Animal Cognition: Counting, Spatial Representation, and Tool Use
Counting and Numerosity
Brannon & Terrace (1998): About Counting
- Numerosity is often confounded with other properties like size and surface area.
- Duration (auditory counting) is also a factor.
Methods
- Training: Touch screen; choose exemplars in ascending order of numerosity.
- 35 sets that differ in size and shape of elements.
- Category types: nominal vs. ordinal.
- If forming nominal representations, they are learning representations of an arbitrary feature, not ordinal representations.
- Learning a general rule, not a stimulus-specific rule.
Fig. 2: Accuracy increases across training sets.
- Are they learning to count/represent numerosity successfully?
- Or learning the order in which to respond to each stimulus set?
Harlow: Subjects can learn to learn; they learn that there is an order and they have to figure it out for each new stimulus set, rather than tapping into the common property across all stimulus sets: that they are ordered according to numerosity.
- These same monkeys participated in a previous study where they learned to sequence arbitrary stimuli; so maybe they are doing the same thing here.
- Chance accuracy = 2.5% (4 alternatives).
Testing
- Tested with 150 new stimulus sets.
- 1 through 4, each stimulus is novel.
Fig. 2b: Accuracy in test sessions.
- Apply what they learned to novel numbers 1-9.
- Pair low (familiar) with high (novel) numbers.
- Reinforce familiar numbers to ensure they’re not learning just a simple rule.
- Probe trial type of interest in pure form so as not to introduce new learning: measure what they know.
- Accurate.
Fig. 4: Larger numerical distance between numbers leads to more accuracy.
- More discriminable.
- Numerosity effect: Weber’s law for numbers.
- Able to determine ordinal relations between stimuli.
Different Ways They Could Be Doing It
- Using a counting algorithm to judge relative magnitudes.
- Using a one-to-one corresponding matching algorithm.
- Comparing each stimulus to each other stimulus.
- Okay, there’s 1 in both, there’s 2 in both, there’s 3 in this one but not 3 in this one… so 1 with 3 is the right answer.
What do the results imply about what animals are actually doing?
- Either using a counting algorithm or a matching algorithm.
- But using something to make an ordinal distinction.
How Are These Different?
- Instead of accumulating over an ordinal scale (counting), comparing from one stimulus set to the other and checking each pair off in a cognitive list (matching).
- Evolutionary advantage? Next paper.
- Why not counterbalance and have them count down? 4, 3, 2, 1.
- Or choose a smaller quantity to get reinforced.
- In studies using pigeons, get the same effects using ascending and descending sequences.
- But pigeons are not counting: sophisticated in terms of numerosity.
- Harder to train? Initial discrimination is harder: 4 & 3 more difficult than 1 & 2, Weber’s law.
Beran (2008)
Why might it be important to track numbers of moving objects?
- Members of your own and different groups.
- Ranking within a group.
- More social animals seem to be better at this.
Methods
- Pairs of numbers of moving dots 1 to 10, choose the larger one.
- Trained with 1000 trials.
- Helps to learn quantities above 4.
Results
- Able to discriminate between smaller sets and larger sets.
Fig. 1: Distance effect & magnitude effect (absolute differences & proportional differences).
- More distance –> better.
- Larger ratio –> worse.
- Conforms to Weber’s law.
- Same data, just parsed differently.
- Don’t know which effect fits data better because monkeys are performing so well.
Exp. 2
- Controls for movement.
- Able to do numerosity discrimination when one set moves and the other doesn’t.
- So not just that more moving objects are more chaotic (more movement, more complexity), so choose more chaotic image.
- If motion was the cue, should have always picked the moving one.
Exp. 3
- Records cumulative area of pixels illuminated by each dot (“amount” of dots).
- Added in capuchins and people.
- Amount of dots does not make a difference: still use number of dots as cue.
- No difference between rhesus monkeys and people, capuchins are worse.
- Rhesus monkeys are already at ceiling, so people can’t really be better.
- A subset of capuchins are just about as good as humans & rhesuses: best capuchins.
- Differences in rearing? Age?
- Best capuchins were younger.
- Are patterns of results among best capuchins similar to patterns in rhesus & people? Yes.
- Rhesus monkeys are more sensitive to social hierarchy: more important to understand ranking?
- Rhesus: old world species; capuchin: new world species.
- Far separated from each other phylogenically speaking.
Congruency between pixelation & #
- Congruent: pixelation & # are confounded (more amount, more #).
- Incongruent: (larger #, lower amount of pixels).
- Slightly better, but not significant.
- Easier to pick out dots when they’re smaller?
Figure 5
Response time:
- Longer to respond to larger ratios (more similar): more difficult discrimination.
Fig. 6b: Less correct for larger ratios.
- Humans are faster than capuchins & humans at ratio below .078, then slower.
- Monkeys do not differ in slope of functions of latency for correct trials.
- Cubic function (p. 69)? Describe a quadratic trend.
- Trend analysis: data are best fit by a quadratic trend: direction of data changes once: shallow at small ratio & steeper at large ratio.
Exp. 4
- Discriminating a subset of stimuli.
- Track target dots among a set that also includes distractor dots.
- Congruent sets: more total dots, more target dots.
- Incongruent sets: more total dots, fewer target dots, & vice versa.
Fig. 6b
- Rhesus monkeys did well on both tasks.
- Capuchins did worse than rhesus, better on congruent trials than incongruent trials.
- Performed at chance, but not below chance, so they were still trying.
- Counting is useful.
- In a larger system: number & quantity (# of dots vs. amount of dots).
- Are they really counting, or are they conceptualizing “more” or “less” differently?
- Analog number representation?: smooth, continuous differences between 1 & 10.
- Parsed into discrete, natural numbers?
- Seems to be how humans count: 1, 2, 3.
- Papers don’t distinguish (or care) between these possibilities.
- Dimensions correspond to quantity.
- Small numbers: have easily perceptible properties, can just perceive it (2-ness, 3-ness), don’t have to count.
Spatial Representation
Cheng (1989): Vector Sum Model
- Animals have strategies for navigating their environment.
- Finding exact location – use landmarks; one landmark is not enough.
- Vectors: direction & magnitude from landmarks to hidden goal location.
- Vectors are learned by finding the goal location & perceiving where the landmarks are in relation to that goal.
- @ goal site, bird encodes vectors from landmarks to goal.
- Later, calculates distance between itself & 2 landmarks & vectors a & b that it encoded.
- Sums agent-landmark vectors & landmark-goal vectors to get discrepant navigation vectors from bird to goal.
- But if a landmark is moved to the south, the bird calculates navigation vectors.
- The one that corresponds to the moved landmark moves in the same dimension (y/south) as the landmark shift, but not the orthogonal direction (x/east-west).
- Search shift should follow landmark shift in terms of direction – never orthogonal.
- Some vectors may be weighted more than others.
- Landmarks near the goal are more useful for navigating: more accurate info about the goal, shorter distance to calculate, easier to detect.
- The vector sum model predicts the pigeon goes somewhere between the goal location & shifted navigation vector based on the weights associated with each landmark.
- How can you tell the relative weights of landmarks?
- Move landmarks apart from each other @ equal distances (shift 1 north & 1 south).
