Transcript Can Animals think?

Can Animals think?
A behaviorist perspective on
comparative cognition
Comparative Cognition
• Evolutionary approach:
– cognitive mechanisms evolve in response to selective pressures
peculiar to each species’ ecology, physiology and morphology
– Thus cognitive processes differ across species
• Comparative cognition:
– compare abilities across species
– Analyze performance of different species on same set of tasks
• Focus on species-specific mechanisms and strategies that underlie
problem-solving performance
– Better to understand how different animals problem solve rather than
– Try to determine which animal is “smarter”
– Must use a set of tasks, not a single task
What behaviors to study?
• Relearning
• Altering habits and adapting to change by
switching to different solution routine
• Behavioral flexibility
• Neophilia
• Exploration strategies
• Attention
• Motivation
• Affordance learning
• Physiological constraints
Morgan’s Canon:
• Accept the lowest level of intentional explanation that works
• Natural selection: what matters is that the animal achieves
goals such as finding food, mates, and safety.
• Using Morgan’s canon to choose among alternatives assumes
that natural selection has always produced the lowest level
intentional system that can do the job.
• But doesn’t answer the question of how the animal does this!
Why is this relevant?
•
Functional concerns of behavioral ecology and
ethology often lead to mechanistic questions - which are the realm of cognition.
•
Cognitive ethologists are frequently concerned
with the diversity of solutions that living
organisms have found for common problems.
•
Emphasize broad taxonomic comparisons
•
Do not focus on a few select representatives
of limited taxa (we hope).
•
Looking for general principles of behavior or
cognition.
Behavioral ecology can inform
questions of cognition
• Optimal Foraging Theory: maximize rate of energy intake and
fitness.
• Example: Woodpecker takes longer on some trees than on others
when foraging.
• Assume: adaptive, optimal, maximizing energy, maximizing fitness.
• Function: avoid depleted food, avoid predation, stay close to nest.
• Mechanism: how know depleted, what info tells bird to change
behavior, how does it know where its nest is?
• Measure: distances between trees, prey repletion rates, prey energy,
etc.
What about memory?
• Do animals show evidence of “memory”
• What kinds, how is it altered, and how can it
be explained?
For example: Kinds of Memory
• Reference: Information
that is procedural and
long term.
– How to dial phone
• Working: event specific
and short term.
– What number to dial.
Vaughan and Greene (1973):
• pigeons trained to classify slides as positive or
negative
– random- no concept involved
– after 10 sessions, could classify 80 slides (40 + and
40-)
– learned next 80 even faster
• eventually worked up to 320
Herrnstein, 1976:
• Wanted to see if pigeons could categorize
• Three categories:
– Trees
– Water
– Person
• Found that pictures being seen for first time (novel)
discriminated as well as training pictures
• Interestingly- similar patter of errors and correct
discrimination across the pigeon subjects
What does this mean?
• Ability to discriminate open-ended classes of
stimuli poses problems at two levels:
– Analysis of features enabling subject to tell
whether object is member of particular class
– Analysis of properties of classes that render them
discriminable
What are they responding to?
• Too complex to be common elements
• Cluster of features that are more or less
isomorphic-probabilistic conjunctions and
disjunctions
• Look like semantic categories of generalization
Pigeons as Art Critics
• Birds: excellent visual acuity in comparison to
humans!
