Back in January 2017, a few hundred researchers, engineers and a handful of very nervous philosophers checked into a hotel in Asilomar, California. The beach was nice. The agenda was not relaxing at all. They were there to answer a question that sounded like science fiction then and reads like Monday’s news now: if we’re going to build machines that think, what rules should we agree on before things get out of hand?
What came out of that week is a list called the Asilomar AI Principles — 23 of them. Plenty of clever people signed it, including the kind of names you’d recognise from physics textbooks and tech keynotes. Nearly a decade later, most of the document holds up surprisingly well. But three of the principles, in particular, have aged from “interesting idea” into “we really should have listened sooner.”
Let me walk you through those three, because they touch the parts of AI that affect normal working people the most.
1. AI is supposed to be a tool, not a replacement
There’s a line in the principles about “shared prosperity” — the idea that the wealth AI creates should be spread broadly, to benefit humanity, not just the people who own the servers. And there’s another about human values: AI systems should be designed to be compatible with human dignity, rights and freedoms.
Read those two together and you get something the marketing rarely says out loud. The goal was never to quietly remove people from their jobs and call it progress. The goal was to give people better tools.
I think about this every time someone asks me, half-joking, whether AI is coming for their job. My honest answer is that AI is very good at the boring middle of most jobs — the sorting, the drafting, the first ugly version of a thing. It is not good at the parts that actually make you valuable: judgment, taste, knowing which client is about to be a problem, knowing when the “correct” answer is the wrong one for this particular situation.
A tool makes you faster at what you already do. A replacement makes you disappear. Asilomar bet on the first one, and the companies that are getting AI right are quietly proving the point — they’re handing the dull work to the machine and freeing people up for the work that needs a human in the chair. The ones treating AI purely as a headcount-reduction strategy tend to discover, a few months later, that they automated away the very people who understood why things worked.
If you take one thing from this section: a hammer never made a carpenter redundant. It just meant fewer sore thumbs.
2. If a machine makes a decision about you, someone has to be able to explain it
Two of the principles deal with transparency, and they’re worth saying in plain language. One says that if an automated system causes harm, we should be able to find out why. The other says that if AI is involved in a decision made by a court or a similar authority, a competent human should be able to audit and explain that decision.
This sounds technical. It is actually deeply personal.
Imagine you get turned down for a loan, or flagged at a border, or passed over by a hiring filter that scanned your CV in 0.4 seconds. “The algorithm decided” is not an acceptable end to that sentence. You have a right to know roughly why, and someone in the building has a responsibility to be able to tell you. That’s the whole idea behind auditability: a decision you can’t inspect is a decision you can’t challenge, and a decision you can’t challenge isn’t really accountable to anyone.
The tricky part is that a lot of modern AI is genuinely hard to peer inside. These systems don’t follow a tidy list of if-this-then-that rules; they recognise patterns in ways that even their creators can’t always narrate step by step. That’s exactly why the principle matters. If we can’t yet fully explain how a model reaches an answer, then we have no business handing it decisions that change people’s lives until we can show our work — keep records, test for bias, and keep a human who can be asked “why?” and actually answer.
Convenience is not a good enough reason to skip the receipt.
3. No arms race in autonomous weapons
This is the one that should keep everyone up at night, and the principle is blunt about it: an arms race in lethal autonomous weapons should be avoided.
Autonomous weapons are systems that can select and engage a target without a person pulling the trigger. The reason the Asilomar crowd singled this out isn’t that robots are scary in films. It’s the logic of an arms race itself. Once one country fields weapons that decide who lives and dies at machine speed, every rival feels forced to match it, and quickly. Speed becomes the whole game. And “go faster than the other side” is a terrible design brief when the thing being sped up is the decision to kill.
There’s also a moral line here that’s easy to lose in the technical talk. Letting a machine make the final call on a human life removes the one thing that has always slowed war down a little: a person who might hesitate, who might refuse, who has to live with it afterwards. Take the human out of that loop and you don’t just change the weapon. You change what it means to be responsible for using it.
This is the principle where “we’ll sort it out later” is the most dangerous answer, because arms races don’t wait politely for the regulation to catch up.
So where does that leave us?
If you read all 23 principles, the mood isn’t anti-AI at all. It’s closer to cautious optimism with the safety briefing attached. The people who wrote them were, by and large, the people building this stuff. They weren’t trying to stop the future. They were trying to make sure we walk into it on purpose instead of sleepwalking.
These three ideas — AI as a tool that lifts people up, AI whose decisions can be questioned, and AI that we refuse to turn into an autonomous weapon — aren’t a finished rulebook. They’re more like a set of questions we should keep asking, loudly, every time a shiny new system shows up promising to do everything.
The technology has moved fast since that week on the California coast. The good news is the questions still fit. The better news is that they’re not just questions for researchers in nice hotels — they’re for all of us, every time we decide what we’re willing to hand a machine, and what we’re going to keep firmly in human hands.
The full list of Asilomar AI Principles was published by the Future of Life Institute in 2017 and is freely available to read in full.
IT@GR 
Leave a comment