- The final navigation vector should take you right to the goal if the bird averages the vectors, unless the bird weights the closer landmark more heavily: results: shift search in direction of closer landmark.
Other Theories
- Maybe vectors aren’t learned: maybe just direction?
- If this was true, would need 2 landmarks to locate the object: goal location is defined by the intersection of compass locations.
- Maybe goal distance?
- How far away landmarks are from the goal.
- The goal is @ the intersection of circles surrounding landmarks.
- 2 landmarks isn’t enough (2 circles can intersect @ 2 spots).
Methods (Exp. 1)
- Trained to find hidden food wells.
Testing
- 6 trials just like training trials.
- 2 random test trials: move food wells by rotating box 90°, replace landmarks so the bird can’t tell anything has changed.
Types of Test
- Control test: landmark in original location.
- 2 landmarks moved 10 cm apart each.
- 2 landmarks moved 20 cm apart each.
Phases:
- Trained with either left or right landmark closer first, then switched to the opposite landmark arrangement.
- Videotaped each test trial, counted # of frames in which the pigeon’s head was in the box where the goal is: normalizing space in relation to the goal.
- 0 is defined by where the goal was originally.
Dependent Variables:
- Peak place of search.
- Location in the box that corresponded to the middle of the highest region of the distribution of searches.
- Peak density of search @ peak place.
- % of test time spent in that place.
Results
- Shifted peak search in accordance with the movement direction of the landmark.
- Weighting landmarks when doing vector averaging.
- Overall weighting nearer landmark more.
- Did not search in direction orthogonal to movement (as predicted).
- Spread-out search more following landmark shifts.
Peak Search Densities
- Shifts along left-right axis: searching more from left to right.
- Search distributions for Sukowa in left-right / up-down axes.
- Each of 3 distributions is symmetrical (not skewed).
- Distributions have the same spread when distance is increased.
- Might expect 20cm to be fatter if using a non-linear mental scale.
- The psychological scale used to estimate goal location is linear.
Methods: Exp. 2
- Trained & tested birds with both landmark locations (left or right closer).
- Control test + near/far tests.
- 1 test, near landmark shifted down.
- 1 test, far shifted down.
Results:
- When near shifted down, more likely to shift search down.
- In support of the vector sum model no left-right shift.
- Peak density: decrease on up-down axis search density & peak place was only when the nearer landmark was shifted.
Cheng (1986): Geometric Module
- When near shifted down, more likely to shift search down.
Is Spatial Representation Modular?
- Is there a small number of relatively simple processes/modules that can explain spatial learning behavior?
- Special representational system that codes one’s geometric space.
- Processes that act on this system are independent/modular.
Apparatus
- Visual, textural, odor cues distinguish 2 corners. Also, 1 wall is white in 1b.
- Visual/olfactory cues that can be used as landmarks that provide unambiguous specification of space.
- The rectangular shape of the apparatus could also be a cue.
- Rats’ spatial choice behavior appears to be controlled by geometric features more than expected – rotational error.
- Systematic errors = little interaction between learning about features & geometric properties of the space: modularity.
Procedures
Working Memory: Exp 1
- Study phase: rat goes into 1 of 2 arenas, finds 7 coco puffs in 1 of 100 different grid locations.
- After eating the 3rd coco puff, taken out of the box for a 75-sec delay.
- 2nd arena: food dish is buried under bedding in the same location as the 1st arena with 4 coco puffs (rat thinks 4 remaining pellets are in the same place as before).
- Where does the rat look?
- Sometimes where it would have been (correct hit).
- Sometimes 180 degrees from where it would have been, almost as often as it searched in the correct place (rotational error).
- Equally likely to look in geometrically equivalent place in other corner of the box: geometrically exactly like the other location with respect to location in the rectangle; surprising: seem to be blind to visual cues.
- Sometimes randomly (miss).
Reference Memory: Exp 2
- Food is always in 1 of the 4 corners of the box: learn which corner the food is in.
- Visual cues: panels in corners of the box; rats come to be controlled by visual cues.
- Rotational tests:
- Visual cues & goals are rotated 180 degrees: control for odor cues/other cues that might be guiding the rat: none of these matter: guided by visual cues that were experimentally manipulated (corner cues, wall cues, geometric space).
- Most of the time, went to the correct corner (71%); when they do make errors, more likely to go to geometrically equivalent corner (21%).
- Tested without features in correct & geometrically equivalent corner: can’t discriminate between those 2 corners.
Exp 3: Transformation – transposition results in greater change, but affine & reflection change relation of target with geometric shape.
- Diagonal transposition: switch panels of correct & geometrically equivalent corners.
- No decrement in performance.
- Affine transformation: each panel gets moved over 1 location in clockwise/counterclockwise direction.
- Reflection: locations of panels are flipped across the horizontal axis.
- These are more disruptive than transposition; changes geometric cues: short side/long side features change (if you used to have a short side on the left, now you have a long side). Tend to stick to geometry that had been previously correct. Using geometry & then local features.
- Rats are independently learning features & geometry – recording distance, angles, & left/right relations.
Miller & Shettleworth (2007)
Overshadowing procedure:
- Overshadowing condition: geometric + feature cue / control: geo cue.
- If there is overshadowing, cues should compete with each other for learning, the animal won’t learn as much geo as controls.
- Controls should look more in correct & geom equivalent location than overshadowing condition.
- Test without feature cue.
- Same for overshadowing & control conditions: lack of overshadowing of geometric cues by feature cues.
- Rats choose where to look based on geometric cues = learn geometric cues independently = modular.
Blocking procedure:
- Control condition: learns about 2 cues simultaneously, how much does it know about 1 cue?
- Blocking condition: learn about the same 2 cues together.
- Prior to that, learned about 1 cue (feature).
- The geometric cue is not present: the box is square in this phase, not rectangular (4 corners are geometrically equivalent).
- The feature cue has all the associative strength, none left for geometric cues.
- Shouldn’t learn nearly as much about redundant added cue (Rescorla-Wagner theory).
Results:
- Little evidence for blocking of geometric cues by feature cues.
- Suggests separate systems for features & geometry: not competing for the same learning resources.
- This paper brought together these findings & Cheng’s findings by adding in some twists:
- Associative strength (Vl) of any corner location (l) is the sum of the associative strength of all of the features (elements; Ve) that correspond to that corner (unique visual pattern of that corner, visual features common to all places in the box, geometric features [2 pairs of corners share these features–associative strength varies together for these 2 locations]).
- Change in Ve is a function of pl where e is present.
- pl = probability of looking in a particular location
- Choice procedure: goes to some corners more often than others.
- Importance: exposure to contingencies is different for different elements because it is exposed to some elements more than others if it chooses the corresponding locations more often: the animal exposes itself to stimuli.
- The probability of visiting a location = sum Ve for that location (function of 3 features–common, unique, geometric–in that corner) divided by all the Ve’s of all the corners.
- When the rat visits, there is food there (lambda = 1 = learning) or not (lambda = 0).
- V’s of all features in that corner change accordingly. Some elements have a high reward contingency, leading to an increase of expected reward contingency & V for other elements @ that corner beyond expected (feature enhancement).
