Cognitive Psychology: Test 3

Hebb rule

"neurons that fire together wire together" 1) Stimulus is first presented 2) As a stimulus is repeated, the synapse is strengthened by causing structural changes. 3) After many repetitions, more complex connections have developed between the 2 neurons, which causes an increase in firing rate and greater neurotransmitter release, even though it is the same one that was originally presented. Change that occur in the synapses that are activated around the same time by a particular experience provide a neural record for that experience. Memories for an experience cause changes in many thousands of synapses, and a specific experience is probably represented by the pattern of firing across this group of neurons.

Serial Position Curve of DRM Lists: Free Recall

-the primacy and recency affect is shown -people falsely recall the critical lure at about the same rate as middle of the list items that were presented -there is a small rate of intrusions (saying words that aren't on the list)

DOJ Recommendations

1) Ask open-ended questions and avoid suggestive questions 2) A lineup should be serial rather than simultaneous -shown one picture at a time 3) Eyewitnesses should be told the culprit might be or might not be in the lineup 4) Ask witnesses to state, in their own words how certain they are immediately before they give any feedback -research shows that high confidence measured at the time of identification is associated with more accurate identifications 5) Use fillers who are similar to the suspect 6) Use a blind lineup administrator

Systems Consolidation: Multiple Traces Model

1) Connections between the cortex and the hippocampus are initially strong and connections between the cortical areas are weak. 2) As time passes, the inter-cortical connections strengthen and the hippocampus-cortical connections remain.

Potential problems with hierarchical semantic

1) Couldn't explain the typicality effect (graded category membership) ex: the statement "a canary is a bird" is verified more quickly than "an ostrich is a bird" but the model predicts equally fast reaction times because "canary" and "ostrich" are both one node away from "bird" 2) Some response times findings don't fit ex: people are faster to verify that a pig is an animal than a pig is a mammal -"a pig is a mammal" should be verified more quickly according to the model because the link from "pig" leads directly to "mammal" and then "animal" 3) Specific categories are stored far away from their most fundamental properties

What are concepts good for?

1) Guiding attention and perception ex: focusing attention on a printer since you know from your concept of a kitchen, a printer is not supposed to be on the stove 2) Memory ex: misremembering what was in the office based on the concept of what is usually in an office 3) Inference, generalization, and prediction 4) Communication -since we share certain concepts, this allows conversations to be better understood

Components of the Illusory truth effect

1) Might occur because repeated information is accompanied by a sense of processing fluency 2) This could lead to a backfire effect when false information is debunked ex: Later on you hear the same claim and you think it sounds familiar but you don't remember if you were told that the situation was false (even worse when there was a distraction while they were telling you information) aka: People might find the information easy to process (high fluency) without remembering that they were told it was false

You can express confidence in...

1) Perceptual judgments 2) Source memory 3) Prospective memory 4) General knowledge 5) Decisions: sure this is the right thing to do vs. not as confident

Properties of Connectionist models

1) The circles are units, each defined by an activation level -represents the neurons in the brain 2) The lines are connections -represents axons that transfer information between units 3) Some of the units have specific labels ex: input units: units activated by stimuli from the environment ex: output units 4) Others don't have labels ex: hidden units: an intermediary step between the inputs and the outputs 5) Information is stored in connection weights: determines how signals sent from one unit either increase or decrease the activity of the next unit -represents the increase or decrease in firing rate of the next neuron -darker shading indicates activity -high weights: strong tendency to excite the next unit -low weights: cause less excitation -negative weights: can decrease excitation or inhibit activation of the next unit 6) Activity spreads in the forward direction (input -> hidden -> output) 7) Error signals flow backwards

Hierarchical semantic networks

1) Well-organized set of propositional statements 2) Avoids redundancy by storing properties at higher levels 3) Supports generalization for newly learned properties and exemplars 4) Lower-level, specific concepts are acquired later and lost earlier But, at the algorithmic level, is this really the representation we use? Is it this neat and organized? -this may not actually be a perfect explanation for using concepts

Concepts overlap

A given item could belong to multiple categories Global level (superordinate): furniture -may not be as informative, but allow for a lot of generalizations Basic level: table Specific level (subordinate): kitchen table Most people pick the basic level name ex: saying fish (basic) rather than animal (global) or trout (specific) Going above the basic level results in a large loss of information and going below results in in little gain of information

Acquiring conceptual knowledge

A kid is shown a picture of a horse and they think it is a dog -this is the phase where they over generalize (every 4-legged animal is a dog to them) As you grow, you are taking big, undifferentiated concepts like a dog and as you learn more, you can fill in the details of the lower hierarchy

Real-life flashbulb memories

A vivid, long-lasting memory for unexpected emotionally laden, and consequential events -feels like you mentally took a picture of it and you can relive it (not actually like photographs though) -refers to memory for the circumstances surrounding how the person heard about an event, not the memory for the event itself ex: can have group flashbulb memories like 911 or when school was closed for covid They may seem more accurate than they really are!

