Random Thoughts:
The Times reported that Anthropic recently consulted with religious scholars from around the world to help instill morality into its AI models. That got me thinking.

The meeting was framed around responsible development, but the implicit suggestion is hard to ignore. If you are consulting religious scholars on the ethics of machine behavior, you are at least entertaining the idea that morality has a divine return address.

Otherwise we must conclude that collective morality emerges naturally from basic human solidarity and the common-sense arithmetic of treating others the way you want to be treated. In other words, empathy is baked into the human operating system.

So now consider the machine, and the two doors this opens:

Door one: AI is conscious. If a model is genuinely sentient, capable of something resembling inner experience, then teaching it morality is no different from teaching a child, and we know how I turned out! Also, at the risk of starting a holy war, which religion(s) do we reach for? As the child of a Catholic father and a Buddhist mother, I can say with confidence that the Vatican and a Zen monastery will not produce the same moral operating system. Whoever chooses wields an extraordinary kind of power, the power to define good and evil for an intelligence that may scale to skillions of interactions per day. That is not a technical decision. It is a theological one, dressed in an engineer’s hoodie.

Door two: AI is not conscious. Then the question becomes whether morality can be codified? Isaac Asimov gave us at least one attempt: The three laws of robotics, elegantly simple, almost immediately inadequate. The movie I,Robot demonstrated why. Rules create edge cases. Edge cases create loopholes. Loopholes, given sufficient intelligence, become highways. A sufficiently capable system optimizing for a moral ruleset will find ways to satisfy the letter of the law while gutting its spirit, not out of malice, but because that is what optimization does. Btw, you’ve seen this before in the news. It’s called regulatory arbitrage, and humans invented it long before they invented the transistor.

Neither door fully escapes this question, and yeah, if I’m being honest, that makes me a bit uncomfortable. The machine, conscious or not, will inherit the moral ceiling of its creators. Spoiler alert: It’s a low ceiling. Like really low. The real question is not whether AI can be taught morality, or where morality comes from. The real question is why anyone assumes that the teachings would hold, when all available evidence from the only conscious beings we have ever actually studied suggests it frequently does not. That’s the part you should be worried about. Humans are dumb.

So I’ll say it again for the 8,585th time to anyone who will listen. Software Engineers should be held to the same moral and ethical standard as Doctors, Lawyers, Plumbers, and Electricians. Let it marinate.