AI is not a teenager
A senior AI exec told me we need to 'teach AI ethics as it grows.' That framing is wrong — and it matters.
Last week in Bangkok, a senior exec responsible for AI strategy told me: "AI is like a teenage child. We need to teach it ethics as it grows."
I understood the sentiment. But the framing is wrong — and the distinction matters.
LLMs are not children
A language model is a mathematical function trained on a snapshot of text. It does not grow. It does not learn from experience. It does not internalise values over time the way a person does.
When you run a prompt, you get a weighted probability distribution over tokens. The model does not remember the last time it was wrong. It has no concept of regret. The "teenager" metaphor anthropomorphises the mechanism in a way that leads to bad decisions.
Why the metaphor leads you astray
If you believe AI is maturing — that it will get more responsible as it gains experience — you'll underinvest in the controls you need today. You'll treat governance as scaffolding to remove later. You'll wait for the model to "grow up" rather than building the systems that make its outputs predictable and auditable now.
Companies that have got this right did not try to raise their AI. They built a controlled operating environment around it: explicit boundaries, defined escalation paths, an immutable record of every decision.
You do not need to teach AI ethics. You need to build systems in which unethical outcomes are structurally impossible — or at minimum, detectable and traceable.
The companies that win
The exec I spoke to works for a large bank. They are running fifteen AI pilots simultaneously. Productivity numbers look good. Audit trail: non-existent.
The companies that will win are not the ones with the most AI running fastest. They are the ones that can prove what their AI did, why, and what the outcome was — on demand, to a regulator, to a customer, to themselves.
AI doesn't grow up. It scales. Treat it like software.
Build the controls before you scale, not after. Because when something goes wrong at scale, "we were still teaching it" is not an explanation anyone will accept.