Simulating semantic dementia

Add random noise to the connection weights What if the boundaries between concepts got blurry? -could start to overgeneralize high-level properties (pine trees have leaves, birds have 4 legs, etc.)

An alternative: Connectionist model

Also known as "parallel distributed processing" (PDP) models or neural network models It is a alternative algorithm-level theory of semantic knowledge -NOT implementation-level: contains very little detail about biophysical dynamics of neurons! A computer simulation of cognitive processes -> can wind up with a network that can give us different answers about the domain of semantic knowledge The general modeling approach can be applied to many different domains of cognition

Large vs. Small Text Metacognition Experiment

Can we manipulate fluency in a way that affects how memory really is? How do people rate their likelihood of predicting an item? -when people see big words they think they are going to remember it more -when people see small words they think they are not going to remember because you have to squint -in some studies, harder to process things might even be encoded better Results: Font size had no effect on memory despite people thinking that they would remember the larger font better. However, the studying twice did have an effect on memory because you saw the word twice. Explanation: Ease of processing might be the factor that is misdirecting you -font size leads to higher ratings but does not make an impact on actual memory

Semantic Network Approach

Categories can be arranged in a hierarchy of levels, from global (at the top) to specific (at the bottom) ex: canary and salmon are at the bottom and living thing is at the top The semantic network consists of nodes (category or concepts) that are connected by links. Related concepts are connected by links and a number of properties are listed for each concept. ex: canary and bird, bird and animal are connected Cognitive economy: storing shared properties just once at a higher-level node ex: instead of saying "can fly" and "has feathers" for every kind of bird, you can just put them for the node bird This model shows how concepts and their properties are associated in the mind.

Concepts vs. Categories

Category: a collection of items ex: the category "bird" is the set of all exemplars (examples of types of birds) Concept: a representation in your head -a concept defines how items should be grouped into categories ex: the concept "bird" is your mental representation of a bird, including its properties (has a beak, can fly)

Connectionist Model Example

Concept items: canary, salmon, rose Relation statements: is a, is, can, has When we activate the concept "canary" and the relation unit "can", some of the representation and hidden units are activated. This eventually results in the property units "grow," "move," "fly," and "sing" to be activated. Shading indicates activity of the units, with darker shading indicating more activity.

2) Systems Consolidation

Connection weights might change at different speeds in different brain structures. The hippocampus learns, quickly and the cortex learns slowly. -complementary learning systems The hippocampus replays memories (even during sleep) and gradually trains the cortical representation.

Memory is constructive

Declarative memory retrieval is an inference about what occurred in the past -influenced by multiple sources of information, including background knowledge to infer what happened in the past Like perception! -memory involves reconstructing, not merely reproducing past events -similar constructive processes underlie both accurate and inaccurate memories Leads to memory illusions similar to perceptual illusions ex: DRM lists and misleading post-event information As with perception, knowing how the illusion works doesn't necessarily make it go away! -can cause problems when literal, verbatim memory is needed -false memories arise from the same constructive process that makes true memories, which is why construction can cause memory errors

Simulating concept acquisition

During a supervised training, connection weights are adjusted to reduce error Thus, a child's learning about concepts begins with little information and some incorrect ideas, which are slowly modified in response both to observation of the environment and to feedback from others. Similarly, the connectionist network's learning about concepts begins with incorrect connection weights that result in erroneous solutions which are slowly modified in response to error signals to create the correctly functioning network Learning is very gradual!

Generalization

Even if you don't know what the bird is, you can infer a few things such as: -it has wings -it can grow Generalization can be made... 1) to newly learned exemplars ex: There is a bird called the tawny frogmouth The model is efficient because you can take the properties you already know and see that you actually do know a lot about a tawny frogmouth by making generalizations 2) of newly learned properties ex: birds have a beak coated with keratin Only have to modify one local place in the network and it can have a general influence on your knowledge about lots of things

Does the prototype or exemplar approach work better?

