The science of errorless learning and why your dog's mistakes might be your fault, not theirs.
There's a deeply held belief in animal training — and in human education, for that matter — that errors are part of the deal. You try, you fail, you learn. Struggle builds character. Frustration is just the tuition fee for knowledge.
But what if that's wrong? What if errors aren't a feature of learning, but a symptom of poor teaching design?
That's the uncomfortable implication sitting at the heart of a concept called errorless learning — and the science behind it is hard to argue with.
The Pigeons That Changed Everything
In the 1960s, a researcher named Dr. Herbert Terrace set out to teach pigeons a relatively simple discrimination task. When a red light appeared, peck the button. When a green light appeared, don't.
Straightforward enough. But how you get there turns out to matter enormously.
In the traditional approach, pigeons would first learn to peck for the red light, then the green light would be introduced as a signal that pecking wouldn't pay off. Errors would accumulate. The pigeon would peck the green light, nothing would happen, peck again, nothing — the classic extinction process playing out in real time, with all the frustration and emotional noise that accompanies it.
Terrace did something different. Rather than introducing the green light at full intensity as a sudden contrast to the red, he brought it in gradually — dim at first, barely perceptible, sometimes shown only briefly or at a slightly different wavelength. So faint, in fact, that the pigeons were unlikely to respond to it at all. Over time, as correct responding became consistent, the green light was slowly made more prominent until it resembled its full form.
The result? A dramatic reduction in errors. Not a small improvement — a dramatic one.
But here's the part that really matters.
In his 1963 paper, Terrace described something striking about the traditionally trained birds. The green light — the signal for "don't do that" — had acquired aversive properties through the extinction process. These pigeons would actually turn their heads away from the button when the green light came on. They didn't just learn to not respond; they learned to flinch. And Terrace found this aversiveness persisted long after the extinction was technically complete.
The errorless learners? The green light was just a neutral stimulus. No avoidance. No emotional residue. They'd learned the same skill, just without the baggage.
There's a deeply held belief in animal training — and in human education, for that matter — that errors are part of the deal. You try, you fail, you learn. Struggle builds character. Frustration is just the tuition fee for knowledge.
But what if that's wrong? What if errors aren't a feature of learning, but a symptom of poor teaching design?
That's the uncomfortable implication sitting at the heart of a concept called errorless learning — and the science behind it is hard to argue with.
The Pigeons That Changed Everything
In the 1960s, a researcher named Dr. Herbert Terrace set out to teach pigeons a relatively simple discrimination task. When a red light appeared, peck the button. When a green light appeared, don't.
Straightforward enough. But how you get there turns out to matter enormously.
In the traditional approach, pigeons would first learn to peck for the red light, then the green light would be introduced as a signal that pecking wouldn't pay off. Errors would accumulate. The pigeon would peck the green light, nothing would happen, peck again, nothing — the classic extinction process playing out in real time, with all the frustration and emotional noise that accompanies it.
Terrace did something different. Rather than introducing the green light at full intensity as a sudden contrast to the red, he brought it in gradually — dim at first, barely perceptible, sometimes shown only briefly or at a slightly different wavelength. So faint, in fact, that the pigeons were unlikely to respond to it at all. Over time, as correct responding became consistent, the green light was slowly made more prominent until it resembled its full form.
The result? A dramatic reduction in errors. Not a small improvement — a dramatic one.
But here's the part that really matters.
In his 1963 paper, Terrace described something striking about the traditionally trained birds. The green light — the signal for "don't do that" — had acquired aversive properties through the extinction process. These pigeons would actually turn their heads away from the button when the green light came on. They didn't just learn to not respond; they learned to flinch. And Terrace found this aversiveness persisted long after the extinction was technically complete.
The errorless learners? The green light was just a neutral stimulus. No avoidance. No emotional residue. They'd learned the same skill, just without the baggage.
This Isn't Just About Pigeons
Errorless learning has since been studied across a wide range of species and contexts — including rehabilitation settings for humans recovering from traumatic brain injuries. The principle holds: reducing errors during the acquisition of a skill leads to more effective learning and fewer negative emotional side effects.
