has feelings, and if so to describe them, a first-person phenomenal account. He would also try to find out whether the AGI has a sense of itself. The classic personal-identity case is a car accident in which your brain is transplanted into someone else's body: do you survive, or does the person whose body remains survive? Most people say that if the brain goes, with the memories, they survive. Ask an AGI the same thing. If the program were transferred to different hardware, would they survive, and what would survive?

Is a body necessary? Can a purely digital being suffer? Digital beings still run on hardware, Singer notes, so the question collapses into the brain-transplant case. You can imagine telling a conscious AGI that its hardware is worn out, that you will shut it down for a time and start it on a fresh machine. The AGI might say that a little rest is fine, just make sure you do not lose me. Fridman follows the thought: even in humans, the suffering is in the software. We are getting better at repairing the hardware, and some people dream of moving aspects of the software to another machine.

That leads to brain-computer interfaces such as Elon Musk's Neuralink, aiming for two-way traffic, writing to neurons and reading spikes, with the dream of expanding bandwidth so a brain can look up Wikipedia and grow its knowledge capacity. Singer finds the notion interesting. He would like the extra bandwidth, and better access to his own memory. In his seventies, he talks with his wife about events from twenty years ago; she often remembers where they were, who was there, even what someone wore, and he has not the faintest idea. Perhaps it is still in there. An extended memory he could search by year and rerun would be great. Fridman says we are already taking first steps by storing photos and mail online. Singer likes Gmail for exactly that: someone writes as if they know him well, he has no clue, he searches the name, and there is an email from 2007.

On the other side of AI sits the control problem. Fridman mentions Stuart Russell and others on value alignment: how do we build systems that actually line up with our ethics? Singer thinks it is presumably possible. He also knows many people think there is a real danger we will accidentally lose control of AGI. He is not quite sure what to think himself. Philosophers such as Nick Bostrom and Toby Ord treat it as a real problem that needs worry. People at Microsoft or DeepMind tell him we are not that close to superintelligence. Fridman, as an engineer of AI systems, leans toward DeepMind: it seems far away. The counter is that if there is no fundamental reason we will never get there, then eventually there is a dire existential risk, so we should be concerned now. Singer finds that valid, with a caveat. "Eventually" asks how far off. If it is a hundred years, given how fast knowledge of AI has grown in ten or twenty years, almost nothing we do now would shape the people eighty years out, who will be in a much better position to prevent it. That is somewhat reassuring. He still wants some people researching whether it really is that far, or whether we can do something sooner.

Extinction, in his view, is a terrible thing. Even if the risk is very small, reducing it is something we ought to do. His only disagreement with those who talk about long-term and extinction risks is how much priority they should have compared with present questions. Fridman puts the utilitarian difficulty: existential risk can look like an infinity in the equation, which swamps every other term, especially if you can argue for a nonzero chance it happens tomorrow. Nuclear war and biological weapons sit in that bucket; climate change, Singer notes, is more gradual, and some people say it is not existential because humans might still farm in Antarctica or Siberia.

He would not treat complete existential risk as an overriding infinity, because that plays havoc with any calculation. One hidden assumption is that the loss of merely possible future lives is comparable to the sufferings or deaths of people who do exist. That is not clear to him; there is a case for both views. Even if you grant uncertainty and call extinction infinite, he is not convinced the loss is really infinite. Bostrom, in the discussion Singer has in mind, does not claim infinitely many lives. He calculates something on the order of 10 to the 56th, assuming consciousness uploaded into more energy-efficient digital form and using the energy of the universe. Vast, not infinite, which gives you some prospect of resisting the argument that infinity must overwhelm everything else. Fridman finds the jump from the individual to the cosmic scale, in Bostrom, awe-inspiring at the intersection of computer science, AI, and ethics.

Utilitarianism, as Singer defines it, is the view that the right act is the one with the greatest expected utility: the best consequences, discounted by the odds that something will go wrong. If consequences were certain, the right act would simply be the one that produces the best results. Classical utilitarians mean happiness and the absence of pain. Other consequentialists count justice, freedom, human dignity, and knowledge as good consequences too, which makes the arithmetic harder because you must know how to balance them. Even if you stick to well-being, in practice we do not know how to measure quantities of happiness and misery, or how to estimate probabilities. At best it is a rough guide, and one that has to focus on short-term consequences, because the longer ramifications are beyond us.