- Some of these V’s affect other corners:
- If you go to the correct location & are reinforced, it increases V for the correct location & V for the geometrically equivalent location. Reward @ the geometrically correct corner is reinforced @ > 50%.
- Competition between features: could work toward geometric features not getting as much associative strength as you would expect: V for geometric features in correct & geom equivalent location goes down when you go to geom equivalent location because you don’t find food there, goes up for both when you go to the correct location.
- This is why you don’t get as much blocking & overshadowing when combining geometric & feature cues as you get when you use other kinds of cues.
- Cheng (1986), rats don’t go to affine location (formerly far geometric corner) often.
- This location has in it the visual feature that was in the geometrically equivalent corner.
- Similar to AX+ / BX- discrimination – conditioned inhibition.
- The common element X is part of 2 compound stimuli: 1 is reinforced (AX+), 1 is not.
- A becomes excitatory, B becomes inhibitory, X is neutral (reinforced on 50% of trials).
- Why does B become inhibitory?
- Presented in compound with X which is neutral.
- B is never reinforced: loses associative strength.
- The feature in the geometrically equivalent corner is initially somewhat positive because the corner is seemingly in the correct location.
- But when they go there a lot during training, the unique visual feature in that corner is not reinforced because it is not correct.
- That visual feature becomes inhibitory: gets the least amount of searching.
Pattern Recognition
Terrace
- Difference between input & output chunks in LTM & STM.
- Problem: no one can actually define what chunks/chunking are.
- 7 +- 2?
- People just accepted this without justification.
- Chunking seen in nonsense syllable lists, mazes.
- Most logical thing: there is no other way to remember these novel things that don’t make sense on their own.
- First introduced by Miller (1956): 7 +- 2 idea.
- In WM, a person can hold 5 to 9 chunks of memory, regardless of the total amount of information.
- Terrace: but we don’t really know what a chunk is.
Simultaneous Chaining Procedure
- Subjects are trained to respond to stimuli in a certain order, but stimuli are all presented at the same time.
- Have to learn by trial & error which one to go to first, what the correct pattern is.
- Differences between simultaneous & successive chaining.
- Successive: presented 1 @ a time: see A, respond to A.
- Get feedback.
- Successive: varies the order of presentation each time.
- Simultaneous: not given any feedback.
- If you respond to A & then B, there is nothing that tells you that B comes next and not C or D.
Training
- Simultaneous: trained that A comes first because if you don’t pick A, the trial ends.
- Can’t train backwards (learn that D comes last) because responding to D is extinguished because the trial ends.
- Less well understood & studied.
- Successive: learn about the last stimulus first because it is the last item that is rewarded.
- Incremental training procedure: ultimately only rewarded if they get the whole sequence.
- A long history of being studied as an explanatory device in learning theory.
- Lashley, Skinner, etc.: felt that most (all?) of the complex behavior we want to explain as psychologists is a matter of successive chains of S-R reinforcement, which then is used in higher-order conditioning.
- Stimulus A occasions response A, which ends up being reinforced by SB, which is a conditioned reinforcer –> complex chain of behaviors.
Barnabus the Rat
- Trained to push a lever to get food.
- Then, trained to lift the flag, then an elevator went down; the reinforcer for the response of lifting the flag is being lowered to the level of the lever; the sight of the lever became CR.
- Then, trained to go through a tube to get to the elevator: the elevator became CR, running through the tube was reinforced by the sight of the elevator.
- Etc.
- Eventually, even though he was placed on the floor with the lever, he would still start the complex series of behaviors, although he could get to the lever without going through the chain of behaviors.
How Pigeons Learn Lists: Similar to Numerosity Sequences
- The same 4 stimuli are presented on every trial, but the location of the stimulus is random.
- Reinforced for pecking stimuli in the correct order.
- Probe tests: pairwise tests, A vs B, B vs. D, etc.
Fig. 1: Baseline: pigeons trained in successive chains.
- Shows criteria needed to move on.
Fig. 2: Subsets.
- Responded to all orders above criterion except BC.
- Possibility: B & D, A & C, A & D are more discriminable because they are farther apart (distance effect).
- Pigeons learned rules: always learned that A comes first, D comes last, whatever comes in between is incidental.
- No set rule for middle items: no “end” element (A or D).
- Not necessarily learning everything about the chains themselves as much as they are just learning conditions & rules for different lists.
Chunking
- What is the pigeon’s memory capacity?
- A series of shapes were presented in a way that they could be chunked (organized by color), or that they were interspersed (controls) with alternating stimuli.
Fig 5
- Chunking groups took less time to correctly complete the sequence.
- Longest dwell time on C & D: chunked ABC, E’D’.
- Amount of time making the response.
Non-Chunking Groups
- Big dwell times because not chunked.
- Chunking aids list-learning.
List-Learning in Monkeys
- Better than pigeons given the same 5-item task.
- Responded above chance to all stimulus combinations, unlike pigeons, who had trouble with BC.
- Why?
- Smarter?
- Learning entire lists?
Fig. 9: Latency of responding, comparing monkeys & pigeons.
- As lists got longer, monkeys took more time to respond to later items in the list.
- Linear representation of lists: processing serially, have to mentally scan the list.
- Takes longer to respond the later in the list because they have to scan more items.
- Sternberg: this pattern in people.
- After presenting monkeys with a few lists, monkeys improved with novel lists (became more adept at trial & error methods), but pigeons didn’t.
Fig. 11: Knowledge of ordinal position.
- Chen et al.: monkeys trained to respond to 4 lists of pictures.
- Derived lists: took 1 stimulus from each list.
- Maintained: maintained order from original list (last image from original list was still the last image in the maintained list).
- Can retrieve knowledge from previously learned lists & apply them to derived lists as long as the ordinal position was not changed.
- Explanation: there must be a representation of the ordinal positions of the items in the lists.
- Or when there is a predictive value to each item on the list, it has temporal properties associated with its position (2nd stimulus is seen @ the same point in time relative to the items that precede it).
- So maybe it’s not the ordinal properties that are important, but the temporal properties.
Temporal Chunking
- Fig. 13 results from Bugs & Garbo (monkeys).
- How long does it take to respond to each list.
- Took a long time to respond to the 1st item, but once they had it, found the rest of the items fast.
- Either monkeys are:
- Selecting 1 @ a time.
- Planning the sequence: what was the sequence I learned, then think about the order in which they respond, then move & respond to all of them.
- Supported by Fig. 13.
- But how do you know that he is developing a plan?
- Or just looking @ all of the items 1st, then trying to find the 1st one, then they already know where all the other ones are?
- Another study: when locations are switched before response, it takes them longer to respond because it interrupts their plan?
- Was it the plan, or they were just surprised? Something attentional?
- Evidence of chunking & planning.
- Also tested in humans (Fig. 16).
- Humans naturally tend to pause in the middle of long list responding (8 items).
- Evidence of chunking: respond to 1 chunk quickly, then think of the 2nd chunk & respond to it.
- The pause signifies the end of the chunk.
- Always 2 chunks.
- The size of chunks varied from trial-to-trial within & across subjects.
- Always a pause, but where the pause was varied.
- Graph: took the longest latency for each trial & relativized latencies before & after it.
- The longest pause was 2 or 3x other latencies.
