BiologicalBasis 8 Rules vs Associations slides and notes.pdf

Lecture 8Rules and Associations PSY2304Biological Basis of Behaviour First –a brief aside…•Don’t believe everything you’re told•And don’t believe everything you read•Even in scientific journals…•Consider the strange case of N rays.•What was reported as discovery of a new physical phenomenon turned...

Lecture 8Rules and Associations PSY2304Biological Basis of Behaviour First –a brief aside…•Don’t believe everything you’re told•And don’t believe everything you read•Even in scientific journals…•Consider the strange case of N rays.•What was reported as discovery of a new physical phenomenon turned out, at best, to be a psychological one.•The important thing is to be sceptical. Judge everything on its merits. With that in mind… A warning from history. What do we mean by intelligence?•Even in humans this is the subject of much debate.•Is intelligence one thing -some kind of general-purpose ability (cf. Spearman’s g)?•Or is it a composite of social skills, verbal skills, spatial skills etc. -and what would this mean as applied to animals?•In humans we use IQ tests, in animals we use a variety of different problems to assess intelligence. One of the key things we look for is evidence of the ability to use rules, to reason, as distinct from performance based on associative learning / conditioning. Can animals use rules?§We’ll look at evidence that points towards rule use in a variety of species. We’llstart with some of the simpler paradigms used to study rule-based learning in animals, before moving to more complex designs.§Today’s problems are more about the logic of the design than the application. In a way, what we’re studying is how, in principle, to test an animal’s abilities. §Ideally, we’d like to come up with something that could be utilisedacross species and so form the basis for a sort of IQ test. Serial Reversal Learning: Mackintosh (1974)++--+-Learn this problem to a criterion of say 90% correct. Correct colourcan occur on either side.Then it reverses -learn to same criterion again. Remember it’s colourin this example that matters, not position.It then reverses again -learn to same criterion -and so on.The result? On later reversals the animal makes less errors in acquiring the discrimination. In extreme cases it makes only 1 error! Evidence for the use of a “win -stay / lose -shift”rule? Analysis§The result is consistent with the idea that something like a rule is being used by the animals. The win-stay, lose-shift rule simply says that on picking a colour(in our example), if reward follows, keep picking that colour, if it doesn’t then switch to the other one. but there are other explanations that must be ruled out first.§It may be that the animal is learning which aspects of the stimulus to attend to –in this case colourrather than position -and this enhances learning -but could it lead to one-trial learning?§What’s needed to counter this objection is some form of transfer test, and fortunately there is a similar experimental paradigm that has this built-in to the procedure. Learning Sets: Harlow (1949)+-++--Learn this problem to a criterion of 90% correct. It’s the object that matters –not its position.Then it changes to different objects -learn to same criterion again.It then changes again -learn to same criterion -and so on.The result? On later problems the animal makes less errors in acquiring the discrimination. In extreme cases it makes only 1 error! Evidence for the use of a “win -stay / lose -shift”rule? Analysis§Learning to attend to certain stimulus features won’t explain these results -as the stimuli and relevant features change from problem to problem. One-trial acquisition of this task is very suggestive of win-stay / lose-shift rule use.§But even here there are other possibilities: Could obtaining the reinforcer be used as a cue to help solve the discrimination? The animal might learn that a very recently seen cue is good if it’s just obtained a reinforcer, but bad if it hasn’t, what’s called a conditional discrimination. §We know that animals can learn that the blue stimulus say (of a blue / yellow simultaneous discrimination) is rewarded when a tone sounds, but that the yellow stimulus is rewarded when a clicker sounds. There is no urgent need to postulate rule use to explain this, so what’s so different about the conditional discrimination explanation just given for learning sets? Tr a n s i t i v e I n f e r e n c e§We can perform transitive inference: If A is bigger than B, and B is bigger than C, is A bigger than C?§Can animals do this? If so -would it be evidence for use of some relational rule (“bigger than”in the example given above).§The answer is that there is evidence for transitive inference in animals, some of which can be explained in simpler terms, but other evidence is quite persuasive.§The basic design of these experiments is to train the animals on a chain of pairwise items -and then test a novel pairing. A+ B-B+ C-C+ D-D+ E-TrainTestB vs. DThe animals chose B over D -evidence of transitive inference?McGonigle and Chalmers (1977, 1992) -Squirrel monkeys. A+ B-B+ C-C+ D-D+ E-TrainTestB vs. DA is always reinforced, E never is. B, C, and D are partially reinforced, but B is paired with A and so can associatively retrieve it and benefit from A’s strong association with reward. D has a similar relationship to E. Hence B is chosen when paired with D. To t a l Associative strength for rewardAn associative explanation: von Fersen et al. (1991). A+ B-B+ C-C+ D-D+ E-TrainF+ G-G+ H-H+ I-I+ J-E+ F-Treichler and Van Tilburg (1996) -rhesus monkeysAn attempt to rule out an associative explanation that's not completely convincing.Test B vs. G and D vs. IAt the end trainFind that B and D preferred. Problem is the “linking” training. Analogical Reasoning§We can reason by analogy: Puppy is to Dog as Lamb is to ?§Can animals do this? If so -would it be evidence for use of some relational rule (“juvenile form”in the example given above).§The answer is mostly no -with one very famous exception. This is Premack’s language trained chimp -Sarah.§Sarah has been trained to communicate using symbols, and has learned to use symbols for “same”and “different”appropriately. The next slide gives an example of the type of problem she is now able to tackle as a result. KeyPadlockClosed painted canPaintbrushCan openerSameGillan et al (1981)Sarah is able to make the right choice -the can opener in this example. Summary–What is intelligence in animals?–They undoubtedly have some pretty impressive abilities.–But many of them can be explained by an appeal to associative learning, or to specialised skills that may be innate (e.g. homing pigeon).–Nevertheless, there is some evidence consistent with rule use in animals, and at least one animal (Sarah!) appears capable of analogical reasoning. ReadingRequired reading: Pearce, chapters 4 (111-121) , 7 (187-89), and 10 (259-62).References: •Gillan,D.J.,Premack,D.&Woodruff,G.(1981).Reasoninginachimpanzee:I.Analogicalreasoning.JournalofExperimentalPsychology:AnimalBehaviorProcesses7,1-17.•Harlow,H.F.(1949).Theformationoflearningsets.PsychologicalReview,56,51-65.•McGonigle,B.O.&Chalmers,M.(1977).Aremonkeyslogical?Nature267,694-696.•McGonigle,B.O.&Chalmers,M.(1992).Monkeysarerational.QuarterlyJournalofExperimentalPsychology,45B,189-228.•Treichler, F. R. & Van Tilburg, D. (1996). Concurrent conditional discrimination tests of transitive inference by macaque monkeys: List linking. Journal of Experimental Psychology: Animal Behavior Processes22, 105-117.•vonFersen,Letal.(1991).Transitiveinferenceformationinpigeons.JournalofExperimentalPsychology:AnimalBehaviourProcesses17,334-341.