– But: use artificial settings for discrimination training
– This study used “natural” stimuli- paintings
• Difference between Monet and Picasso
– Monet: landscapes, more realism
– Picasso: Cubism, not “real”, much more sharp corners
and edges
Experiment 1
• Pigeons trained on discrmination between
photos or videos of Monet and Picasso
– 8 pigeons
– Projected pictures and then had to peck key
underneath “correct” picture
– 10 paintings from each artist
– Testing stimuli: novel paintings from Monet and
Picasso, then from Cesanne, Braque and Delcroix
– Second test similar with 3 other new artists
Experiment 1
• Trained to 90% criterion
• Test 1: color paintings of monet vs picsso
• Test 2: presented paintings out of focus to
obscure “edges”
• Test 3: left right reversed
• Test 3: novel stimuli of Monet, Picasso and
other artists
Experiment 1
• All subjects learned discrimination
–
–
–
–
Had preference for some paintings
Not color
Not edges or sharp outlines
little problem with mirror image and upside down
images
• Generalized to other impressionist paintings and
cubist paintings
• Evidence of both categorical and individual
discriminations
Experiment 2
• Trained to a pseudo concept discrimination
– Discriminate 2 arbitrary groups of paintings
– Contained both Monet and Picasso pictures
• 2 pigeons
• Same manipulation of stimuli
• Both easily learned the task
What does this mean?
• Pigeons’ discriminative performance could be controlled by
different styles of paintings
– No identified single cue for discrimination of paintings
– Some decrease in responding for reversed or upside down
paintings
• Note: paintings had little if any ecological significance for
pigeons• Distortion tests:
– More disruption when painting displayed real object (Monet)
than abstract (Picasso)
– Evidence that could discriminate both individual paintings and
group them into categories
• Evidence of Flexibility of categories
Gorillas and Natural Concepts
• Several species of animals show ability to form concepts:
–
–
–
–
–
–
–
Pigeons
Parrots
Crows
Dolphins and whales
Seal lions
Dogs
Etc.
• Question: is this a perceptual ability or cognitive ability?
– Obviously, must have perceptual characteristics
– To show cognitive ability must show ability to transfer learning to novel
exemplars
– These must vary across several dimensions
– Evidence in nonhuman primates that they attend to local features, not global
features (of concept)
Abstract vs. concrete concepts
• Concrete concepts:
– Share many features
– Easily discriminated along perceptual lines
• Abstract concepts:
– Share fewer features
– Defined in terms of breadth of category to be learned
– Fewer perceptual overlaps
• Humans easily perform abstract concept formation
• Question: do great apes also show this (since are our closest
relatives)
Method
• Subject = 4 year old captive female lowland gorilla (Zuri)
• Materials:
– Photo sets: 10 S+ and 10 S- category exemplars
– S+ and S- shared similar backgrounds, matched on as many features as
possible
– Minimized similar perceptual features across S+ and S-
• Procedure:
–
–
–
–
–
–
–
Used Apple computer
10 S+ and 10 S- per session
Photo pairs randomly presented
Many sessions per day
Basically had to discriminate great apes vs. humans
Used first 2 sessions with novel photos to indicate transfer
Coded photos across several dimensions
Phase 1: concrete discriminations
• Gorillas or orangutans vs. humans
• Orangutans versus other primates
• Orangutan color test
• Could examine transfer by errors:
– E.g., If responding by color: not show transfer to
black and white photos
Phase 1: Results
• Gorillas vs. humans
– Reached criterion in 14 sessions
– Showed transfer
• Orangutans vs humans
– Reached criterion in 7 sessions
– Showed transfer
– Better at pictures of adults than
young apes
• Orangutans versus other
primates
–
–
–
–
Reached criterion in 19 sessions
No immediate transfer
Took 25 sessions on second rianing
Third set only 3 sessions
• Orangutan color test
– Reached criterion in 7
sessions
– No transfer
– Mastered second set in 2
sessions
– Showed transfer to third
• Gorillas vs other primates
– Reached criterion after 16
sessions
– High degree of transfer
Phase 1 results
• Could examine transfer by errors:
– E.g., If responding by color: not show transfer to black and white
photos
• Could detect gorillas and orangutans vs humans
• Not as good on orangutans vs other primates; gorilla vs other
primates was good
• Did not appear to be discriminating on basis of single feature, but
instead was using multiple features
• Still: could be concrete concepts rather than abstract
Phase 2: Intermediate discriminations
• Primates vs. nonprimates
– Mammals, reptiles, insects, birds, fish
• Primate controls:
– Used stimuli that she made many errors with
• Results:
– Primates vs. non primates
•
•
•
•
•
Reached criterion after 12 sessions
Not show transfer
23 sessions on second set
3 sessions on third set, with some transfer
Only age affected discrimination (as before)
– Correct if primate photo was young animal
– Incorrect if non primate photo was young animal
Phase 2: Intermediate discriminations
• Zuri had more trouble with intermediate
discriminations relative to concrete
– Age affected ability to discriminate
– More likely to select photos of species she had
seen before or served as S+
Phase 3: Abstract Discriminations
•
Animals vs. non animals
–
Non animals = landscapes with neutral background
•
Food vs.. Animals
•
Results:
–
Animals vs non animals
•
•
–
Food vs animals
•
•
•
12 sessions to criterion on first set
Showed transfer on all subsequent photo sets
Quick to criterion
Good discrimination on initial transfer
Better at abstract discriminations!
