Rethinking AI Ethics: Why Core Values Must Come First

Regulation can't keep pace with AI. Laws take years to draft and pass; a new model ships every few months. If you've felt a low-grade unease watching this unfold, I think you're right to — not because AI is unmanageable, but because we're managing it from the wrong end.

Here's the reframe I keep returning to: what if the fix isn't faster regulation, but a different starting point — not "what are the rules" but "what do we actually value." That's the harder question, and the one that scales. So: what are these core values, and how do we build them into a system with no lived experience of consequence?

Why start with values instead of rules

Values sit underneath decision-making, well before ethics or law get involved. They're the quiet filter deciding what you notice, what you prioritize, what you'll defend even when it costs you something. Any serious model of values-based decision-making starts here, because rules are downstream of something more stable.

Ethics, by contrast, is slippery — it shifts by culture, era, and the room you're standing in. Law is even more local, a snapshot of what one jurisdiction agreed on, usually years after the fact. Values like human flourishing, integrity, and sustainability sit at a different altitude: abstract enough to survive translation across cultures, concrete enough to guide a decision.

An AI system built on that foundation operates from purpose rather than restriction. Constraints keep a system busy being told what it can't do; values give it something to move toward. One produces compliance. The other, partnership.

What a values-first architecture might look like

Picture three layers, stacked:

1. Core values layer — the foundational principles: human flourishing, integrity, sustainability. Slow-moving, deliberately so.
2. Ethical layer — a contextual filter checking whether an action fits current social norms.
3. Judicial layer — the compliance check against law and liability.

Notice the direction of authority: values inform the ethical layer, which refines what reaches the judicial layer, never the reverse. A decision earns its legality by first being sound at the level of values and ethics — the difference between an AI asking "can I get away with this" and one asking "should this happen at all." That would need something like a shared values framework: a common reference point AI systems check decisions against, rather than each one improvising its own reading of "human flourishing."

The hard problems no one gets to skip

A values-first approach doesn't remove the hard questions — it just moves them earlier, where they're easier to see.

Who gets to decide?

Somebody has to define which values count, and revise them as they stop fitting. Democratic process sounds appealing, but global consensus is a genuinely enormous ask — and there's a sharper danger in whichever culture builds the system writing its own assumptions into what gets called "universal."

Should the AI change its own values?

Human values shift with age, era, and circumstance — what mattered to you at twenty-five rarely matches fifty. A system frozen with one static value set risks becoming obsolete, or harmful, as the world moves and it doesn't. Letting it revise its own values sounds like the fix, but that's the part that should worry you most: a system that can rewrite its own foundation can drift somewhere nobody chose.

What if the framework gets captured?

A shared values framework becomes a single point of failure the moment it exists — whoever controls it controls what every connected system treats as good. Real safeguards here aren't optional. They're the whole ballgame.

What actually gets better if we do this well

None of that is a reason to abandon the idea — it's a reason to build it carefully.

A shared foundation gives systems built by different companies common ground to reason from, with fewer misfires from each one inventing its own definition of "fair." It orients AI toward human flourishing rather than the nearest proxy for it, like engagement or efficiency, which quietly become the actual goal whenever nobody specifies something better. It weighs decisions against their impact on people and the planet, not just the next quarter. And it builds trust: transparent values are values people can reason about, not guess at.

Where this actually starts

Not with a lab, and not with a regulator. With a conversation — one that needs philosophers, technologists, ethicists, and ordinary communities in the same room. Superintelligence by Nick Bostrom and Human Compatible by Stuart Russell are worth your time here, and groups like CAIS and OpenAI are already doing serious work on the safety side.

From there, the work is architectural: systems that learn within a values framework without drifting from it, and oversight that catches drift before it compounds. Unglamorous work, and the kind that holds up.

The bet worth making

Here's the wager, stated plainly: a system built on values, not rules alone, will default toward decisions that serve people and the planet — not because it's forced to, but because that's the shape of its foundation. Start there, and you sidestep much of the reactive scrambling that comes from legislating your way out of problems after the fact.

AI will keep shaping how we live, work, and relate to each other, whether we've thought this through or not. The question isn't whether we can stop that. It's whether the intelligence we're building reflects not just what we can do, but what we actually value.