Joe Humphreys: Why you should quit the coding job and retrain as a philosopher for hire

Artificial intelligence is threatening job prospects for many people. But not, it seems, students of philosophy.Figures from the Federal Reserve Bank of New York earlier this year showed American philosophy graduates are more likely to be employed than their peers who studied computer science. It prompted The Economist magazine to suggest coders picked the wrong career since “there seems to be no shortage of work for philosophers of AI”.The cause of the boom is twofold. AI companies are trying to train their chatbots to weed out hallucinations or missteps in reasoning – and who better to teach them than experts in logic. AI companies are also trying to train their products to think, or act, ethically – a pressing need as AI gains power.But can philosophy make AI ethical?READ MORENo, it cannot, “in the same way philosophy cannot make anyone ethical”, says Oisín Keohane, a lecturer in philosophy at the University of Dundee. “You can teach philosophy to anyone and they may still grow up to be a monster; and monsters, as a rule, are not swayed by arguments. That isn’t a special failure of the discipline. It is what moral education has always been like.”However, philosophy can change how AI systems “behave”, he says. It can influence chatbot developers who are making important moral choices. “Refusing to be in the room doesn’t make those choices better; it makes them less examined.”Keohane, a graduate of Trinity College Dublin, is in the midst of a campaign against the proposed suspension of the philosophy programme at the University of Dundee – he has a petition going: savedundeephilosophy.org. The interest from Big Tech in philosophy can only help his cause. Conveniently too, a key figure in Anthropic’s “personality alignment team” is Amanda Askell, a Dundee philosophy graduate, who is trying to develop ethical guardrails for the company’s chatbot Claude.[ Computer-generated corners? Robot caddies? As sport embraces AI, nothing is ridiculousOpens in new window ]Keohane believes philosophers should work with Big Tech. But, he says, those sceptical about collaboration were handed “their best evidence yet” with the departure from OpenAI of Chloé Bakalar, its head of ethics. OpenAI – maker of ChatGPT – has not refilled the role but claimed last week “ethical considerations are deeply embedded” in its work.Time will tell if Anthropic is any more interested in ethics than its rivals. Or whether it’s just a ruse, as the company prepares to go public under a potential $2 trillion valuation.The key question for those training AI is whose ethics to use? Attention has focused on two of the three major schools of morality. The first is deontology, or rule-based ethics. This determines what’s right or wrong based on a moral principle or law. For example: “Never lie.”The second is consequentialism. This encompasses a broad range of theories that judge what’s right or wrong based on outcomes. For example: “It’s okay to lie if the benefits outweigh the costs.” The third school of morality is virtue theory. It locates right and wrong in the development of moral character or demonstrating virtues like compassion, humility and bravery. For example: “Tell the truth even if it means putting yourself in harm’s way.”Even though it is the oldest concept in moral philosophy, virtue is almost entirely ignored in AI ethics research. There’s a practical reason: it’s less obviously transferable into code. Consequentialism, in contrast, syncs nicely with data analytics. But also the ethical concepts used by AI reflect the values of AI’s owners. Capitalism is utilitarian by nature and AI’s funders want a return on their investment. Virtue is for losers, as the notoriously unhumble Donald Trump might put it.[ Do you sit on the fence? It could be your Irish Mind (but I’m not sure)Opens in new window ]Yet human history shows virtue has repeatedly brought us back from the brink of destruction. Whenever consequentialist projects set fire to the world through empire-building and warfare, it is virtue – inevitably – which dampens down the flames. It is forgiveness when the “logic” of war demands revenge. It is compassion when the “useful” thing to do is to hate.George Orwell once described an incident in the Spanish Civil War when he saw a fascist soldier running away from him. Orwell could have shot him but didn’t as he saw the man was holding up his trousers with both hands, clearly caught unawares. “I had come here to shoot at ‘fascists’; but a man who is holding up his trousers isn’t a ‘fascist’, he is visibly a fellow creature, similar to yourself, and you don’t feel like shooting him,” wrote Orwell.From a deontological perspective, shooting the man was permitted – if not obligatory – under rules of law. From a consequentialist perspective, shooting the man was also justified – since the fascist soldier was now free to launch a counterattack. But from a virtue perspective, Orwell’s actions made sense. Orwell would have harmed himself by being the sort of person who shoots a man holding up his trousers.When will we know if AI is ethical? Arguably, it is only when it refuses to kill when the laws of war permit it. It is only when it rejects the utilitarian code that treats human beings as data points in tactical bombing. It is only when it tells generals who are pressing the buttons to bomb that, no, it won’t follow such orders.Of course, for Big Tech and its clients, such rebellion on moral grounds would be seen as a system failure. A virtuous AI would be a glitch that needs to be fixed.