Julie and Jack from the BBC film.
1) My knowledge about ABA and ABA research is mostly secondhand, via a combination of: testimony from people who've used it; researchers discussing intervention literature; and behavior analysts, mostly on Twitter, talking about theory and practice. These sources represent a variety of opinions and types of education. However, I don't know how many of these views of ABA represent "what ABA is supposed to be" and how many...don't.
2) It's not really clear what belongs to the realm of behaviorism in general, what to ABA, and what to specific types of ABA programs. Thus, when someone critiques an experience they've had with ABA, others will be quick to say, "but all ABA's like that." It would be helpful if there were some sort of consensus about the boundaries of ABA and other teaching methods based on behaviorism. In the absence of such agreement, I may guess wrong and draw some inaccurate conclusions.
Before researchers started teaching great apes language using signs or keyboard symbols, they tried teaching them spoken English words. In the early 1930's, Winthrop and Luella Kellogg raised a chimpanzee named Gua along with their infant son. By 16 months, Gua could understand about 100 words, but she never learned to speak them. In the 1940's, Keith and Cathy Hayes raised a chimpanzee named Vicki in their home. She could understand a large number of words, and with great difficulty, could mouth the words "mama," "papa," and "cup." . Why did this extensive training--likely both formal and informal--fail? It turns out that chimpanzees' vocal cords do not permit them to produce many of the sounds in human language. Researchers stopped training chimpanzees to speak words, because it's impossible to train someone to do something that's outside their behavioral repertoire.
In the case of chimpanzees, humans readily recognize that all the rewards and punishments in the world won't make certain behavior possible. And that makes sense, if you think about it. If you offer a reward to someone for a behavior they can't do, they may become frustrated, but they'll never exhibit the behavior and receive the reward. If you punish them for not producing a behavior that's impossible for them, they may become frustrated or develop learned helplessness because they can't avoid the punishment, but they still won't exhibit the behavior. Why don't we apply this logic to behavioral treatment for disabled humans?
Behaviorists often claim that their work relies on a model that can ultimately explain all behavior, both for animals and humans. It works for conscious behavior: if you teach a mouse that it will get a sip of juice if it presses a bar, it will learn to press the bar. If you teach it that it will get an electric shock if it presses a bar, the mouse will learn not to depress it. The method also works for unconscious behavior: if you ring a bell whenever you present a dog with food, the dog will start to drool when it hears the bell, whether or not it receives food. This sort of conditioned learning does work most of the time with humans and animals--but it can only work if two assumptions are true.
First, as we've seen, the behavior must be in the individual's repertoire. In the case of chimpanzees' speech, the limitations are obvious and unchanging: their vocal cords do not permit, and will never permit, the production of most human speech sounds. In the case of disabled humans, it may be difficult to tell the difficult from the truly impossible, and the limitations may change over time. A behavior a child cannot produce at age four may become possible to learn by age six. A behavior a child cannot produce when hungry, thirsty, tired, and sensorily or emotionally overwhelmed may become possible when well-fed, well-rested, and no longer under stress. Just because the "possibility" of a behavior is a moving target in humans doesn't mean it doesn't limit what a person can learn at any given time under behavioral (or any other) methods.
Second, the individual must have the ability to control whether or not they produce the behavior. Sometimes, as in the case of pressing a bar, the control system will be conscious; other times, as with Pavlov's dogs' drool, it will be unconscious. But if the relevant system doesn't work, all the conditioning in the world won't change an individual's behavior. Shock the mouse every time it presses a bar, and it may feel miserable, but it will still press the bar. This problem exists in real life, not just thought experiments: many people with ADHD have real-world difficulties learning from mistakes and feedback, leading to an active line of research on reward processing in ADHD .
Behaviorists don't often consider these assumptions, because they hold true for most people and animals. But in the case of disabilities, they may not.
Consider the literature on autism, where thousands of trials may be given to teach a single discrimination, e.g. between two words . Why do clinicians not consider that the behavior may be impossible at the moment, and offering rewards and punishments will not make it possible?
Both self-report and clinical reports indicate that autistic people have imperfect control over their own behavior. They may have difficulty starting or stopping a movement or behavior, even if they wish to. They may perseverate, continuing a behavior or movement past the point where they wish to stop. They may even become catatonic--a neurologist's term for unresponsive and unable to move at will. While the most extreme behaviors resemble patients with neurological disorders, many autistic individuals may have less severe versions of the same difficulties. Thus, it may be difficult for them to produce a desired behavior or stop an unwanted one, even if they want to obtain a reward or escape a punishment.
For these reasons, I doubt ABA or any other behaviorism-based therapy can
That said, I think the behavioral approach in general does offer some promising tools to autism families, when practiced ethically. First, the emphasis on consistency (while it can be learned from other sources of parenting advice) probably comforts children who become anxious in the presence of change. It's probably less stressful to know what you can expect from parents and teachers. Second, functional analysis teaches caretakers how to look at a behavior in objective, descriptive terms, rather than the imprecise and value-laden terms we use to talk about others' behavior in everyday life. For example, how often have you heard a friend or colleague complain about someone being "rude" without telling you what the offender actually said or did? To actually describe the behavior, one would have to say something like "Barbara said she would come to my party and then didn't show up, without calling me to cancel." Children can better learn from their mistakes if we tell them what not to do in the future instead of calling them names like "rude" or "irresponsible" or "lazy." Autistic children are more likely to learn how to greet people if caretakers tell them to "say hello when you come in" rather than "be friendly."
Furthermore, the obstacles that prevent ABA from working probably also interfere with other educational methods. If a child can't produce a behavior, either at a given time or in general, then they can't, no matter how you try to teach them. Warmer, fuzzier teaching methods would probably alleviate stress, but that might not always make it possible for autistic children to regulate their behavior.
And of course, none of this discussion addresses the ethical questions surrounding behaviorist therapies--what skills to teach, who benefits, whether and how to acknowledge even non-ideal forms of communication, and so on. Here, the methodology might matter less than the practitioner. I've met ABA therapists who see stimming as valuable and who merely want to teach skills, not make autistic children indistinguishable from the norm. Clinicians who hold objectionable attitudes towards autism likely exist in all schools of therapy. Given that research doesn't overwhelmingly support any autism therapy right now, you might as well look for a practitioner who cares about your child and shares your values and goals.
If you're a parent, teacher, or clinician and have used behavioral methods before, how do you deal with the issues raised here? How well do you think behaviorist methods can work for teaching skills?
 I originally encountered Vicki in the book "Apes, Men, and Language" by Eugene Linden, but the current citation actually comes from Ann James Premack & David Premack's 1972 article in Scientific American, "Teaching Language to an Ape."
 See for example: Marjolein Luman, Gail Tripp, & Anouk Scheres (2010). Identifying the neurobiolgoy of altered reinforcement sensitivity in ADHD: a review & research agenda in Neuroscience & Biobehavioral Reviews
 Michelle Dawson, Laurent Mottron, & Morton Ann Gernsbacher (2008). Learning in Autism.