He Left School at 16 to Clean Toilets in the Outback. Now He's Trying to Keep AI Honest.

Simon Longstaff's mother died when he was seven. He was sent to boarding school, and when he turned sixteen his family told him plainly: everything they had to give him, they'd already given, in the form of that education. He was on his own from there.

With no trade and no university place, he took a job as a "service attendant" on Groote Eylandt in the Gulf of Carpentaria in the mid-1970s — cleaning toilets, emptying bins of stale milkshake and prawn shells under clouds of mosquitoes. He talked his way into becoming a paramedic and fire officer, got his driver's licence behind the wheel of a red fire truck, and drove ambulances on an island with no resident doctor, including his first case of a life he and his colleagues couldn't save.

He was also adopted into the Lalara clan of the Anindilyakwa people — one of fourteen clans on the island — given his own Indigenous name and songline, and remains connected to that community more than fifty years later.

Learning to see patterns, not objects

The lesson that shaped his entire career came at the end of a wharf. Elders on Groote Eylandt took him out to the shiploading port and told him they wanted to teach him to see. Look, they said, and you'll see the dolphins. He looked. He couldn't see them. They laughed — warmly, he's careful to specify — and told him the problem was that he was looking for the dolphins at all. Their way of seeing the world wasn't a search for discrete objects, a dolphin here, a cup there. It was reading the pattern the ocean made, and noticing how that pattern changed when dolphins moved through it.

He didn't understand what he'd been given until decades later, mid-interview on a radio program, describing how his ethics work moved across an enormous range — preparing soldiers for war at one end, end-of-life hospital decisions at the other. "For me, it's all about the patterns," he said, and then realised, out loud, that the ability he'd been using his entire career as a philosopher had been taught to him on that wharf.

Ethics is not an optional extra

Longstaff has spent almost thirty-five years as Executive Director of The Ethics Centre. His central complaint about how the tech industry treats ethics is that it's bolted on after the fact, as a compliance checkbox, rather than built in as the foundation. His test for whether you're doing ethics at all is almost insultingly simple: any time you stop and ask "what should I do," as opposed to just doing what habit or instinct suggests, you're doing it.

"A lot of people in tech say, because I can do it, that means it's okay that I should do it. Well, no. One should think about the affordances which are created by technology, and how you can restrict the affordances to those which are reasonable."

His proposed model for governing AI is deliberately borrowed, not invented: treat it like pharmaceutical approval. Don't mandate which specific values a company builds into its system — that stays the company's choice, preserving room for genuine innovation. Instead, require that they can name the values and principles the system was designed around, demonstrate those principles are actually present in what they built, let an independent third party verify it, and use something like blockchain to prove the system hasn't been quietly altered since. "None of this limits innovation," he says. "You're entirely free to do it in ways which might be novel. But you must be able to tell us."

The necessary fiction

The hardest problem he names isn't malicious use — it's that nobody, including the people who built a given model, can fully explain why it produced a particular output. His answer borrows from government: a minister is legally responsible for their department even though it's physically impossible for one person to monitor every public servant and every system underneath them. Society imposes what he calls a "necessary fiction" — you are held to have known, even though you couldn't have. He argues AI needs the same fiction applied to the executives deploying it. If a company's model has discharged its ethical obligations as designed, the CEO using it is still responsible for what it produces, black box or not.

What survives automation

Asked what happens to work and meaning as AI and robotics take over more of it, he reaches for two historical precedents most people wouldn't put side by side: ancient Athens, whose citizens were freed for civic and artistic life by an economy built on slave labour, and pre-colonial Indigenous Australian societies, where life was rich and structured without anything resembling a wage job. Neither, he argues, collapsed into meaninglessness for lack of employment in the modern sense — which is his case that a society built on AI-driven abundance doesn't have to either, provided it deliberately redistributes the value that abundance creates, an idea he's worked on with economist Alan Schwartz under the name "the universal commons": if clean air, mineral wealth and public infrastructure are held in common, a share of what's extracted from them belongs to everyone, not just whoever owns the machine doing the extracting.

What he insists won't be automated away is a very specific kind of contact: a doctor's hand on a patient's shoulder delivering a terminal diagnosis, a colleague saying "I care about you enough to tell you to stop" instead of a sensor silently halting a conveyor belt. A machine can simulate the gesture. It cannot share the mortality that makes the gesture mean something. "That is non-accidental," he says. "I think that only another person could do that."