Need Learning Theory Nerd

[corvus]

I have a burning, burning learning theory question.

If a behaviour appears to be under stimulus control, (for, say, approximately 90 trials over 3 days) can increasing the reward value (not frequency) cause the behaviour to drift and the stimulus control to slip? I.e., behaviour is no longer under stimulus control.

[Mrs Rusty Bucket]

random jackpot payouts are supposed to be more effective at increasing a behaviour and maintaining it in the absence of any reward than reliable vending machine payouts. Ie the dog that is used to a vending machine payout will quit trying / responding sooner than one accustomed to random payouts.

The trick is figuring out and planning the transition from a vending machine to pokie machine. I had a link for that somewhere. Hmm.

Sorry for the non-geekness of the explanation.

Is that what you are talking about? I thought you were the learning theory/science nerd around here.

[corvus]

Not quite. The reward rate has been constant. The volume just went from base level to lower, then back up to base level again. The behaviour was maintained in one dog on the lower reward volume, but only semi-maintained in the other (about 50-90% response rate to cue). When the reward volume went back to the base level, both dogs started performing the behaviour when not cued as well. I'm wondering if learning theory can explain this, because I'm really excited about the alternative and I need someone less excited to be rational about it.

My learning theory knowledge is just the basics, really. I never did psychology. There is a huge mountain of data about this stuff, but it's so hard to find anything specific in it!

[Mrs Rusty Bucket]

I'd say one dog was more persistant with a broken vending machine than the other.

How long would you keep putting in the money for a drink when you didn't get the drink. Some people let it eat their money once, and some people try more times (eg maybe only once today and again tomorrow and one more time the day after).

My dog was getting a pretty reliable payout at the bottom of the dog walk even if I wasn't there that she started checking the end of the dog walk before she would do any of the rest of the course. Sigh. I've stopped the treat when I'm not there completely but she still goes and checks occasionally. I guess I'd still check the chocolate shelf at my mum's place if I wanted a chocolate... tho that's a pretty reliable place for chocolate.

Edited September 23, 2011 by Mrs Rusty Bucket

[corvus]

I'd say one dog was more persistant with a broken vending machine than the other.

Yes, but it's what happened when the vending machine got fixed that interests me. Suddenly they were whacking away at all the buttons rather than just the one that gives them the good stuff.

[Mrs Rusty Bucket]

Ah - you may have found the secret to a successful transition from vending machine to pokie machine...

I will have to try it.

I have sort of tried it. Every now and again I forget to take the treat bag with me to dog walk, especially the beach ones.

[Aidan3]

My immediate thought is that, because the magnitude of the reinforcer changed unexpectedly, they might be exploring to see what it is that caused the change. If you press a button 200 times and get a single M&M on whatever schedule there is in place, you have an expectation of a single M&M. Then all of a sudden you start getting 10 M&Ms, you're going to wonder what you did.

If you want to get a visual representation of what happens to any associated respondents, this java applet allows you to run repeated trials while playing around with the strength of the unconditioned stimulus. The output is determined by the Rescorla-Wagner model. You can also investigate blocking and shadowing effects using different stimuli, which might be a consideration if you 'step back' and view your apparatus from the dog's perspective.

[Mrs Rusty Bucket]

I caught a bit of the drs that I'd recorded that talked about "intermittant re-inforcement" and all the happy brain chemicals people (dogs too?) get when they get this kind of re-inforcement. And they linked it to sport and making a good play. I suspect seeing your favourite team make a good play is almost as re-inforcing as doing it yourself.

http://www.thedoctorstv.com/videolib/init/3839

Because gambling is rewarding and unpredictable, which is known as intermittent reinforcement, it can be extremely addictive.

Ie the pokie machine payout.

[corvus]

Thanks.

Just to be clear, the reinforcement wasn't really intermittent at any time. It was quite predictable, as that's kind of integral to the study design. They have to know when the behaviour will be rewarded and when it won't.

[Jigsaw]

Not quite. The reward rate has been constant. The volume just went from base level to lower, then back up to base level again. The behaviour was maintained in one dog on the lower reward volume, but only semi-maintained in the other (about 50-90% response rate to cue). When the reward volume went back to the base level, both dogs started performing the behaviour when not cued as well. I'm wondering if learning theory can explain this, because I'm really excited about the alternative and I need someone less excited to be rational about it.

My learning theory knowledge is just the basics, really. I never did psychology. There is a huge mountain of data about this stuff, but it's so hard to find anything specific in it!

I had an analogy late last night very similar to Aidan's about MM's and lost it Yep, I'd certainly be going hammer and tongs at the machine trying to get more chocolate!

A continuous reinforcement schedule can produce response reduction and if your reward fell below expected levels you may have seen a drop off of the behaviour to a slow extinction curve. When the machine started to reward properly I'm guessing it suddenly became more salient to the dogs and you saw an increase in behaviour. If you were running concurrent schedules of reward it can induce excited behaviour and apparent disregard of the stimulus in some cases.