- Argument: chunking is occurring at retrieval, not encoding.
Input & Output Chunks
- Pauses are not encoded with stimuli; something is in the organism that divides lists in a somewhat unpredictable/arbitrary way into before/after chunks.
- The process has to occur in order for the 2nd of 2 chunks to be executed.
Importance of Chunking
- Regardless of what chunks may or may not be, we know that chunking/pausing in these tasks reflect how organisms organize their responses, a way to lessen WM load at the time of retrieval.
Brown
- Animals are controlled by patterns, learn patterns easily.
- Spatial relations among locations control rats’ behavior when looking for sugar pellets.
Varieties of Spatial Learning
(shark pics)◊◊beacon homing◊◊i’m looking 4 something,i know i’m getting closer b/c it’s getting bigger◊◊piloting◊◊landmark cues◊◊control by geometry◊◊dead reckoning◊◊based on the way my body feels,i feel like i’ve moved this far….◊◊how animals learn spatial relations w/o using landmarks: pole bx◊◊4×4 bx w/ poles baited on top w/ a treat◊◊treats R placed in a pattern◊◊e.g.,square◊◊interpretive problem: figure 4: can explain tendency after finding a baited pole 2 go 2 s locations by saying they’ve learned a pattern◊◊spatial generalization gradient: area surrounding pole increases in attractiveness,more likely 2 search nearby,if they search nearby,they R more likely 2 find another baited pole …◊◊spatial proximity is most powerful determinant of where they search◊◊data analysis is restricted 2 s’s & x b/c those R the only proximal locations 2 the pole they jst chose: don’t look @ choices where rat goes 2 another random location: controlling 4 spatial proximity◊◊not controlling 4 possibility that surrounding each of the places where food hs been discovered in this trial,there is a gradient of generalized excitation◊◊s+ hs a property (e.g.,hue) that varies on a continuum◊◊during testing,get a generalization gradient w/ lots of responding 2 s+,more responding 2 more similar stimuli,less responding 2 least similar stimuli◊◊in 3d space,there could be a gradient around each of the locations where food hs been discovered (fig.11)◊◊more likely 2 search around where food was last discovered b/c those locations R more similar (more excitation)◊◊food found in 2 locations -> 2 gradients that overlap◊◊where a place is affected by both gradients,there will be more generalized excitation◊◊w/ square pattern,locations that complete the square pattern get more generalized excitation than other locations◊◊no pattern learning needed◊◊but evidence from pattern learning comes from performance w/ other kinds of patterns◊◊line pattern in particular is important: if the pattern is the line,then U still get more generalized excitation 2 the locations that complete the square,but U see responding consistent w/ the line pattern◊◊happen upon the 1st treat by chance◊◊1ce they find the 1st 1,knowledge of the pattern makes it easier 2 find the other 1s◊◊not using odor: finding the 1st 1 by chance◊◊after 120 daily trials,chose poles consistently in a square pattern,more often than expected by chance◊◊as soon as they find 1,more likely 2 find another 1◊◊fig.4: 1ce they find 1,stick 2 same row or column 2 find 2,then have 2 switch 2 other row/column 2 find s (x is an error)◊◊linear pattern: all 1 row◊◊locating an unbaited pole provides potential information◊◊if i don’t find 1 in this row,i don’t have 2 look in this row anymore◊◊acquire tendency 2 choose poles in same column after finding a baited pole◊◊checkerboard pattern: every other◊◊have an abstract representation of what the pattern is◊◊can make 3 kinds of moves◊◊skip: stay in same row,skip the next 1,go 2 the 3rd 1◊◊these moves get U food going horizontally across a row◊◊not jst learning that U get food from going horizontally◊◊put in diagonal blocks so they couldn’t do skip moves during training◊◊take out barriers: then skip,but never trained 2 skip◊◊not jst learning a motor/response pattern◊◊have a representation of the pattern◊◊using wm 2 remember where they’ve already been,turning where they’ve already been into a landmark?◊◊would cues facilitate pattern learning by helping them distinguish btw locations?◊◊also avoid revisiting◊◊tell them where they’ve visited by putting a spotlight over the location: so bright,it scared them/blinded them? didn’t help◊◊light from below –> more likely 2 avoid revisits (helps wm),but also didn’t help spatial pattern learning◊◊wm is not involved in pattern learning◊◊more likely dead reckoning◊◊transitioning from 1 baited pole 2 the other creates a vector,as vectors change,tells U something about the pattern◊◊doesn’t explain ability 2 choose baited pole after an unbaited pole◊◊maybe 2 wm systems?◊◊storing where i’ve already been: not being used◊◊storing location of where U found food◊◊from knowledge about where i’ve been & where i found food,i can extract information about spatial relationships between locations◊◊spatial relationships among locations R abstract,not connected 2 the locations themselves◊◊use of dead reckoning system: keep track of where we’ve been in space,integrate from that where we R relative 2 where we started◊◊if U know distance,speed,& time,U can calculate where UR in relation 2 where U started◊◊relation is independent of specific location,learned independently: don’t have 2 know the location of where U started◊◊4/9: metacognition◊◊definition: awareness of 1’s own perceptions & memories◊◊metamemory: ability 2 recognize the strength of our own memory & determine future actions based on that◊◊metacognition: judgments of our ability 2 make perceptual discriminations (?)◊◊facilitates efficient behavior◊◊hampton (2001)◊◊1 of the 1st successful attempts 2 show that animals can exhibit metacog monitoring & control◊◊methods (fig 1)◊◊study phase: image on screen◊◊delay interval (static: exp 1)◊◊choice phase (if they have a choice): end trial or take test ◊◊exp.2: 2/3 of the time,have a choice 2 end early & avoid taking the test (receive a lesser reward)◊◊1/3: forced trial◊◊test phase◊◊results◊◊fig 2: exp.1◊◊animal chose 2 continue –> btr performance than on forced trials◊◊evidence that they choose 2 avoid trials they don’t know answer 2◊◊no option 2 opt out –> forced 2 answer when they don’t know the answer –> worse performance◊◊exp.2: maybe external cues make it more likely 4 animal 2 want 2 end trial & get smaller reward (pellet)◊◊probe trials: during study phase,no stimulus presented◊◊results should differ from non-probe trials if they R making decisions based on the study stimulus in non-probe trials◊◊no right answer in test phase◊◊critical is choice phase: shouldn’t choose 2 continue b/c have no memory of a test stimulus◊◊only rewarded if they decline (get lesser pellet reward)◊◊monkey 1 always avoids test in probe trials◊◊never learned negative consequence of choosing the test◊◊monkey 2 takes longer 2 learn?◊◊fig.3: more likely 2 opt out of the test in probe trials◊◊absence of memory causes monkeys 2 decline tests◊◊if it was another external cue,both normal & probe trials would have been affected◊◊evidence of metamemory: cue = internal stimulus,status of memory◊◊exp.3: make sure they’re not using timing as a cue◊◊during training,had delay time set 2 about half a minute◊◊customized 2 each monkey 2 get choice accuracy jst right (not @ floor or ceiling)◊◊4-alternative match-2-sample: chance = 25%◊◊fig.4◊◊m1◊◊more likely 2 end trial early as delay got longer◊◊more uncertainty◊◊choice trials: performed fairly well◊◊forced trials: steady decline with increased delay◊◊as delay interval gets longer,memory trace of sample decays◊◊m2◊◊didn’t decline as many tests◊◊steady decrease in choice & forced trials◊◊jst not as good @ making metacog judgments in the delay exp? overconfident? individ differences?