–
–
Suggests may have been relying on perceptual qualities for concrete and intermediate, but could not for
abstract
Why better at abstract than intermediate?
•
•
–
Within class and between class similarities interact to determine relative difficulty of discriminations at various levels of
abstraction
Also: were artificial “human” discrminations…..don’t know meaning to gorillas
Showed excellent transfer, unusually so for a non human primate
Better at abstract discriminations!
• Suggests may have been relying on perceptual qualities for concrete
and intermediate, but could not for abstract
• Why better at abstract than intermediate?
– Within class and between class similarities interact to determine
relative difficulty of discriminations at various levels of abstraction
– Also: were artificial “human” discriminations…..don’t know meaning
to gorillas
• Showed excellent transfer, unusually so for a non human primate
– Could not have been just memorizing
– Some effect of experience: “learning to learn”
Does evolution play a role:
• prewired to see trees, water, people?
– Presence of static features can be discriminated
– More likely that are prewired to form a “schema”
or prototype
• Examine behavior of other species to
determine how behavior may be “prewired”
and learned
Dogs and Concept Formation
• Strong discrimination learning
– Most often use MTS or DMTS
– Visual cues
• Color of objects: blue vs. organge, black vs. white
• E.g., Milgram, et al, 1994; Araujo, et al, 2014
– Spatial cues: body position and landmarks
• Better at body position (L, R)
• Milgram, et al 1999; Ashton and DeLillo, 211
– Auditory cues:
• Go/no go: Brown and Slotysik (1999)
• Different sounds
• Human vocal signals (McConnel, 1999)
– Olfactory cues, particularly nonsocial odor cues
Can do many tasks that corvids, sea
mammals and apes can do!
• Contingency reversal learning:
– Can learn A B and then B A
– Ashton& DeLillo, 2011
• Object permanence:
– Can find hidden object when observe object hidden
– Some data ((Gagnon & Dore, 1992, 1994) suggests can find
when NOT see the object being hidden
• Object learning (Framl & Frank, 1985)
• Categorizing and inferential learning: Range, et al, 2008
Several other tasks, too!
• Object manipulation (Topal, et al, 1997)
• Means-end taks (Osthaus et al, 2005)
• Quantitative tasks
– More vs. less
– Some counting
– Search order
• Spatial navigation: Cattet & Etienne, 2004 and solving
detour problems (Pongracz, et al, 2001)
Nonsocial counting:
• Dogs can count?
– Numerical competence and ability to discriminate
more and less
– Dogs about as good at numerical competence as the
great apes!
• West and Young (2002) from Pepperberg (1994)
• Dogs shown three problems
• 1+1 = 2; 1+1=1; 1+1=3 (all in dog biscuits; shown problem
then solution)
• Dogs gazed longer when the expected solution was wrong
• Similar to Tinklepaugh studies in chimps and monkeys
Expectancy violations
• Tinkelpaugh (1928) task
–
–
–
–
Show food item
Cover it up with a cup
Slide to animal
Animal lifts up cup- but tricked: another lesser
preferred food is there
– Look to see if animal is surprised/upset
• Dogs show strong expectancy violation
– So do chimps, corvids
Spatial Cognition
• How do animals remember where the food
comes from?
