I’ve read many books about AI in recent years. Almost every one of them has a section on ethics, usually a paragraph near the end of a chapter, sometimes a chapter of its own near the back of the book. At some point I started testing these sections against one question: what would a reader do differently the next day after reading this. Book after book, the answer was nothing.
The sentences are interchangeable. “It is important to consider ethical aspects.” “Responsibility ultimately lies with people.” Move either line from the marketing chapter to the HR chapter and it still fits, because it was never really about marketing or HR.
A lightning rod does one job. It carries a charge to the ground before it hits anything that matters. The ethics section in an AI book does the same thing to a different kind of charge: the one that would build if a reader stopped and asked whether what the book describes is actually a good idea. The paragraph discharges that question before it can become a problem for the rest of the chapter.
The paragraph is doing its job. A book about AI with no ethics section reads as reckless, so ethics goes in, sometimes as the closing lines of a chapter, sometimes as its own chapter at the very back, present in just the amount needed to show the author considered the question. It is there to establish that the book covered the topic, and nothing in it changes what the reader does next.
When we were writing our own book about AI, I sat in front of the same problem. I knew a section on ethics had to go in somewhere, and I found out how hard it is to get past a warning, to write something with enough consequence that a reader could not just turn the page. Many authors solve this by not trying at all. I would have liked that option too.
I’m not happy with my own result. Not at all. Why else did I spend that time writing one essay after another about it, always asking the same question and ending up in the same place? I have been wrestling with solutions for two years now.
A reader who finishes one of these sections walks away with the sense that the section has handled the ethical question, free to move on to the next chapter. The real questions are not in there. What happens to the people who lose their job to AI. Where do they go, who pays for their retraining. What about decisions nobody can trace any more, because nobody knows how the analysis arrived at its answer.
None of these questions fit in a paragraph. Answering any one of them properly would put the rest of the book in question, so they stay out of it. They’re in here because I’m not happy with my own work on the page where I was supposed to raise them.
I doubt I could have done it better. Two years of trying have not produced one I would call finished.