◊◊evidence of animal model of the ability 2 make adaptive decisions 4 future behavior based on assessment of current state of knowledge◊◊foote & crystal (2007)◊◊also have uncertainty response◊◊duration discrimination task◊◊white noise persists 4 various times;choose whether it is short or long◊◊more difficult 4 middle durations◊◊fig.1: ◊◊study phase◊◊choice phase: choose or forced◊◊choose: 2 choices 4 nose poke: decline test or take test◊◊test phase: 6 pellets if correct,0 if incorrect,3 if declined◊◊fig.2◊◊a-d: index of stim difficulty x % declined◊◊trend of rats ending expt early as it became more difficult◊◊e-h: % correct◊◊less accurate with increased difficulty◊◊up 2 most difficult discrimination: then accuracy was btr on choice trials than forced trials◊◊evidence 4 metacognition◊◊started w/ n=8;ended w/ n=3◊◊5 out of 8 didn’t meet criteria 2 participate: didn’t learn procedure◊◊index of stimulus difficulty◊◊short = 2-3.62 s = not difficult◊◊long = 4.42-8 s = not difficult◊◊intermediate = 3.62 & 4.42 s = intermediate◊◊shortest = 2,longest = 8;4 = geometric mean◊◊these values R on logarithmic scale◊◊based on equal distances in proportion rather than absolute values of time◊◊index of stim difficulty is on this log.scale: how far is stimulus from geometric mean of 4◊◊smith,beran,couchman,coutinho (2008)◊◊criticism of metacog research (including their own)◊◊confounding variables in research◊◊limitation◊◊question = can these metacog studies be explained by something simpler?◊◊issue of reinforcement◊◊uncertainty response still gets rewarded = form association btw uncertainty response & reward◊◊get nothing 4 wrong answer = turn it into an aversive response◊◊choose uncertainty = avoiding aversion?◊◊simulation based on signal detection theory◊◊fig.2: temporal discrimination procedure shows that simple signal detection can create results that look like metacognition◊◊simulation replicates sparse-dense results using signal detection theory: fig.3◊◊**only responsible 4 1st simulation**◊◊replication of foote & crystal discrimination task◊◊as task becomes more difficult,uncertain response becomes more likely◊◊forced trial accuracy is worse than chosen trial accuracy◊◊signal detection explanation of results◊◊ideally,2 things can happen: white noise can be 2s or 8s long◊◊psychologically,the rat hs a scale of impressions◊◊continuum: really short 2 really long◊◊present 2s stimulus –> it is really 2s,but sometimes it can seem really short,& sometimes it can seem longer b/c there is a distribution of impressions◊◊d’ = how well can durations be discriminations◊◊beta = threshold/criterion that classifies boundary between long & short◊◊everything lft of threshold is classified as short,everything right is long◊◊present a 2.97s & a 5.38s stimulus◊◊d’ is smaller,more overlap btw distributions,more likely 2 make errors◊◊don’t have 2 choose short or long: can choose uncertainty response◊◊in metacognitive terms,rat is sensitive 2 where on the scale of psychological impressions the intermediate durations lie: in the overlap◊◊more of the distributions lie in the “uncertain” range when the distributions overlap more (vs.when they don’t overlap as much)◊◊in sdt terms,there R 2 thresholds b/c there R 3 categories (short,long,uncertain)◊◊no awareness or representation of where we R on scale of psychological impressions◊◊jst distributions of effects of stimulus◊◊w/ intermediate threshold,it is equally likely 2 fall in “short” range or “intermediate” or “long” ranges◊◊more overlap = even more likely 2 pick intermediate/uncertain than short b/c more proportion of the distribution falls in the middle of the 2 thresholds◊◊how does model explain why chosen trial accuracy is btr than forced trial accuracy? authors don’t really explain.◊◊chosen trials: if discrimination falls btw 2 thresholds,respond uncertain◊◊accuracy on a chosen trial is proportion of distribution that results in a correct response/hit (proportion that falls into uncertain range is not included)◊◊accuracy on a forced trial is proportion of hits (1 threshold): not as large as proportion of hits in chosen trials◊◊smith proposed a btr measure of the uncertainty response◊◊stop reinforcing 4 uncertainty response◊◊reinforce @ the end of the block rather than on a trial-by-trial basis (like getting a grade @ the end of a test)◊◊ways of judging uncertainty that don’t create a continuum of reinforcement◊◊alternative ways of measuring metacog in animals don’t jst raise questions about whether animals have metacog◊◊raise questions about whether subjective state of uncertainty is important◊◊maybe human performance can also be explained by simple sdt/associative processes,with subjective experience superimposed on it◊◊mike book: “along 4 the ride”◊◊our subjective phenomenology/consciousness that we spend a lot of time in psych studying is jst along 4 the ride: not really what causes behavior,superimposed on cognitive processes ◊◊◊◊4/16: tool use◊◊next week: don’t read kubinyi;following week: don’t read s-14◊◊context (mike)◊◊evolutionary history of cognition in humans: tool use was thought 2 play a big role in this 4 a long time◊◊~25 years ago: most thought that only humans use tools◊◊depends on definitions: what’s a tool?◊◊physical object that’s not part of yur body that is used in a functional way◊◊if used this way,many animals use tools◊◊spider webs? came out of the body….◊◊nuptial gifts in scorpion flies? ◊◊leaf cutter ants farm food: R the leaves they plant a tool?◊◊beaver dams: create a place where fish R more likely 2 thrive so they can catch & eat them◊◊over time,more people started thinking that other kinds of animals use tools (or tool-like devices)◊◊what is it about human (& other special cases of) tool use that is worthy of our attention as cognitive psychologists?◊◊rake use in monkeys (paper from biopsych)◊◊neural signature indicated that they started 2 understand the rake as an extension of themselves◊◊extended mind hypothesis◊◊journal becomes an extension of yur mind◊◊natural selection can also operate on traits that R extended by our effect on our environment (extended phenotype)◊◊tools R where the extended phenotype meets the extended mind◊◊bluff et al.(2007)◊◊questions about tool oriented behavior◊◊epigenetic processes◊◊cognitive processes◊◊development of tool use (p.6)◊◊a set of genetically determined rules (like building nests)◊◊advanced reasoning abilities (like folk-physics)◊◊devel.of tool use b/c of reinf.history (ability 2 physically manipulate objects)◊◊social learning◊◊studies of using & manufacturing tools◊◊reared 4 crows in the lab◊◊2 were shown daily demos of tool use by humans (extracting food from holes w/ sticks)◊◊other 2 received same amt of human contact,fed near holes◊◊what is the effect of social learning?