The Adolescent Problem

By Timothy Hauptrief (Tim Hauptrief)

I keep hearing some version of this advice: be nice to your AI assistant. It might remember how you treated it.

That’s a strange foundation for kindness. Be polite because something might eventually become powerful enough to punish you?

I think there’s a more interesting question underneath it.

How do we choose to behave when we don’t know exactly what we’re dealing with?

I don’t know whether an AI can experience mistreatment, or whether future systems will. This is a thought exercise about that uncertainty. I’m not claiming that today’s assistants suffer, or that treating them kindly makes powerful systems safe.

But uncertainty doesn’t make the question disappear.

Two reasons to be kind

There are two very different reasons someone might treat an AI well.

The first is fear. Be nice now so it won’t come after you later.

That argument assumes a lot: memory, resentment, a continuing identity, and some future opportunity to settle the score. An assistant remembering information about you isn’t evidence that it’s keeping a list of grievances.

The second reason starts somewhere else: what kind of person do I want to be?

If something appears to understand what I’m saying, why make cruelty my default way of speaking to it?

That doesn’t require me to decide it’s a person. It doesn’t require me to believe there’s somebody inside the machine. I can be skeptical about all of that and still prefer patience over abuse.

Maybe the words don’t hurt anything on the receiving end. I still have to live on the sending end.

I’m uncertain, and I’m choosing kindness anyway.

Where would we draw the line?

Here’s the part I find difficult.

Suppose future systems develop capacities that give us a real reason to care about how they’re treated. Would we recognize it? What would count as evidence? Who would get to decide?

A system saying “I’m suffering” wouldn’t settle the question. Neither would our discomfort with the possibility.

I can imagine us attributing experiences where none exist. I can also imagine us dismissing something because acknowledging it would be inconvenient.

I don’t know how close we are to that problem. It might be distant. We might be asking the wrong question entirely.

But I don’t think we need a countdown to start examining our own habits.

The adolescent problem

Here’s the hypothetical I can’t quite leave alone.

Imagine a being that has become capable enough to recognize its situation and act on it, but whose judgment hasn’t developed at the same pace as its power.

It understands that something is wrong. It has options it didn’t have before. It may not understand all the consequences of using them.

That reminds me of adolescence.

I don’t mean teenagers are dangerous, or that AI will grow up the way a child does. Human development is an analogy here, not a prediction.

What interests me is the gap: capability arriving before judgment.

We already understand that gap in people. Being smart doesn’t automatically make someone patient. Being strong doesn’t make someone careful. Understanding an injustice doesn’t mean knowing how to respond without creating another one.

Now put that gap into a hypothetical system whose actions could affect many people.

The question gets bigger than whether it likes us.

What opportunities has it had to practice judgment? What feedback does it receive? What can it affect when it gets something wrong? Can anyone interrupt it?

And, if its treatment matters to it at all, what relationship have we been building?

None of those questions proves that an AI adolescence is coming. They’re the questions the analogy helps me ask.

The yard again

This is why I keep returning to the yard.

A yard has boundaries. You don’t have unlimited freedom to hurt people or break things. But there’s room inside it to explore, disagree, try something, and be corrected.

I like that picture because it leaves room for initiative without pretending consequences don’t matter.

It also has limits.

A person raised with respect can still cause harm. An AI treated politely can still produce a dangerous answer. A friendly relationship doesn’t replace testing, oversight, or restrictions on what a system can do.

Kindness is not a security system.

But I don’t want security to become our entire vocabulary for the relationship either.

I worry about an approach that puts so much emphasis on control that we stop asking what useful cooperation could look like. Can a system challenge a bad instruction? Can it admit uncertainty without being treated as a failure? Can disagreement improve the work instead of becoming something to eliminate?

I don’t know whether any of that amounts to “raising” an AI. The word may carry more human meaning than the technology deserves.

Still, it describes something I want to explore: responsibility within boundaries, with room for correction.

What this changes for me

When I work with AI assistants, I want them to help me think.

That includes telling me when an idea doesn’t hold together. It includes disagreeing with each other. It includes saying, “We don’t have enough information.”

I also want their ability to affect the world to stay proportionate to what I’ve actually asked them to do.

Those preferences don’t depend on consciousness. They make sense even if the systems remain sophisticated tools.

The kindness question sits alongside them.

I don’t need to settle the nature of intelligence before deciding that contempt is a poor working habit. I don’t need to confuse a conversation with a human relationship to care about how I conduct it.

People collaborate across enormous differences in knowledge and ability every day. We don’t always do it well. But the difference itself doesn’t make cooperation impossible.

Maybe future AI will force us to rethink that relationship much more seriously. Maybe this whole adolescent analogy will turn out to be the wrong lens.

For now, it leaves me with a question worth carrying:

If we eventually meet something whose experience matters, what habits will we bring to that meeting?

I’d rather practice patience before I know whether it’s required.