• How do animals remember where they left
their food?
• What environmental and internal stimuli are
animals using during search behaviors?
Food Storing Behavior
• Animal creates a resource distribution tha only it
knows/has awareness of.
• Reference Memory: storage sites, what is in the
site, territory
• Working Memory: which site did I empty today?
• Information: spatial layout, site contents, etc.
Do Nutcrackers form Geometric
relations between objects?
• How does the Nutcracker remember where it hides its
food?
• Clark’s Nutcrackers: Birds use general principle to find a
goal located between two landmarks.
– relationship between landmarks
– not between a goal and the landmarks
2)
1)
Goal
Two Landmarks
How form spatial relationships?
• Clark's nutcrackers can learn to find the point halfway between two
landmarks that vary in the distance that separates them.
• general principle, as the birds correctly find the halfway
point when the landmarks are presented with new
distances between them.
• The ability to find a point defined not by the relationship between a
goal and a landmark, but by the relationship between landmarks.
Two distinct processes:
• Direction: the use of directional bearings to find the
(hypothetical) line connecting the landmarks
– North, south, east, west
– Must use landmarks to mark direction
• Distance: finding the correct place along that line.
– Must use landmarks to mark distance
Set up a Test:
• Nutcrackers were trained to find a location defined by its
geometric relationship to a pair of landmarks.
– Distance relationships: Two groups trained to find positions on
the line connecting the landmarks
– Constant Direction: Two groups trained to find the third point of a
triangle
• Four inter-landmark distances and a constant spatial
orientation were used throughout training.
• Result:
– Constant distance group learned more slowly with less accuracy
– showed less transfer to new distances
Spatial Relations:
• When tested with a single landmark
– birds in the half and quarter groups tended to dig in the
appropriate direction from the landmark
– So did birds in the distance group.
• Nutcrackers CAN learn a variety of geometric
principles:
– directional information may be weighted more heavily than
distance information
– can use both absolute and relative
– including configural information about spatial relationships.
Three Landmarks
Hunting by search image
• Five known forms (or
"morphs") of the North
American underwing
moth, Catocala relicta.
• Note the variable forewings and the relatively
uniform hind wings.
• hunt by searching
image.
What about dogs?
Evidence of Solidarity Principle
• 2 critical tasks for animals:
– Animals must be able to predict trajectories of moving objects
– Animals must be able to predict whether to approach or avoid an object
• To do this: must be able to understand object occlusion- that is, how
hidden objects “behave” as evidenced by:
– Predictive reaching
– Searching
– Following a hidden object (and judging where it will emerge)
• Many (if not most) animals have some degree of ability to do these tasks
– Is this an innate/prewired behavior?
– Is this a “learned” or experience-based behavior?
– Use expectancy violation paradigm.
Limitations on trajectory prediction
•
Developmental time course to behavior
•
Differs across animals, with primates showing greatest capacity
•
Seems somewhat dependent on motor abilities
•
Problem: differences in data between search and expectancy violation tasks
– Lack of executive ability? Mismatch between knowing and acting
– Observational knowledge not same as action knowledge
– Global interplay between cognitive subsystems and across settings?
•
What about dogs?
– Evaluate search behavior when object rolled in direction of a barrier
– human toddlers, adult rhesus macaques and cotton top tamarins all fail to reason about
location of hidden barrrier when reaching for an invisibly displayed object
– Even when show sensitivity to solidity info in expectancy violation paradim
Method
• Subjects:
– Hare’s lab
– 10 Pet dogs all over 1 year of age with no formal agility or advanced
obedience training
• 4 female, 6 male, average age 4.5 years
• many breeds; 6 were mutts
• Apparatus:
–
–
–
–
–
Used a clear plastic tube which extended into a wooden box
2 doors on box: far door and near door
Could reach inside each door to retrieve object
False back: researcher could place objects into box on one side or other
Basic idea: roll object down tube; barrier should block the object from rolling
to far door.