Exemplar advantage: 1) By using real examples, it can more easily take into account atypical cases like flightless birds ex: rather than comparing a penguin to the "average" bird, we know with this approach that some birds don't fly -doesn't discard information that might be useful later 2) Can deal more easily with variable categories like games ex: difficult to imagine what prototype might be for the category that connects football, computer games, and solitaire -exemplar approach only requires that we remember some of these examples People may use both approaches: As we initially learn about a category, we may average exemplars into a prototype, then later in learning, some of the exemplar information becomes stronger.

How do we know that people with semantic dementia are not just loosing the words for the particular concept?

Experiment: Task was to view a line drawing, wait 10 seconds, then draw it from memory ex: the person was asked to draw a swan and they drew one with 4 legs Explanation: they are taking general properties (a lot of animals have 4 legs) and generalizing it for all animals

Newly formed memories are fragile and susceptible to interference.

Experiment: 2 groups of participants learned lists of nonsense syllables Immediate group: learned both lists back to back Delay group: learned first list and then waited 6 minutes before learning the second list Results: Immediate group: remembered 28% of the syllables -experienced greater retroactive interference Delay group: remembered 48% of the syllables Explanation: Immediately presenting the second list to the immediate group interrupted the forming of stable memory for the first list (consolidation did not happen)

False memory experiment

Experiment: Enlisted the participant's parent or older sibling and got 2 real childhood events (from around age 5) for them to try to remember -they also added in one made-up event Half the participants also saw their real class picture to supplement the fake event -didn't show them a photoshopped picture of the event like other studies do Both groups were told to spend time each day over the next week remembering more about the event Results after 1 week: No-photo condition: 23% of people said they remember the false memory Photo condition: 65% of people said they remember the false memory -the illusion is powerful!

Consolidation and Sleep: Enhancing Memory

Experiment: Had high school students learn a list of 24 pairs of English-German vocabulary words. Sleep group: studied the words and then went to sleep within 3 hours Awake group: studied the words and then stayed awake for 10 hours before resting Tested both groups 24 hours later. Results: Students in the sleep group forgot much less than students in the awake group. Explanation: 1) Sleep enhances consolidation! 2) Sleep eliminates environmental stimuli that might interfere with consolidation

Source monitoring example: Famous people experiment

Experiment: Participants read a list of non famous names. They then read a list of old and new non famous names and new famous names and were asked to point out who are celebrities. Results: Immediate testing: Participants were able to correctly identify non famous names as non famous people. Waiting 24 hours: Participants were not as accurate and some non famous names were misidentified as famous people. Explanation: Sometimes you think that the names you have seen on the study list are celebrities because they seem familiar to you. -source monitoring problem because you are unsure as to where you know that name from (the study list vs. from them being a celebrity)

Evidence for the Multiple Traces Model

Experiment: Participants viewed pairs of stimuli, such as an alligator and a candle, while undergoing fMRI. They were told to imagine the items interacting with each other. 10 minutes and 1 week later, they saw a list of the original pairs and some they have not seen. They were told to say if they "remember" the pair being originally presented, "know" of the pair, but don't remember when they were originally seeing it, and "don't remember." -"remember" indicates episodic memory and "know" indicates semantic memory Results Part 1: After 10 minutes: more remember (episodic) responses than know (semantic) After 1 week: more know (semantic) responses than remember (episodic) Results Part 2: The hippocampus response remained high for RR pairs (remember both at 10 minutes and 1 week). The hippocampus responses dropped to near 0 for RK pairs (ones that lost their episodic character after 1 week) Explanation: the response of the hippocampus can change over time, but only for stimuli that have lost their episodic character.

Source misattribution example: Misleading post-event information

Experiment: Participants viewed photos depicting a car accident and then had to respond to questions group 1: did another car pass the red car while it stopped at the stop sign? group 2: did another car pass the red car while it was stopped at the yield sign? Later on, they are given 2 pictures one with a stop sign and one with a yield sign and they were asked to identify which picture they originally saw Results: With accurate information (group 1): 75% correct With misleading information (group 2): 41% correct This is an example of source misattribution because they are confused about where that information about the yield sign is coming from (didn't realize it wasn't from the picture but from the trick question)

Compared with everyday memories from the same time, are flashbulb memories more consistent?