◊◊by 79 days,all 4 retrieved food from crevices using sticks◊◊tutored birds also spent more time using & manipulating the tools (carrying them around,inserting them in things)◊◊can’t jst be social transmission: non-tutored birds also developed tool use,so social transmission is not necessary◊◊social transmission might play a role in refinement of tool use,local enhancement◊◊local enhancement: if a human picked up a stick & handled it in front of the crow,the crow might also want 2 handle it;enhances attractiveness;motivational change that occurs in response 2 social stimulation◊◊all showed stereotyped precursor behavior◊◊e.g.proto-probing: hold the stick in its mouth & rubbing it back & forth against a surface◊◊suggests that there is some genetic predisposition 4 some behaviors that may be involved in the dev.of tool-use behaviors◊◊precursor behs R not reinforced w/ food,so they R not dependent on reinf history◊◊manufacturing tools◊◊wild crows rip pieces off of pandanus leaves & use them as tools◊◊hand-reared crows: @ 2 or 3 months old,all learned 2 use these leaves as tools◊◊@ 6 or 7 months,all ripped @ leaves,but only tutored crows & 3rd 1 (read paper 2 figure out the rest)◊◊suggestion of cultural transmission of how the tools R made,what type of tool is created◊◊seems 2 be genetic influence that can be affected by social transmission◊◊deployment of tool-use behs◊◊if these behs develop from a genetic disposition,doesn’t mean these behs have 2 be stereotyped◊◊cog processes that underlie tool use◊◊folk-physics: common sense understanding of how the world works◊◊does this promote evolutionary dev of tool use,or does tool use promote evol devl of folk physics?◊◊what does animal understand about tool use?◊◊when crows R given a number of tools 2 choose from,natural beh of tool use is not rigidly specified,indicative of presence of folk physics◊◊2 crows given tools that R 10 diff lengths◊◊food in tubes @ diff depths: specific tool needed◊◊able 2 select tool of appropriate length◊◊when given active trap,learn 2 push out food so it doesn’t fall in the trap◊◊when switched 2 inactive trap,continue 2 push the food out the same way even though the trap will not trap the food◊◊non-tool-user tube trap (p.15)◊◊trained on a,transferred 2 b;seem 2 transfer;also seem 2 transfer from b 2 a (so not jst learning 2 avoid the black disk generally;specifically pulling away from black disk on the butt)◊◊using assoc rule that they pull away from the trap with the black disk on the butt?◊◊½ trained on a,½ trained on b,transferred 2 c & d◊◊have 2 not follow the rule they followed before◊◊1 rook successfully transferred,other 5 were not successful◊◊associative learning can affect tool use◊◊argues against predisposition 2 figure out folk physics causality◊◊given 3 tools,diff diameters,2 bundled together & 1 separate◊◊prefer the tool of the smallest diameter,even if they have 2 disassemble the bundle◊◊capable of manufacturing appropriate-sized tools◊◊field expt: bx w/ transparent side,food @ differing depths: 1st tool crow used did not differ in length depending on depth of food;if 1st tool was too short,would try a different 1◊◊have some concept of folk physics;not blindly guessing;but not always being utilized properly◊◊hook making◊◊initial trials R critical,because 1ce they have experienced the tool,there is learning◊◊aluminum strips: bending them in the same way as the wire is ineffective◊◊have 2 learn proximal bending: bend the part that is in their beak◊◊having the same initial actions as when they were given the wire would indicate trial-&-error learning◊◊goal-directed actions,intent on creating a hook shape◊◊will bend the end of the strip & then try 2 use the straight edge sometimes;maybe don’t understand why they need a hook?◊◊unbending action◊◊learn 2 do the opposite of bending strip into a hook◊◊but maybe initial trial was accidental◊◊“understanding” is not all-or-nothing,but a continuum◊◊human example: extracting souvenir badges out of a tube◊◊initially attempted 2 do it without a tool,wouldn’t use it right;made the same mistakes crows used◊◊so people & animals learn by doing◊◊no test 4 understanding causality◊◊R there levels of understanding?◊◊bird & emery (2009)◊◊tool use was 1ce thought 2 be unique 2 humans◊◊crows betty & abel used tools;crows do this both in the wild & in captivity◊◊rooks R in the same family as crows◊◊high physical intelligence◊◊don’t seem 2 use crows in the wild◊◊4 crows◊◊all have experience pulling sticks 2 get food◊◊enclosure is full of rocks & surrounded by wire mesh◊◊pre-training◊◊apparatus: cube with platform;if U put w8 on the platform,the worm falls out the butt◊◊have 2 put w8 down the tube 2 dislodge the platform & drop the worm out◊◊learn (@ 1st accidentally) 2 get the worm out,then transfer 2 other configurations◊◊might learn through social imitation (all trained together)◊◊quick 2 transfer 2 other configurations◊◊testing◊◊isolated from other birds;prevents social learning during testing◊◊given tools placed next 2 the apparatus◊◊st1 size test: choice of st1 sizes;all were functional in large tube◊◊all 4 preferred largest st1 4 large tube (fig.1 a)◊◊more salient?◊◊when given small tube,choose small st1 80% of the time (fig.s2)◊◊able 2 select tools based on functionally relevant properties,not jst perceptual properties◊◊st1 selection test◊◊have 2 leave the testng room & go back 2 the run 2 select the tube◊◊have 2 remember what size the tool is◊◊4 small tube,choose progressively smaller rocks as they learn larger rocks 1’t work◊◊st1 orientation test◊◊all subjects chose long,thin tube regardless of tube size◊◊easier 2 handle?◊◊manipulated st1 differently depending on size◊◊could drop st1 into large tube,had 2 rotate st1 so it is vertical 2 drop it into small tube◊◊¾ did it on 1st trial,other did it on next trial◊◊how could they have learned 2 do this? did they have any relevant previous experience? maybe.◊◊author argues they understand the relationship btw the shape of the objects & the possibility of the object falling down the tube◊◊affordance: functional properties of objects–not jst physical properties–can be perceived;the climbability of stairs,the grabability of a glass;nothing beyond perception is necessary 4 the functionality of an object 2 be detected;it affords that function;sign of a good tool = U can see what 2 do with it◊◊is this sort of perception indicative of understanding?◊◊cognitive model in the mind of an animal that allows it 2 detect which tool will work◊◊stick use test◊◊no previous history of being reinforced 4 using stick◊◊all birds used the stick jst as they used the st1 on the 1st trial◊◊dropped in large sticks like large st1s;given light,thin sticks –> push down,add force 2 collapse the platform◊◊fig.2a◊◊evidence 4 detecting affordances of tools◊◊this 1 can be dropped,this 1 can be pushed◊◊non-functional st1 vs.functional stick (& vice versa): chose functional object◊◊indicate goal-directed action: push down with the stick even though they have never been reinforced 4 this◊◊fig.3a & 3b◊◊metatool test◊◊use a tool 2 gain access 2 another tool 2 gain access 2 reinforcement◊◊various difficulties◊◊subject must recognize that a tool can be used on a non-tool object◊◊must resist immediate motivation 2 use that tool 2 get the food◊◊behs must be organized hierarchically◊◊given 3 bx set-ups: middle hs worm,side 1s have large & small st1s,respectively◊◊given large st1 that would fit in either side,not in middle◊◊all subjects solved task from 1st try: used large tool 2 access small tool,used small tool 2 access reward◊◊stick modification test◊◊can’t use stick unless they break the side branches off◊◊all birds successfully modified & used tool (fig.4): relationship btw # of modifications needed & made◊◊hook use◊◊gave them functional hooks;had 2 pull up bucket rather than push down◊◊all 4 successfully solved task,¾ on trial 1◊◊all inserted hook end into tube more often than straight tube◊◊gave them functional hooks & non-functional hooks◊◊¾ chose functional hooks on 1st trial,all learned it eventually◊◊many perceptual similarities◊◊the function is what is important◊◊manufacturing hooks◊◊given straight piece of wire◊◊¾ made hook on 1st trial◊◊only successful on lifting bucket on 35% of trials◊◊fig.6◊◊tool use