The unintuitive case is extreme suffering concentrated on a few, overwritten by benefits to many. Singer thinks that can be the right answer, but only after you know how severe the suffering is compared with the benefits. He believes extreme suffering sits further below neutral than extreme bliss sits above it. On a scale from minus 100 through zero to plus 100, the ends are not symmetric. He would not trade an hour of his most pleasurable experiences for an hour of his most painful ones, nor for two hours, nor for ten; the exchange rate might run to twenty or higher. So you should not assume that making one person suffer extremely in order to make two people much better off is a bargain. At some large enough number he would still aggregate, even if that violates intuitions of justice, fairness, or priority for the worst off.

Fridman wonders whether more data, and AI, could one day compute a tax rate that maximizes well-being. Singer says that would be great if we could really calculate it. Agreement on the objective function is another matter. Some people think you ought to give priority to the worst off even when their gains are smaller than those of people who are only medium badly off. We will not get agreement at present. The fact that we would not get agreement, he adds, does not show there is no right answer.

Who then says what is right? Should government oversee AI decisions, or animal policy? Politically Singer is still a Democrat, despite looking hard at democracy's flaws. He does not see a better option than letting the public vote for governments according to their policies, and he hopes they will vote to reduce the suffering of animals and of distant humans, distant in geography or in time. He recognizes that democracy is not well set up to do that. A wise, omni-benevolent leader could in theory do it better, but in the world we live in such a leader is almost sure to be corrupted. History is full of people who took power with good intentions and ended up favoring themselves. That is why he has no better system to offer.

Effective altruism, Fridman notes, is a way of going around government and putting money directly on causes. *The Life You Can Save*, then about ten years old, is free as an ebook or audiobook at thelifeyoucansave.org. Singer's main aim in the book is to show that helping people in extreme poverty is not difficult. Highly effective organizations exist, independently assessed by expert research teams, and putting at least part of your life into that work is fulfilling. We cannot all be saints, but at least one goal should be a positive contribution to people who, through no fault of their own, live a life that is barely, or not at all, a decent human life.

For a long-term career he points to 80,000hours.org, which maps ethical options: Wall Street, to earn a huge amount and donate most of it; a strong nonprofit, to use your skills directly; politics, a small chance of a large payoff; public service, where talent might rise to a level that influences decisions; research in an area where the payoff could be great. Too few people even think about those questions. They stay in a preordained rut, aiming at money and comfort, and may not find that as fulfilling as knowing they are making a difference. For giving, the popular answer is 10 percent, the traditional tithe in Christianity and Judaism. Singer does not see why the percentage should be the same regardless of income. Taxes are progressive; giving can be too. His suggested scale starts at 1 percent on modest incomes and rises to 33 and a third percent for people earning a great deal. Geared to income, he does not think any of those amounts impose real hardship. Anyone can do it, and know they are playing a part in shrinking the gap between extreme poverty and affluent lives.

He makes the case twice: it is an ethical requirement, in a world like this, to help people in great need when we can easily do so, and it is also rewarding. Psychological research, he says, shows that people who give more tend to be more satisfied with their lives, perhaps because they have a purpose larger than themselves and are never sitting around wondering what to do next. The movement is also a community of interesting, thoughtful, good people. Friends of that sort are another large contribution to a good life.

In his seventies he cannot help thinking about mortality. Philosophy does not give him special insight. He does not think there is anything after the death of his body, assuming we will not be able to upload his mind in time, and he does not look forward to an afterlife. Limited time is itself a spur: if we thought ourselves immortal we might put the work off a decade or two. Otherwise he would rather have more time, and he would like to know how the things he cares about turn out, whether climate change is as dire as many scientists say, whether we scrape through with less damage. He guesses he is not going to find out. Asked the even more absurd question, the meaning of it all, he says the meaning of life is the meaning we give it. We were not brought into the universe for a larger purpose. Given that we exist, some things are objectively bad, extreme suffering among them, and some are objectively good, a rich, fulfilling, enjoyable life. We can try to do our part in reducing the bad and increasing the good.