Experiment: Participants were asked to remember memories from everyday life (blue line) and memories about the 9/11 attack (red line) Results: 1) Details you can provide: -Remembered both events pretty well and everyday events were a little higher -Memory decreased at the same rate over time for both events 2) How confident are you in the accuracy of your memory? -participants' confidence and vividness in memory decreases at the same rate as the details you can provide -However, for the flashbulb memory, people say they have more vivid memory and they believe their memory is more accurate compared to their everyday memory Conclusion: Flashbulb memories may lose their details without losing their vividness -not a difference in losing their accuracy overtime, it is a difference in how much you believe the event is vivid and impactful to you -thus flashbulb memories are vivid (likely to be remembered) and ordinary (not accurately remembered) at the same time

Decisions to share fake news on the internet

Experiment: Used a false headline in a study and people had to say if it is accurate or fake. They then had to say if they were to share the headline for both headlines they knew were accurate and fake. Results: People were capable of judging whether a headline was accurate or not. But, this accuracy did not affect their inclination to share it. Even though, when asked they agreed it was important to share only stories that were accurate. -people's stated goals are inconsistent with the sharing behavior that they endorsed Explanation: Why do people share fake headlines? They are momentarily inattentive and forget about their goal of wanting to share accurate information. A possible intervention: Prime the accuracy goal Just by jogging someone's memory into the mode of assessing the truth value of something (their actual goal), they are more likely to share only accurate information

Flashbulb memories: Challenger Experiment

Experiment: asked people how they first learned about the loss of the space shuttle Challenger But, how can the researchers know the truth of when they actually learned about the event? -repeated recall method is not that accurate (memory instantly after the event is used as a baseline of how accurate your memory is later on) Solution: Can see whether details are changed or lost over time -they asked people where they first learned about the Challenger explosion right away and 2.5 years later Results: Right away: 21% say from the TV 2.5 years later: 45% say from the TV Why does this happen? -lots of interference: saw followup coverage on TV after they learned about, talked to people about it, confused with how other people found out about it

False memory of DRM Lists: Free Recall

Experiment: showed a list of words that all connected to each other based on meaning and you had to write down as many words that you can remember Results: most people falsely remembered the word "window" because it was similar to the list of words that were in the list like "glass" and "screen" DRM (Deese, Roediger, and McDermott) Lists: 1) Deese showed recall was better if the items in the list were semantically related compared to if they were random and unconnected -participants also often falsely recalled non-presented items that are related to the list 2) Roediger and McDermott confirmed this task systematically elicits false memory reports

1) Misattribution of familiarity experiment

Experimental group: some saw a video of a male teacher reading to students Control group: some saw a video of a female teacher reading to students Then both of them viewed a film of a female teacher being robbed and were asked to pick a robber from the photo spread -the robber in the video has physical resemblance to the male teacher and the other decoy suspects Results: When the robber was in the photo spread, 20% of the experimental group incorrectly picked the male teacher as the robber, and 10% of the control incorrectly picked the male teacher as the robber When the actual robber was not in the spread, 60% of the experimental group incorrectly picked the male teacher, and 20% of the control group incorrectly picked the male teacher When the experimental group did not see the correct robber, they choose the person who looked most like him -familiarity is being misattributed!

Remember/know judgements for real events

Finding semanticization of remote memories Semanticization could be useful in that it transforms event-specific memories into generalizable knowledge

Are some kinds of memories especially resistant to change?

For example, for memories for personally significant, affectively laden events? Two approaches to studying this: 1) laboratory experiments with emotional stimuli 2) real life events

Connectionist model vs. Hierarchal network

Hierarchal: concepts and their properties are represented by activation of different nodes Connectionist: representations are far more complex, involving many more units for each concept -knowledge is represented by distributed activity -also what is much more likely to what happens in the brain

How do we organize concepts?: Idea #1

Idea #1: we organize concepts as definitions, sets of necessary and sufficient conditions This can work as part of a specialized theory: ex: definition of what a triangle, prime number, and a cousin is are very clear because there are exact characteristics that define them But this doesn't always work! ex: what is a game? Dictionary definition: "a form of play or sport, especially a competitive one played according to rules and decided by skill, strength, or luck" This definition is a little fuzzy -trying to define what people usually define what a game is based on their concept of a game is difficult A lot of the conceptual knowledge that we has isn't very much like definitions ex: people's concept of lions are that they have manes, but this isn't the definition of a line