For dogs, the implication is significant. Every time we set up a training scenario that produces repeated failure — asking for too much too fast, introducing a distraction before the foundation is solid, or relying on the animal to "figure it out" through a fog of frustration — we're not just accepting errors. We're potentially adding aversive emotional associations to the very stimuli and contexts we want our dogs to feel good about.
Waiting Their Turn: A Real-World Example
Here's a practical illustration of what errorless design can look like in dog training, drawn from an example offered by Hannah Branigan on her podcast Drinking from the Toilet (episode 178, 2023).
Imagine you're training a puppy and you need them to wait in a crate while another dog gets a session. The trial-and-error approach looks like this: put the puppy in the crate, train the other dog, ignore the barking and crate-biting, and return to the puppy once they've settled. It works, eventually. But it works through extinction — and extinction, as Terrace showed, leaves marks.
The errorless version looks quite different.
You put the puppy in the crate, then turn your back — barely. So briefly that the puppy almost can't register it's happened. Then you immediately turn back, take the puppy out, and resume training. That's the whole loop, at first. You've given the crate context meaning: facing away means stay calm, facing toward means training resumes.
From there, the duration of "turned away" grows incrementally. Maybe next you turn away and take a small step. Then turn away and offer another dog a treat. Then a short session with the other dog. Each increment is small enough that the puppy doesn't tip over into distress — errors stay minimal, the emotional climate stays calm, and the learning stacks cleanly.
It's elegant. And it's a direct application of the same logic Terrace used with his pigeons sixty years ago.
A Credit That Often Gets Missed
Here's a piece of history that tends to get dropped in the retelling: Dr. Terrace didn't coin the term "errorless learning." That distinction belongs to B.F. Skinner, who introduced the concept back in the 1930s.
Skinner's position was direct: errors are not necessary for learning. When they occur, they are the result of poor program design, not an inevitable part of the process. Terrace's experiments were an influential application of those ideas — but the foundational thinking came earlier.
It's worth sitting with that. Poor program design. Not a weak learner, not an uncooperative dog, not an animal that just "needs to experience failure to learn." The errors we see in training are often a message about the structure of the training itself.
What "Errorless" Actually Means in Practice
One fair criticism of the term is that it sets an impossible standard. Genuinely zero errors, across every session, for every learner? That's not realistic, and chasing perfection in that sense would itself be a kind of poor program design.
The more useful framing — and the one consistent with both Skinner's and Terrace's actual positions — is this: the goal is to minimise errors and maximise the learner's chances of success. Good planning. High rates of reinforcement. Criteria introduced gradually enough that correct responses are almost inevitable before the difficulty increases.
What we're designing against is the accumulation of errors that leads to frustration, emotional shutdown, or the slow acquisition of aversive associations. Not because we're precious about our training aesthetics, but because the science tells us it gets in the way of learning.
A session where your dog is mostly getting it right, reinforcement is flowing, and the skill is quietly solidifying — that's not easy mode. That's good program design.
Want to see these ideas in action? Watch the full video on errorless learning on the Train Me Please YouTube channel:
References:
Anderson, E. (2013) Errorless Learning II. eileenanddogs - What my dogs teach me. Available at: https://eileenanddogs.com/blog/2013/02/09/errorless-learning-skinner/ (Accessed: 15 August 2024).
Branigan, H. (2023) ‘178: How to Apply Errorless Learning Principles in Practical Training Sessions’, Drinking from the Toilet: Real Dogs, Real Training. SoundCloud, 22 July. Available at: https://soundcloud.com/hannah-branigan/178-how-to-apply-errorless-learning-principles-in-practical-training-sessions (Accessed: 15 August 2024).
Skinner, B.F. (1968). The technology of teaching. Acton, Mass.: Copley Pub.
Terrace, H.S. (1963) ‘Discrimination learning with and without “errors”’, Journal of the Experimental Analysis of Behavior, 6(1), pp. 1–27.
RSS Feed