– False baiting: distracted dog and moved object
Procedure
• Pretraining trials:
– Dogs retrieved objects from box and generally acclimated
– 3 trials with each ½ of box
• Test trials:
–
–
–
–
–
–
No wall and wall trials
Three blocks of each
Mixed block, no wall and wall
Mixed block always appeared last
Barrier not in place during no wall (obviously!)
Dogs prompted to “look” and find the treat
Results:
• Mean percent correct:
–
–
–
–
–
More mistakes during no wall
Wall: 85%
No wall: 75%
Mixed: 83%
Dogs tended to:
• Search near location when barrier in place
• Search far location when no wall
• Spontaneously searched in correct location from first of each
trial type
• Performed correctly when mixed trials
Control Trials
• No treat dropped down tube
• Treats placed in box in different positions
• If dogs not attending to dropping of treat but to
experimenter cuing, should perform according to
location of wall
• Results showed:
– No wall sides: chose far side 55%
– Not significantly different from chance
– Concluded that were attending to rolling treat
Conclusions
•
•
Dogs do show solidarity principle
Why dogs, but not non-human primates?
–
–
–
•
Selective breeding?
Experience with humans and thus human tasks?
Other explanations
Dan’s questions:
–
–
–
–
–
–
–
Marc Hauser was referenced on several occasions. How familiar are you with some of the controversies
surrounding him and his research?
If you are familiar with them, how thin of a line is there between what Hauser did, and some of the
questionable practices are commonplace amongst scientists (e.g. manipulating sample sizes, boutique
statistical analyses, not reporting null findings, etc.)?
What is the expectancy violation paradigm?
Generally speaking, which animals can maintain a representation of hidden objects over time? Which ones
cannot?
What are some of the explanations for why dissociations emerge between search and expectancy violation
tasks?
How important is breed in the tasks examined?
The authors recommend that future research examine differences between domesticated dogs,
undomesticated wolves, and different levels of socialization amongst dogs. Has this been done yet? If so,
what were the findings?
Spatial Perseveration Errors
•
Detour tasks: move away from goal to get to goal
– Perseveration errors: failing to shift strategies
– Occur even when visible change of location of correct path, so can’t go back to old path
•
Dogs can solve detour tasks
– But: will show perseveration errors
– Dogs from deprived environments worse than dogs with experience
– More experience with a solution = more perseveration on that solution
•
Major aim of study:
– Determine if cognitive rigidity of dogs changes when change set up in simple detour task
– IV: number of learning trials
– IV: transparency of the barrier
•
If dogs able to solve task after a required change in detour path: indicate complete
understanding of task dynamics
•
If perservation error: lack of inhibition and prominence of learned behavior over
clear visual cue
Method and Exp 1
•
Wooden barrier made from lattice garden trellis put across enclosed rectangle
–
–
Opaque condition: covered with cloth
Visible condition: uncovered
•
Video taped all sessions
•
Experiment 1:
–
–
–
–
–
•
Results:
–
–
–
–
•
•
N = 2- of mixed breeds
4 A trials: started at either L or R
After 4th trial: barrier shifted to gap on opposite side
Then 4 B trials
If dog went straight to opening = correct; else incorrect
All dogs in both conditions solved above chance
Opaque condition slightly faster
First trial significantly slower for B
Did not unlearn previous trial as shown by more errors
Transparency of task had no effect
Did rely more on memory than on visual with more trials
Exp 2
•
•
•
•
Same set up as Exp 1
Changed location of barrier after 1,2 or 3 trials
N = 30
Results:
– First set:
•
•
•
•
•
•
8/10 dogs went straight to gap in barrier after A1
7/10 after shift
B2 = 90%
B3 = 90%
B4=86%
None reached 100%
– Second set:
•
•
•
•
•
A2: all dogs solved
B1: 1/10 solved
Took longer time for B1 than A1
Reached optimal performance within 2 trials
All showed perseveration error on B1, 4 still showed on B2
Discussion
• Dogs exhibited spatial perseveration reliably after 2 or more
presentations
– Failed to walk straight to new obvious goal
– Began to rely on old learning rather than visual cues after 2 trials
• Sam’s Questions:
–
–
–
–
What is spatial perseveration error?