How do we organize concepts?: Idea #2

Idea #2: Concepts can be defined based on similarity ex: What is a good example of a bird? -a sparrow is a classic example of a bird -> very central to the category of what a bird is -a penguin is a bird, but maybe not the first one you think of Exemplars vary in their typicality! ex: examples of a fruit -> strawberry is a classic one, tomato is not a classic one Exemplars vary in their typicality

The network needs to be trained to obtain this result

In an untrained network, activating the "canary" and "can" units sends activity out to the rest of the network with the effect of this activation depending on the connection weights between the units. Since the weights are the same in an untrained network, activity spreads throughout the network and property nodes such as "flower," "pine," and "bark" are also activated mistakenly. How can we prevent this? Back propagation: erroneous responses in the property unit (like flower) cause an error signal to send the signal backward -error signals are sent back to the hidden units and representation units and they adjust the weights so that the correct property units will be activated -causes a learning process to occur

Metacognition

Insight into your own cognitive processes 1) Retrospective confidence judgements: confidence judgements express your level of certainty about what you've seen, remembered, or decided (How sure am I?) 2) Prospective memory judgements: judgements of learning express how sure you are you'll remember something later (Will I be able to remember this later?)

What does the Connectionist Model propose?

It proposes that concepts are represented by a pattern of activity that is distributed across a network to the other units Activation of units in a network depends on... 1) the signal that originates the input units 2) the connection weights throughout the network

Why does looking at a photo of yourself give more false memories?

It provides extra imagery and context to your childhood ex: what your classmates and teacher looked like It is easier to construct a scene in your head if you can picture the faces better

Prospective memory judgments

Judgments of learning: How likely is it you'll be able to remember this later? -important for allocating study time Experiment: Study arbitrary words pairs (OCEAN-TREE) How confident are you that in ten minutes you'll be able to recall what word was paired with OCEAN? -people rate their confidence on a percentage scale This could be asked either right after study, or following a short delay -can plot the results the same way we plotted the confidence calibrations Results: For the delayed judgements of learning, they are pretty close to the accurate metacognition line This is much better than the calibration for the immediate judgement of learning (less accurate and has more of a horizontal line)

1) Synaptic consolidation

Learning and memory are represented in the brain by physiological changes that take place in the synapse involves changes in synaptic efficacy over minutes to hours -Individual inputs vary in how strongly they drive the output (synaptic efficacy) Long term potentiation: strengthening of synaptic transmission causes enhanced firing of neurons after repeated stimulation

Collins and Quillian's experiment

Measured the reaction time to a number of different statements that involved traversing different distances in the network. Results: statements that required further travel up the levels from "canary" resulted in longer reaction times. -this is the same for verifying statements about properties of canaries (a canary can sing, can fly, has skin) and categories of which the canary is a member of (canary is a canary, is a bird, is an animal)

People with highly superior autobiographical memory

Memory is not like a video recording There isn't good evidence that anyone has perfectly accurate photographic memory -these people just see images very vividly There are people with highly superior autobiographical memory who can describe in detail what they did on specific dates in the past -happens automatically and does not need conscious control -performed at levels similar to normal participants on most standard laboratory memory tests Such individuals are still susceptible to memory errors in the DRM and "misleading post-event information" tests

Summary of metacognition

Metacognitive judgements contain valid information, but are imperfectly calibrated. Some factors systematically influence metacognitive judgements without influencing true accuracy.

Reconsolidation does not need to be pharmacological manipulation.

New learning during the reconsolidation interval could change the pre-existing memory. Experiment: Participants were told to remember 24 items in a tape. Test group: Participants took a cued recalled test to remember the items. No-test group: Participants played Tetris instead of the cued recalled test. Both groups were given misinformation after this and then were given the cued recall test again. Results: The test group said yes 50% of the time to incorrect items. The no-test group said yes 30% of the time to incorrect items. Explanation: The misinformation got in the way of the reconsolidation process. When might this be beneficial? -in certain types of therapy, a goal might be to reduce the intensity of an intrusive memory. Ex: PTSD response

Hebbian learning in a neural network

Nodes (defined by activation levels) -circles Connections (defined by weights) -lines 1) Suppose a certain experience repeatedly activates some of the nodes. 2) Their connection weights strengthen. (Activity-dependent plasticity) 3) Now suppose another experience activates one of the same nodes. 4) Pattern completion occurs: the whole assembly can reactivate even if it is just one of the nodes that activates.

