The Logic of Aversive Control

The question

I've been listening to Stewart Hilliard while I work, and one idea kept following me home: the force-free position says it's backed by science, but when I trace the logic, it doesn't hold together.

This morning in the shower I tried to work it out the way I used to work through code. What goes in, what comes out, and what decides the difference. This entry is where that landed. I'm not writing it to win an argument. I'm writing it because Gunner and I are still early, and I want to understand the tools before I ever pick them up.

The model: input in, behavior out

A dog learns by comparing results. Every repetition runs the same loop: a situation comes in, the dog decides, a behavior comes out, and the world answers. That answer shapes the next decision.

The answer can only be one of three things:

• Good. Something the dog wants arrives, or something it doesn't want goes away.
• Bad. Something the dog doesn't want arrives, or something it wants is taken away.
• Nothing. No change.

That's the whole system. Everything else in this entry follows from it.

Two ways to write the rule

If you write training as a simple rule, purely positive looks like this:

• IF right → reward
• ELSE → nothing

A balanced approach looks like this:

• IF right → reward
• IF wrong → cost
• ELSE (still working it out) → nothing, or help

The second rule is stronger for three reasons.

It has more resolution. In the first rule, "wrong" and "not yet" produce the same result. The dog can't tell them apart. In the second, each repetition carries more information.

"Nothing" is ambiguous. No reward could mean wrong, not now, or nobody noticed. A clear cost answers the question directly: not that.

It contains the first rule. A balanced trainer can run the positive-only version whenever it fits the dog or the stage. Purely positive can't add the missing branch. One is a full toolkit; the other is a subset with a rule against expanding it.

The world adds one more complication. The handler isn't the only one handing out results. A squirrel chase is a huge "good," and a handler can't always outbid it with a better treat. A calibrated cost on the competing choice changes the math without an endless escalation of rewards.

Having the cost branch doesn't mean using it often. Good balanced training uses it rarely and precisely.

What makes a cost fair

A cost only teaches when the dog can use it as information. Two conditions decide that.

• Predictable. Something warns the dog it's coming.
• Controllable. The dog's own behavior turns it off or prevents it.

When both are true, pressure becomes a problem the dog can solve. It starts as escape (pressure on, the right behavior turns it off) and becomes avoidance (the dog answers the cue, and the pressure never comes). The cue ends up doing the work. When either condition breaks, the dog stops solving and starts coping. Remove both, and you get learned helplessness: nothing the dog does matters, so it stops trying.

The honest limit

Results don't only change what the dog learns. They change how the dog feels. A cost that's too large or unpredictable overwhelms the dog, and it stops thinking. Every cost also attaches a feeling to whatever predicted it: the handler, the place, the cue.

This is the force-free side's real concern, and it's a fair one. But it argues for skill and calibration, not for removing the branch entirely.

Values, not science

The science clearly supports reinforcement. It works, and it should be the foundation. What the science doesn't establish is that costs should never be used. That second step is a values judgment, and it gets presented as if it were a finding.

That's the core problem. "Does it work?" is a question for evidence. "Is it justified?" is a question of ethics. Both are worth asking, but they shouldn't be blended into one claim.

The definition doesn't hold up cleanly either. Withholding a reward is a cost. Frustration is a cost. So force free doesn't remove costs from the dog's experience. It limits which ones the handler is allowed to use.

Force-free trainers would say their "else" isn't empty. They use no-reward markers, redirection, and management. That's fair, but every one of those still sits in the "nothing" or "less good" range. None of them gives the dog a clear not that.

Hilliard describes the two camps as idealistic practitioners and pragmatic practitioners. The idealist starts from what training should look like. The pragmatist starts from what this dog needs to learn reliably. I think that's the most accurate description of the divide I've heard.

Where Gunner and I are now

Gunner isn't working with pressure yet. Right now we run the simple rule: reward what's right, help with what isn't. That's the correct tool for where he is, and for where I am.

The difference is that I'm choosing it, not bound by it. When he's ready, and when I've earned the skill to apply a cost fairly, the other branch is there. Until then, it stays in the toolkit, unused but understood.

Removing costs by rule doesn't make learning kinder or clearer. It makes the dog's information incomplete.

Balanced training isn't the opposite of positive training. It's positive training with the full set of branches.