Animal intelligence and problem solving Flashcards
what do we mean by intelligence?
Even in humans it is a debate
Is intelligence one thing – general purpose ability (Spearman’s g)?
Or composite of social skills, verbal skills, spatial skills etc. – and what would this mean as applied to humans
In humans use IQ tests, in animals use variety of diff problems to assess intelligence – look for ev of ability to use rules, reason, as distinct from perf based on associative learning
warning
Problems with assessing animal intelligence:
1. Human like behav in animals may appear to be intelligence when not
2. E.g. clever Hans
Couldn’t answer certain problems when over a loudspeaker
- Looked for people’s expressions – has to be someone in the room that knows the answer
“In no case may we interp an action as the outcome of the exercise of a higher psychical faculty, if it can be interp as the outcome of the ex of one which stands lower in the psych scale”
historical background
2 schools of thought
- Gradual (quan) diff between humans and other animals
- Sharp (qual) distinction between humans and other animals – MacPhail hyp – do diff in way vertebrates learn – might learn at diff rates esp. in diff circumstances – humans can just do more
trial-and-error
Thorndike
- No insight
- Law of effect
insight
• Köhler about Thorndike’s exps (1927)
o “If essential portions of exp apparatus cannot be seen by animals, how can they use intelligence faculties in tackling the situ?
o Setting problem that cannot solve
Gestalt psych
Perception of relations
No PA involved
If too difficult: trial and error
E.g. monkey’s and crates to reach bananas
o Realised problem – discovered solution – then executed solution
o Anecdotal ev – limitation
Epstein et al. (1984)
study with birds o Plastic banana – peck = grain o Cannot reach it o Uses block o Took months to train o Not same at all? o Proof that animal could learn to do this
Don’t know what experiences chimp had – boxes always in compound and played with/interacted with them – cannot claim it happened without training
insight (2)
Insight learning ev often open to interp
Köhler – anecdotal, no knowledge of prior experience of animals
Epstein – training experience produced intelligent and insightful behav – existence proof
Insight v instrumental learning
spatial intelligence - detour tests
Köhler (1925)
Key feature
Distance of food from (transparent) fence
Poucet (1983): cats – transparent v opaque barriers
- If wants to go towards it = naturally takes In down the longer path
Opaque = takes shorter path
Some animals able to solve problem by moving away from food to take optimal route to goal
Intelligence? /triumph of instrumental over Pavlovian? – which can be difficult to achieve (e.g. omission schedules)
tool use
Use of external object as functional extension to attain immediate goal
Many animals observed engaging in primitive tool use – indicator of intelligence?
Dropping stone on mussel (using tools) v dropping mussel on stone (random behav/chance)
tool use as ev of causal inference - Povinelli (2000)
Some animals can learn to push from current side – intelligence?
50/50 which end go for at first
One side more successful than other
Put food at other side of trap – has to push from other end – mostly get it wrong at first
see notes
tool selectivity
Chappell and Kacelnik (2002): new Caledonian Crows
o Have to learn to pick right stick length
see notes
tool construction
Weir, Chappell and Kacelnik (2002): tool shaping in new Caledonian Crows
o Best ev
o Use handle to pick bucket up
see notes
can animals use rules?
Claimed they can – same/diff relation studied using transfer tests in matching to sample studies – successful transfer observed, results explained in terms of discrimination based in estimated of recency/familiarity
Animal has some notion, based on trace strength – how long ago last encountered stim – perceived recency
In matching to sample, of learns to select “most recent one”, show transfer to novel stim
serial reversal learning - Mackintosh (1974)
Learn problem to criterion of 90% correct
Then reverses – learn to same criterion again – colour that matters
Reverses again – learn to same criterion etc.
On later reversals, animal makes less errors in acquiring discrimination – in extreme cases just 1 error (chimps) – ev of “win – stay / lose -shift” rule?
see notes
analysis for Mackintosh (1974)
Result consistent with idea that something like rule used by animals, but other explanations must be ruled out first
Animal may learn which aspects of stim to attend to – and this enhances learning – but could it lead to one-trial learning?
Most animals don’t just make one error
Gets faster as attends to correct stim quality
Need some form of transfer test
learning sets - Harlow (1949)
Learn problem to criterion of 90% correct – object that matters – not position
Then changes – learn to same criterion again
Changes again – learn to same criterion etc.
On later problems, animal makes less errors in acquiring discrimination – in extreme cases just 1 error (chimps) – ev of “win – stay / lose -shift” rule?
see notes
Harlow analysis
Learning to attend to certain stim features won’t explain results – as stim change from problem to problem – one-trial acquisition of task very suggestive of win-stay/lose-shirt rule use
Could obtaining reinforcer used as cue to solve discrim? – animal might learn that recently seen cue good if just obtained reinforcer, but bad if hasn’t conditional discrim
Know animals can learn blue stim rewarded when tone sounds, but yellow rewarded when clicker sounds – no need to postulate rule use to explain this
transitive inference
Can perf transitive inference: if A bigger than B, and B bigger than C, is A bigger than C?
Can animals do this> if so – would it be ev for use of relational rule?
Ev for transitive inference in animals, some can be explained in simpler terms but other ev persuasive
Basic design is to train animals on chain of pairwise items – and then test novel pairing
McGonigle and Chalmers (1977, 1992) - squirrel monkeys
If animals choose B over D – ev of TI?
see notes
an associative explanation - von Fersen et al. (1991)
A always reinforced, E never is
B, C and D partially reinforced, but B paired with A and so can associatively retrieve it and benefit from A’s strong association with reward
D has similar r’ship to E]B chosen when paired with D
see notes
Treichler and Van Tilburg (1996) - rhesus monkeys
Attempt to rule out associative explanation that’s not convincing
see notes
analogical reasoning
Can reason by analogy: puppy is to dog as lamb is to ?
Can animals do this? – if so, would be ev for use of some relational rule (“juvenile form”)
No – one exception – Sarah the chimp
Sarah trained to communicate using symbols, and learned to use symbols for same and diff appropriately
Gillan et al. (1981)
o Sarah able to make right choice – can opener – gets right on first trial