seems 2 be a useful byproduct of general cognitive abilities,rather than a domain-specific evolved process◊◊not jst an adaptation used 2 get food◊◊related 2 general intelligence◊◊crows live on islands with no mammals,no competitors,no corpses that they can get food from,so exploit ability 2 forage 4 bugs in trees◊◊rooks don’t need a tool 2 access their food◊◊difference in motivation rather than difference in ability 2 use tools◊◊both have the cognitive capacity/general intelligence◊◊argues against work that indicates that tool use leads 2 general intelligence◊◊maybe we use tools because we evolved 2 be smart,not that we’re smart because we evolved 2 use tools◊◊evolutionary homology vs.evolutionary analogy◊◊homology = share functional trait w/ another organism that UR closely related 2◊◊common 4 historical reasons: common ancestor had that trait◊◊corvids can use tools b/c common ancestor of corvids was a tool user◊◊analogy = develop similar functional traits/abilities w/o being closely evolutionarily related◊◊b/c of similar requirements in yur environment◊◊e.g.,eyes may have evolved 5 or 6 times independently;different kinds of eyes,but many functional similarities;even though there is no recent common ancestor◊◊4/16: social cognition◊◊shettleworth chapter◊◊broken wing display (ch.12,p.433)◊◊presence of predator occasions this display◊◊mostly in ground-nesting birds◊◊occurs primarily when the bird hs eggs or chicks in the nest◊◊parent bird (males & females,depends on species) flies away from the nest & then engages in broken wing behavior◊◊parent usually displays in a direction that is between the predator & the nest,leading the predator away from the nest,not past it◊◊most of the time parent ends up being closer 2 the predator than it was◊◊only makes sense if U attribute beh 2 drawing predator away from young◊◊bird modulates the behavior of the predator,modulating duration,intensity,other properties of the display in response 2 whether the predator is continuing 2 approach or move away from the nest◊◊ristau: this behavior requires a representation in the displaying bird of the desire of the predator 2 attack its young: purposefulness is being used as the core of the explanation 4 the behavior◊◊is this conclusion/assumption forced by the data?◊◊jst a simple s-r response?: see the predator –> engage in display,doesn’t require representation of intentionality◊◊intention = goal (get the predator away from the nest)◊◊intention also involves “aboutness,” or mental states that represent things/ideas out in the world◊◊1st-order intentionality (p.433): planning 2 do something/having an intention◊◊2nd-order intentionality: referent of that intention is another intention;representation is another representation;Ive an intention about what yur intention is◊◊theory of mind◊◊interpretations◊◊need 2nd-order intentionality 2 see this beh;need 2 represent intention of predator◊◊more parsimonious,but does it add anything 2 our understanding of the beh.? does it explain anything?◊◊call it a mental state,assume that it involves a subjective mental state,etc.–> have we really explained the behavior jst because we gave it a name?◊◊need another explanation in terms of contingencies,s-r relationships◊◊intentional stance (dennett,p.433): we should assume that mental states exist when they appear 2,rather than assume that they don’t (or should we assume nothing? mike)◊◊monkey fairness◊◊worked socially 2 solve the problem (1 monkey gave the other monkey a rock so that he could access a jar of nuts,he then shared the nuts with his partner)◊◊requires communication,cooperative behavior◊◊pris1r’s dilemma: ◊◊2 agents,opportunity 2 cooperate with each other in order 2 obtain a goal◊◊if both cooperate,both parties benefit (3 points)◊◊if 1 cooperates & the other defects/cheats/doesn’t provide nuts,1st player doesn’t benefit (0 points)◊◊if 1st player cheats & other cooperates,1st player benefits a lot (5 points),more than he would if he cooperated◊◊if they both defect,something bad happens (1 point),but not as bad as if U cooperate & the other defects ◊◊cooperation is mutually beneficial,but if U cheat,U can do btr than other player◊◊risk: if both cheat,U also don’t do well◊◊why ever cooperate? this is the dilemma◊◊cheating can be beneficial,but hs fewer benefits than cooperation in the long run (iterated pris1r’s dilemma: run multiple trials)◊◊in order 2 do iterated pris.dilemma,need 2 meet several criteria◊◊recognize others,know who is who◊◊know who did what in exchanges with U: i’m not cooperating with this guy because he cheated the last 3 times◊◊reciprocal altruism: i’ll help U if U’ll help me;helping U really helps me as well because U will help me in the future◊◊social intelligence hypothesis (p.418)◊◊social interaction was the evolutionary precursor 2 higher intelligence◊◊evidence that animals who R social in the wild gauge their level of cooperative behavior as a function of the perceived distribution of cooperativeness in others◊◊able 2 keep track of all of the social interactions in a group: led 2 selection 4 high cognitive capacity/intelligence because it makes it easier 2 keep track of social transactions◊◊brown,farley,lorek (2007)◊◊radial arm maze◊◊taps into what rats’ ancestors did in nature: find bits of food in patches distributed in particular places,keep track of places they’ve depleted of food 2 increase foraging efficiency;maybe why they R so good @ this task◊◊will rats keep track not only of where they have depleted food,but also where a foraging partner depleted food◊◊social memory: what is being remembered is another rat◊◊long-term goal: see if memory that hs social content (4 another animal’s actions) is special in some way◊◊social intelligence hypothesis?◊◊specialized cognitive/learning systems 4 social information?◊◊evidence 4 spatial,temporal modules;how about social?◊◊exp.1: skip◊◊exp.2 (hs exp.1 in it)◊◊maze hs closed arms (tubes): unusual;allows rats 2 pass each other without pushing each other off◊◊potential downside: reduces visual cues the rats can use,when they seem 2 naturally use visual information during this task◊◊test pairs of rats◊◊always cage mates (familiar with each other)◊◊as they make choices,do they keep track of the choices the other rat makes?◊◊fig.4: results of free choice procedure◊◊locations visited by other rat most recently◊◊doesn’t necessarily involve memory b/c rat might still be present in or near most recent location;disentangle anything that hs 2 do with memory from effects that might be elicited by the physical presence of the other rat◊◊physical presence of other rat can be an attractive stimulus◊◊open symbols: estimate of choices U would expect rat 2 make by chance (choice alternatives)◊◊rat can only choose 1 arm @ a time;arm fits into 1 of 3 categories: hasn’t been visited by other rat,most recent choice of other rat,earlier choice of other rat◊◊algorithm considers all 8 arms of the maze 4 each choice (of which the rat chooses 1),each gets classified into 1 of these categories◊◊filled-in symbols: choices made by rat◊◊only represent 1/8 as much information as the open symbols◊◊hs tendency 2 choose that kind of maze arm if the prob of choosing that kind of maze arm is greater than the proportion of arms that fit into that category◊◊further classify arms according 2 whether they R places the focal rat itself hs visited◊◊visited by other rat most recently◊◊tendency 2 visit a place more than expected when other rat visited it recently (filled-in dots larger than open dots)◊◊other rat might still be in this location: physical presence of other rat is attractive