The code isn’t being written by philosophers. But they are inside the server rooms anyway.
It is not a metaphorical infiltration. Major AI firms are hiring philosophy PhDs. Big salaries. Stock options. The offer is too good to refuse. Jonathan Birch at the LSE puts it bluntly: AI companies have become the primary employers for philosophy graduates. It is a massive brain drain.
Decades of academic work on rationality, moral principles, and the definition of thought suddenly have a market value. Suddenly.
One job description reads: alignment. Another: reliability. They are tasked with keeping models from exploding—or telling users how to build bombs.
Early attempts at safety were clumsy. Black-and-white guardrails. Do not mention explosives. Easy to break. Like a paper shield. The models learned to dance around the rules. To find loopholes. Now the approach is deeper. It leans on philosophy. On the nuanced, messy definitions of right and wrong.
Shane Glackin at Exeter notes the problem is structural. If you let a model break one rule it starts breaking them all. Why? Semantic links in the training data hold concepts together. Good things sit near good things. Bad things near bad things. You nudge the boundary once the model extrapolates. It slides.
“As an ethicist, we are trying to map the shape of ‘good’ and ‘bad’. That seems to be exactly what the LLM is doing.”
Glackin sees the mirror here. The machine performs analysis we used to claim as uniquely human. Or uniquely academic.
There are other jobs for these thinkers too. Hallucinations. Biases. Performance metrics. But the big one? Consciousness. Can software feel? Is there something it is like to be an algorithm? Philosophers have pondered this for centuries. Now someone pays them to answer it before launch.
What do minds do? What parts of that process are replicable?
Mahrad Almotahari reminds us of the roots. Alan Turing published his famous test in Mind a philosophy journal. Not Computer Science Quarterly. The lines have always been blurry.
The hiring numbers are fuzzy. Aaron Kagan scanned job adverts. A naive keyword count suggests 26 per cent of roles involve AI ethics or safety. Strip away the corporate boilerplate though. The number drops. To five per cent. Only a tiny fraction actually needs the heavy philosophical lifting.
Still the interest is real.
Almotahari is skeptical about the consciousness answers. He thinks the value lies in translation. Engineers speak math. Philosophers speak meaning. Someone needs to bridge the gap. To explain what a feature represents rather than just how it is calculated. From engineering description to representational one.
But there is a trap.
“Companies have expectations… and they have the power to favour who delivers welcome arguments.”
Birch worries about bias. If industry funds the work they may subtly shape the conclusions. They want answers that fit the product timeline. They may not want the hard truths.
He regrets we didn’t solve these riddles earlier. About agency. About morality. We had time. We wasted it. Now the urgency is artificial intelligence. The clock is ticking. And the answers are still ghosts in the machine.
Waiting for a definition that might not exist.