How similar are the network's representations of various concepts?

Once the network has been trained, does it represent it similarly or dissimilarly? Further apart: fewer properties in common Closer together: share more and produce patterns of activity that have more resemblance

Typicality effect on sentence verification

Participants are presented with statements and are asked to answer true or false. ex: An apple is a fruit. A pomegranate is a fruit. Results: participants responded faster for objects that are high in prototypicality

Prototypes experiment

Participants have to classify the randomly placed dot pattern as A or as B as fast as possible. On the first few trials they will not be able to properly classify the patterns and so they will have to guess. -variations of the prototype are made by taking 10 of the 25 dots and rotating them The participant does not know what the prototype is for A and B and there are also 2 unseen variants for A and B. Do you form a mental representation of the unseen prototype for each category? Yes! There are faster response times for the unseen prototype than for unseen variants Explanation: Information about the what the prototype looks like is abstracted from the stored instances (variants)

Wilhelm's Sleep Consolidation Experiment

Participants learned a task and group 1 was told that they would be tested on the task later and group 2 was told they would be tested on a different task later. After a night's sleep, participants were tested on the task to determine if what they expected to be tested on had any effect on consolidation. Results: The expected group performed significantly better than the unexpected group. Explanation: When we sleep after learning, memories that are more important are more likely to be strengthened by consolidation. Some memories are more likely to be consolidated than others.

2) Effects of feedback experiment

Participants viewed a video of a crime and were asked to pick a perpetrator from the photo spread. After choosing, they gave participants confirming feedback, no feedback, or disconfirming feedback. Received confirming feedback: participants rated their confidence as 5.4 Receive no feedback: participants rated their confidence at 4 Receive disconfirming feedback: participants rated their confidence at 3.5 Ideally, the person showing you the photos should not know who their suspect is but in reality, they usually know and will show subtle cues hinting to who it is

Systematic biases in predicting your own later memory success

People have worse metacognition immediately after learning because it is so easy to follow in the moment and think that you understand, but that is not a good guide to your true level of understanding With time after studying and a delay, people are showing better metacognition

Eyewitness testimony

Persuasive but flawed memory recall subjected to interpretation by person reporting it Eyewitness identification ex: photo or live lineup of suspects (can you match your memory to the people?) Potential pitfalls: 1) Misattribution of familiarity ex: this is the face I recognize from the crime scene, but it was not the perpetrator, it was a bystander 2) Effects of feedback 3) Judgements of relative similarity

Typicality effect of priming

Priming: presentation of one stimulus facilitates the response to another stimulus Experiment: Participants heard the prime, which was the name of the color green. 2 seconds later, they saw a pair of colors side by side and pressed a key if they were the same or different. colors were paired in 3 different ways: way 1: colors were both good examples of green way 2: colors were poor examples of green (light green) way 3: colors were different Results: Priming results in faster same judgments for the prototypical green color than for the nonprototypical color.

Metacognition graphs

Proportion correct: -if someone had perfect calibration and really good metacognition, you will have a straight constant slope -to have perfect metacognition, if you are 100% confident, you should get all of them correct -if someone had no metacognition at all, it would be a horizontal line (nonexistent calibration) -no shared information between the confidence you expressed and the proportion you have correct How is actual behavior? -a little bit shallower than perfect cognition If you said you were 80% confident and actually got 60% correct you are overconfident (accuracy can't back it up) -most points from .6 to upwards, are expressing overconfidence because their accuracy is lower than the confidence they expressed

Hupbach's reconsolidation in humans experiment

Reminder group: On Monday, the reminder group was presented with 20 objects one by one and places them in a blue basket. Participants then recall these objects, creating list A. On Wednesday, participants were told to remember the testing procedure from Monday. They met with the same experimenter, in the same room, and the blue basket was present. Participants repeated this procedure with 20 new items which creates list b. On Friday, participants are asked to recall all the objects in list A. No-reminder group: Same procedure expect on Wednesday, they were not told to remember list A. Results: Reminder group: Could recall 35% of list A Mistakenly recalled 24% of list B No-reminder group: Could recall 45% of list A Mistakenly recalled only 5% of list B Explanation: When the reminder group thought back to list A during Wednesday's training session, this made list A vulnerable to being changed. This is why more of the objects from list B were integrated into their memory from list A. The reminder reactivated memory for List A and "opened the door" for List B objects to be added to the participants' memory for that list. The original memory was changed (reconsolidation).