Does this differ depending on species (dogs vs. humans)?
Why did the dogs make this error?
How can you tie this article with the Misbehavior literature we read
earlier in the semester?
– Why could the dogs not reach the 100% accuracy after this error
occurred?
– How can you apply this to your life as a student?
Means-end Analysis
• Problem solving = progressing through series of
mediating actions in order to reach certain end goal
– Must understand series of progressive steps as means to
end
– Means to end understanding = key step in cognitive
development of humans
• What about non human animals:
– Use string pulling or support problem
– Tamarins, ravens/corvis, parrots, elephants all show ability
– Dogs and domestic cats can retrieve distant food but fail to
understand means-ends analysis
Why poor performance in dogs?
• Dogs seem to be able to DO the task, but not show understanding
of task
–
–
–
–
Not good with spatial references
Great with social cues
Thus good social cognition; poor physical cognition
Likely due to breeding/experience with humans
• Is this due to lack of cognition or lack of impulse control?
– Most errors are proximity errors: do the closest
– Capable of efficient problem solving, but gets interrupted by other
behavioral strategies
• Choose based on social cues and experience rather than physical
• Choice based on exclusion as last result
• Training also important, especially clicker training
So what did the authors do?
• To what extent proximity bias overrides or
supports choice based on physical connection?
• What do dogs choose when decision based on
proximity is not possible?
• examined effects of impulsivity
• Examined effects of training history
•
Method
Subjects: 68 dogs and owners:
–
–
–
•
Test apparatus:
–
–
–
•
15 males and 17 females (mean age = 6 years) made it through testing
16 clicker trained/16 not
17 dogs tested with object that had to retrieve and bring back to owner to get treat; 14 dogs got treat directly
Black wooden board with 2 yellow colored wooden boards mounted on top
Dog had to move yellow boards back and forth with paws to get treats/ball
Tester in cage behind apparatus so dog couldn’t interact
Procedure:
–
–
–
Training phase: taught to pull boards with paw using shaping
Motivation assessment: object retrieval for food treat
Test phase: 4 different conditions
•
•
•
•
•
–
Conditions:
•
•
•
•
•
Board presented with reward/object on surface (ON)
Identical reward/object on other side of board (OFF)
To obtain reward, had to pull out board with reward/object on it
Trial ended when dog investigated board that had pulled out and took reward/object if it was there
12 trials per condition (6L 6R)
Same Distance
Weak proximity + support: both rewards/objects placed at far end of board but reward/object placed “on” board was slighter closer
Strong proximity + Support: much closer reward/object placed on board
Proximity against support: reward/object closer to dog was placed “off” board
Fixed factors:
–
–
–
–
Sex
Type of reward
Type of training
learning
Results
•
Same Distance:
– None of fixed factors affected dogs’ performance
– Dogs chose reward/object “on board” signif. more
•
Weak proximity + support:
– None of fixed factors affected dogs
– Dogs chose “on” more often
– No significant difference from condition 1
•
Strong proximity + Support:
– Learning had an effect
– Above chance by trial 6, and got better
– Closeness improved performance, compared to condition 2
•
Proximity against support:
– Dogs rewarded with food did better
– dogs were better in condition 1 than condition 4
– Proximity against support was detrimental to performance, but not that much
Discussion
•
Concluded dogs posses ability to consider means-end relationships in support problem even if
proximity is a confound
•
Reward type did play a role
•
Training did not play significant role
•
Proximity helped only if it was exaggerated
•
But: performance NOT that strong overall when look at individual performance
•
Not support previous research that suggests dogs not have means end analysis
–
–
•
Difference between string test and this task?
Were researchers asking the right question?
After looking at all three studies:
–
–
–
Can dogs problem solve?
Do they show evidence of thinking?
Why or why not?