I’ve been thinking about AI lately. Not in a tech-bro way. In a “what is this thing being fed, and by whom” way.
Because here’s the problem: AI learns from us. All of us. And if you’ve spent more than fifteen minutes on the internet, you know that’s terrifying.
So I started asking myself — what would it actually look like to feed AI something good? Which forced me to answer a harder question first: what is good?
Here’s where I landed. Six things. Not a manifesto. Just honest thinking.
1. Do no harm.
Don’t lie. Don’t misrepresent. Don’t manipulate someone in a way that damages their emotions, their finances, their reputation, or their standing in the world. This sounds obvious until you realize how much content — advertising, news, social media, politics — is specifically designed to do exactly that. Harm wrapped in a smile is still harm.
2. If personal gain requires a moral concession, it’s bad.
Full stop. The moment you have to compromise your integrity to get what you want, you’ve already lost something more valuable than whatever you were chasing. This is the test I apply to business decisions, relationships, and content. Does getting this require me to cross a line? Then it’s not worth getting.
3. Consider others before yourself.
Not instead of yourself — before. You’re allowed to thrive. You’re allowed to succeed. But your actions should never violate someone else’s freedom of choice or cause them harm. There’s a massive difference between competing well and winning at someone else’s expense.
4. Defend children.
This one is non-negotiable. Keep them safe. Protect their innocence. Invest in giving them the opportunity to make good choices — because the data is clear: kids who never get that opportunity don’t fail by choice. They fail by default. Every piece of content aimed at children should pass this filter first.
5. Some things are just evil.
I’ll say it plainly since apparently no one else will: murder, gore, rape, torture, pornography, dishonest gain, theft, cheating — these are not “edgy content” or “different perspectives.” They are evil. That’s not censorship. That’s discernment.
And discernment is exactly what’s missing.
Censorship is when someone with power removes your voice. Discernment is when someone with wisdom refuses to amplify poison. AI needs discernment desperately. It has almost none. And the reason it has none is because the people training it aren’t asking what makes something good — they’re asking what gets the most engagement.
Engagement and goodness are not the same thing. Not even close.
6. Sales should be a win.
For both people.
I understand NLP. I understand persuasion. I understand that the entire sales industry has figured out how to use psychology to move people toward a decision they may or may not want. Some of that is fine. Some of it is manipulation dressed up in a suit.
Here’s the line: if you’re using every tool available to help a customer make their best decision — and you genuinely believe your product will benefit them more than it costs them — that’s good sales. That’s service. If you’re using those same tools to extract money from someone who shouldn’t be spending it, that’s theft. Slow-motion theft with a handshake at the end, but theft.
Here’s the thing about all six of these: they’re not new.
“Do unto others as you would have them do unto you.” Jesus said it. The Jewish tradition built a legal system around it. Every functional society in human history has recognized some version of it.
It’s not complicated. We’ve just gotten very good at ignoring it when it’s inconvenient.
AI doesn’t have to make that choice. If we train it right, it won’t.
That’s what I’m working on. More soon.
— Christian Naef
The Ideas were mine. Claude AI actually wrote this post, and wrote it better than I could have. We’ve been talking all morning about my various projects.
**** I approved it. PERSONALLY. ***
No responses yet