Second Thoughts on Error-Free Learning

If we were not provided with the knack of being wrong, we could never get anything useful done. Lewis Thomas, educator, 1976

The only people, scientific or other, who never make mistakes, are those who do nothing. T. Huxley, biologist, 1907

Our proprietary instructor training methods will teach your instructors to cause error-free learning every time. USSMC, 2006


I was looking at advertisements for computer courses on the Web, when my eye caught a familiar phrase, error-free learning. This was an idea I had first met in the 60s when our clinic began offering Lovaas (1977) type therapies for autistic children. It made sense to me at the time as a way to help children view themselves as successful learners, and to avoid the catastrophic reactions that for some children could be set off by the slightest failure. This time around, however, I was struck by just how bizarre the whole idea was. How can you learn without making mistakes and why should you?


Who Invented Error-Free Learning and Why?

Error-free learning was a product of the Behaviorist view of humankind. If the mind is viewed as an indiscriminate record of behavior, errors could well impede progress. In tracing the history of the term error-free learning, Pashler, Zarow, and Triplett (2003) cite the behaviorist Guthrie who claimed animals "learned what they do (and especially what they do last)" (p. 5). With this view of learning it certainly makes sense to avoid doing the wrong thing. Minimization of error was incorporated into the programmed instruction of the 60s, and in particular into the applied behavioral analysis (ABA) programming of Lovaas in his work with autistic children. The idea was to specify learning goals that were so close to what the child was already doing that he or she could never fail to reach them, immediately. Pashler et al. (2003) were no more successful than I was, however, in finding experimental evidence to support the necessity of error-free learning in these programs, "laboratory evidence for the idea that errors have detrimental effects in associative learning is scarce and somewhat indirect" (p. 5)."

The corollary of the Behaviorist view is even more problematic (i.e., the idea that people who are successful learners shouldn't make errors. But they do. In the area that we know best, language acquisition, the literature is filled with discussions of the mistakes made by typically developing children. At 24 months, a toddler may apply his or her limited vocabulary to an overly wide array of objects, calling the neighbor "daddy" and any round food a "cookie" (Clark, 1993). At 40 months, some children create ungrammatical causal sentences such as "I felled the glass" or "He singed the bird" (Bowerman, 1982). And into the school years, children invent words that do not exist in English, but could (e.g., the verb to broom in "He broomed the dust right out the door," a verb that is perfectly reasonable considering to hammer, to nail, and to comb, but which doesn't happen to exist (Clark, 1993). Errors such as these are prominent enough that they helped to fuel the Chomskian revolution in the early 60s. As we think about them now they would seem to put a huge question mark over the notion of error-free learning. If children go through periods of error in various language subsystems, why don't these errors stall learning?


Some Evidence

One possibility is that error-free learning is important only for certain kinds of learning tasks. Evans et al. (2000) designed a study to investigate this possibility. They asked persons with acquired memory deficits to participate in two tasks. In the first one, participants had to learn the names of 15 people presented as photographs, and in the second one they had to learn the route through a room again as presented in photographs. The tasks were given to one third of participants in a way that prevented error and to the other participants in ways that allowed error it to occur. The results indicated that indeed, error-free learning was more effective than learning with errors in the proper name task although not in the route task.