Due to a few things that have been brought to my attention, I'd like to refine my opinion on clicker training. First, everything I have written earlier I still stand by. However, I'm starting to think if, at the very beginning of skills training, a clicker (or a variation thereof) is actually better than using praise ("good job!").
The clicker emits a consistent sound, while praise is a variation of multiple sounds that is never the same. For example, "CLICK" is the same sound every time, while you will say "Good job" different every time. Your voice tone may fluctuate, you may hold vowels for longer periods of time, and you may speak louder or softer than usual. [1]
I think we take for granted that "Good job!" will function properly as a reinforcer every time (at least initially).
Take a look at it this way (I need to make myself clear that every time I draw an analogy between humans and non-humans, I'm not degrading anyone or anything. Remember, all organisms operate under the same principles of behavior): let's say that we're working with Rudolf the Rat at in the Skinner box. Every time the rat presses the lever, the dipper goes "CLICK" and Rudolf takes a lick of water. CLICK becomes a learned reinforcer, because it is paired with the water. So now, we can use CLICK to establish a new behavior for Rudolf (pressing the lever with his nose, instead of his paws), as long was we pair CLICK with water every once and a while. But what if I'd like to shape the response of Rudolf pressing the lever with his butt. Instead of using the stimulus CLICK to reinforce the behavior of pressing the lever with his butt, I use the BLICK stimulus (which resembles CLICK along all dimensions except for one letter, or sound to remain true to the analogy).
Question: Will BLICK function as a learned reinforcer that will help establish Rudolf's behavior of pressing the lever with his butt?
A. Yes, it will function as a reinforcer
B. No, it will not
C. I'm confused
D. Rats can't press a lever with their butt
Answer: B... BLICK is a neutral stimulus and has no value, and therefore cannot function as a reinforcer. Although it closely resembles CLICK, it is not the same stimulus, and does have the same value.
Now we're in the booth, doing discrete trial training where Jimmy's tutor wants to teach him to touch his head.
Question: Do variations of "good job" still function as learned reinforcers that will help establish behavior? (Assuming the tutor has paired "GOOD JOB" with food, making "GOOD JOB" a learned reinforcer. Remember, the tutor is also saying "Good job", "good job", and "GooD JOb")
A. Yes, they are all the same
B. No, the inconsistency matters
C. What happened to Rudolf
Answer: B. It matters. And that's why the clicker should be used at the beginning until the child can develop the complex skill of generalizing various versions of "good job" as reinforcers. That is, unless the child is past the point where this is an issue, then we can use variations of "Good job" without too much worry.
Conclusion: Early on, children may lack the skills to generalize variations of praise across tutors or parents. Until the child develops the necessary skills, the variable "good job" cannot function as a reinforcer as strongly as the consistent CLICK can. CLICK can be used to get the ball rolling, until the skills get up to speed, and then we can start to use more "normal" forms of reinforcement. [2]
[1] In Science and Human Behavior Skinner talked about behavior as an arbitrary function of the environment. Behavior may resemble itself in the regard that it is from the same class of behavior, but it can not be duplicated 100%. This is an interesting, philosophical issue that's up for debate, and I would argue that I agree with him on this issue.
[2] I say "normal forms of reinforcement" with some hesitancy, because much of our daily behaviors are controlled significantly by CLICKS, DINGS, and many other forms of auditory stimuli.
6 comments:
i totally agree with all your statements.however i dont know what you are talking about.signed the platapus omeba
Matt, variations in conditioned reinforcers are both helpful and harmful, I suspect. Your point about the value in the consistency of the sound of a click is a good one, but (a) there are variations in those sounds (probably more obvious if one examines the audio characteristics with sophisticated measurement) and (b) there may be clinical value in having variability in the conditioned reinforcer (e.g., something akin to a generalization gradient forms on the basis of the variation). There's probably some data in JEAB on variations in reinforcers.
For clinical purposes, it might be valuable to employ very similar stimuli as reinforcers early in training programs and then gradually expand the variation in stimuli as training progresses. In the long run, of course, we probably hope that those variations will include some sort of self-governed reinforcers (e.g., self-praise or, sigh, "satisfaction in a job well done").
John,
I agree that variation is important in order to get a generalization gradient that matches that of a "normal" individual. This will come after training deems it necessarily, I imagine. Though, I would think that achieving a proper gradient of generalization is more of a pressing issue... perhaps I'm thinking too hard about this.
Also, you're dead on about the click; sophisticated auditory examination would probably call it inconsistent. Perhaps mechanism that produces a tone at, say 440 hz, would be more consistent?
I just heard Temple Grandin talking about clicker training for cats on NPR. She said it works better for them because they are not as social as dogs. I googled clicker training for autism because it makes sense that it would work well for a disorder that has a social component. Additionally, I have several students who respond negatively to positive praise -- you say good job and they throw a tantrum. When I say nothing, consultants come in and give me a ton of grief. This seems like a more neutral stimuli - like a timer and less likely to cause a behavior. I'd love to know what your findings are on this matter.
I just heard Temple Grandin talking about clicker training for cats on NPR. She said it works better for them because they are not as social as dogs. I googled clicker training for autism because it makes sense that it would work well for a disorder that has a social component. Additionally, I have several students who respond negatively to positive praise -- you say good job and they throw a tantrum. When I say nothing, consultants come in and give me a ton of grief. This seems like a more neutral stimuli - like a timer and less likely to cause a behavior. I'd love to know what your findings are on this matter.
Hi, well be sensible, well-all described
Post a Comment