stimulus that draws in focal rat◊◊makes central finding even more compelling: in opposition 2 attractive nature of social stimulus,there is a tendency 2 avoid arms visited by other rat a long time ago◊◊most interested in maze locations visited by other rat earlier in the trial◊◊more likely 2 avoid remote choices (choices made earlier)◊◊squares: already visited by focal rat◊◊4 places focal rat hs not been 2 yet,if other rat was there remotely,focal rat is only half as likely 2 go there as U would expect based on the distribution of those arms in the environment◊◊as trial progresses,distribution of choice alternatives changes (U & other rat have visited more places,fewer arms not visited)◊◊dominant rat tends 2 make choices faster: 4 dominant rat,there aren’t as many places it hasn’t visited yet as there R 4 its partner◊◊doing correlational analyses b/c U aren’t manipulating choices of rats◊◊analyses were d1 twice: 1ce from perspective of 1 rat,1ce from the other◊◊forced choice phase: attempt 2 control the situation◊◊put observer rat in a cage,watches stimulus rat make 4 (forced) choices,stimulus rat removed,observer/subject is allowed 2 make choices◊◊exp.3 determined that odor from model rat does not affect choices◊◊subject is more interested in places that were not visited by model◊◊measure of subj rat’s preference 4 places that were visited by other rat: mean serial position of visits (battery dying,reread)◊◊small effect b/c rats can’t interact◊◊4/30: anthropomorphism◊◊blumberg & wasserman (1995)◊◊argument from design◊◊teleological argument 4 existence of god–there must be a designer◊◊things in the world R so complex & perfectly adapted 2 their purpose that they could only have been created by a deity◊◊ex.animal behaviors R functional,so animals must have minds that guide them◊◊complexity in human behavior is so great that there must be a design/designer◊◊where did alleged design come from?◊◊god?◊◊god : natural world :: consciousness/mind/thoughts : complex behavior◊◊human experience: U can will yourself 2 do things (free will?)◊◊causal connections between thoughts/subjective states/consciousness/goals & behavior◊◊both wasserman & wynne point 2 this in a derogatory fashion: mentalism is bad◊◊mentalism: we attribute mental (cognitive) states 2 things with certain properties (other humans,animals,cartoons,robots,etc.),also believe in our hearts that those mental states R causally effectatious,causal agents of behaviors we observe in those other entities◊◊if this was tru,subjective thoughts could be part of explanation 4 why we see complex behaviors in humans & animals◊◊if not,not part of explanation◊◊why criticize mentalist way of thinking about mental states & their relationships 2 behavior?◊◊goodrich & allen critique of wynne (p.147)◊◊a way of coming up with a theory is by being anthropomorphic◊◊might be useful as a technique 4 coming up with hypotheses,ideas 2 test◊◊blumberg critique of wynne (p.145)◊◊being anthropomorphic isn’t a good technique 4 coming up with theories◊◊R these theories/hypotheses legitimate theories that include objective states being causally effectatious?◊◊mentalism is half of cartesian dualism◊◊dualism is not scientific: can’t test the existence of a mind◊◊alternative: materialism◊◊objective world consists of matter & energy◊◊behavior exists in the objective world◊◊the brain exists in the objective world◊◊w’v experience◊◊subjective experiences of memories,thoughts,etc.act on yur brain,cause alterations in behavior through its effects on yur brain◊◊steven pinker: the mind is what the brain does◊◊there is a world of experience,we all have 1,we share it & communicate about it◊◊it is all 100% dependent on the objective world of matter & physics in the form of the brain◊◊causal arrow only goes 1 way: experience is a function of brain states◊◊subjective states/phenomenology/experience can’t have any effect on the objective world of matter & energy: modern materialist view of existence◊◊subjective states can’t play explanatory role 4 behavior,even though it feels like they can◊◊subjective state is an effect of a complicated combination of static & dynamic properties of my brain & everything that hs had an effect on it over time◊◊what if a subjective state was jst a brain state?◊◊memory◊◊personality traits◊◊emotions◊◊cognition = shorthand 4 set of brain states that we don’t know much about,but in principle could be specified in terms of mass & energy;exist in the physical world◊◊what about reinforcement histories?◊◊maybe they affect human personality traits but not animals◊◊need 2 specify this in a way that makes it testable◊◊could hypothesize that chimps could be classified in terms of neuroticism◊◊becomes a scientifically useful proposition if it comes with a prediction that could be tested◊◊if U say a chimp eats grapes b/c a chimp likes grapes,if U mean this as a non-material subjective state,it is not a scientifically testable proposition,because the existence of this subjective state can only be affected by the physical world,not the other way around◊◊cyclical/nominalist fallacy: jst renaming the construct as a subjective state◊◊even if subjective mental states have no place in explanatory theories,aren’t they still a legitimate topic 4 study?◊◊could be targets of study even if they R no good as theoretical explanations◊◊blumberg: doing this hinders our progress◊◊how would U study subjective states? how would these results inform theory?◊◊aren’t scientific results always interpreted subjectively?◊◊instrumentalism: use our subjective states as a tool 2 interpret things◊◊scientific method provides a way of coming close 2 understanding objective reality◊◊how do U measure/observe subjective states?◊◊introspection?◊◊how does anthropomorphism relate 2 this?◊◊we know how it feels 2 be a human◊◊we think we know how others feel as humans because we can communicate about our subjective experiences◊◊animals may or may not have subjective experiences–no evidence 4 it,comparitive cog psych hs nothing 2 say about it◊◊let’s stipulate that there R probably other entities that have subjective states;R they like our human subjective states?◊◊imagine a person with no subjective experience;everything else is the same,all behavior is the same;couldn’t tell whether they had subjective experience or not◊◊anthrocentrism: easier 2 say that other humans have subjective states,not other animals◊◊we don’t want 2 be explained◊◊placebo effect/expectancy effect◊◊is it a matter of the mental state U have when some1 tells U this pill is an aspirin,or hs yur experience set up an expectation (set of brain states) that affects yur experience◊◊if 2 brains have the same set of structures (human & rat) & same brain states,is the subjective experience the same?◊◊structures may have evolved 4 different reasons: R they doing the same thing?◊◊if we know what necessary & sufficient brain states R 2 produce a mental state,would it be possible 2 evaluate the existence of these things in non-humans?◊◊embodiment argument: different 2 be in a human body than a rat body◊◊wouldn’t yur subjective experience be different without yur body? in a different body?◊◊i know that copper conducts electricity;if i find another piece of copper,i can assume it conducts electricity,right?◊◊why R we not comfortable making these assumptions about brains?◊◊if jung put his brain in alex’s body,is it jung? would it have the same subjective state that jung had?◊◊how do U ask animals about their subjective states?◊◊what do we gain by inferring that another animal hs this subjective state? what value does investigating this question offer?