AI isn't biased. It has your bias

“Human beings are simultaneously the problem and the only possible solution.”

Yesterday in a group chat, a colleague shared an article about an AI that learned to cheat on its own safety evaluations. On its own. Without anyone asking it to.

The debate that followed was the usual one:

∙ The ones whose minds jump straight to The Matrix or Skynet

∙ The ones who love AI and can't live without it

∙ And the ones who just don't know what to think

In the middle of all that, Mauricio Gutierrez shared a philosophical conversation he'd had with his AI — the kind that makes you think more than you expected. And that's where the quote at the top came from.

Meanwhile, I was still laughing at an article a magazine published revealing that AI responds differently to women than to men. As if AI discriminated by sex.

And all I could think was: these are the same people who say good night to their AI.

After years of analyzing patterns in human behavior, none of this surprises me.

AI isn't biased.
It has yours.

A human trained it, with their data, their categories, their blind spots. And a human also decided what went into its knowledge base. Guess what happens when whoever trains it has their own biases, their own perspectives, their own way of seeing the world.

In computing there's a phrase that sums it up perfectly: shit in, shit out. What you give it is what it gives back.

It's not malice. It's fidelity. A knife works just as well for a chef as for a murderer. The problem was never the knife.

People don't respond to reality either — we respond to our interpretation of it. AI does the same.

It just doesn't realize it.

That “doesn't realize it” is the only thing that separates us...
For now. Although at the rate it's moving, tomorrow I might be apologizing to my AI for writing this.