What about causal knowledge?

Sometimes multiple properties are meaningfully related ex: birds have wings, has feathers, has hollow bones, and can fly The PDP model doesn't represent any casual theory of how these fit together -but it does know that they consistently co-occur This is enough for the model to perform rapid generalization ex: knowing that the Tawny Frogmouth is a bird

Source misattributions

Source monitoring: determining where remembered information came from -asking if this is external or self-generated information Ex: Did I come up with that idea or you? Where do I know that guy from? Did I turn off the stove or just think about it? Source misattributions: errors of source monitoring Ex: Subconscious plagiarism George Harrison wrote a song that sounded a lot like the Chiffons song -George thought that the source of information was attributed to self-generated information despite it being from something he encountered in the past

Strengths and Weaknesses of the Connectionist model

Strengths: 1) can simulate learning and generalization -because similar concepts have similar patterns, training a system to recognize the properties of one concept (like a canary) can provide information about another related concept (like a robin) ex: learning about canaries helps us predict properties about other types of birds that we have never seen 2) graceful degradation: the disruption of performance occurs only gradually as parts of the system are damaged -connectionist models are not totally disrupted by damage because information in the network is distributed across many units -similar to how brain damage only causes partial loss of functioning 3) a powerful method for representing large amounts of knowledge Weaknesses: 1) requires supervised learning 2) high complexity reduces explanatory power

False memory of DRM Lists: Recognition

Study list: bumpy, tough, sandpaper, jagged, coarse, uneven Experiment: First the participant studies a list of words, and then there will be a mix of unrelated lures and a critical lure (ex: rough) and the participant has to say if it is in the study list or not -the critical lure is a lure that is strongly related to the words in study list Results: 86% of targets are called old 2% of lures are called old (pretty good at rejecting words that have nothing to do with the list) However, 84% of the critical lures were called old -"rough" feels super familiar to participants Strength theory: targets have strong familiarity in memory and lures have weak familiarity in memory -the critical lure has high familiarity, but not because it was actually seen, it is because the word was indirectly activated by words that are related to it (semantic associations)

Can you do things to change someone's confidence without changing their accuracy?

The Illusory truth effect: merely repeating information can make it seem more true -people have an enhanced probability of evaluating something as true if they hear it a lot -a type of source misattribution Experiment: Presented true and false statements to participants and then asked them to rate how interesting they were. In the second part, they asked participants to indicate whether the statements they had read previously, plus a new number of statements, were true or false. Results: People rated new statements that were correct "true" 56% of the time. People rated repeated statements that were correct "true" 62% of the time. -repetition increased perceived truth, even if the person knew the correct answer

How does activity in the representation level change during learning?

The bars represent activity in the 8 representation units in response to 3 different concepts: "rose," "daisy," and "canary" 1) The experimenters set the connection weights to 0 at the beginning of the learning process. 2) As learning progressed, the computer changes the weights slightly after each trial in response to error signals. By trial 250, the the pattern for "canary" and "daisy"/"rose" look different 3) By trial 2,500, it is now very easy to tell the difference between "canary" and "daisy"/"rose." "Daisy" and "rose" have similar patterns but it is easier to tell apart compared to trial 250. The model acquires broad categories first, and then makes finer distinctions later, just like kids do during development!

The Illusory truth effect graph

The first time you see a statement you have to rate how interesting it is The second time you rate it, you tend to rate its truth rating as higher than if you did not see it in the first section at all In the intermediate level of confidence, you are most susceptible to repetition boosting your level of confidence -because you have heard it before, it might be easier to process

The connectionist model learns really slowly

The same set of weights holds information about many concepts Any one connection is going to contribute to your knowledge about lots of things Learning too fast about one thing could interfere with other knowledge -could mess with previous things that you learn One possible solution: complementary learning systems This slow-learning cortical system might interact with fast-learning MTL system -training this slower learning semantic system System consolidation -a memory can quickly be represented in the MTL, but only after a lot of time is passed will it begin to be represented in the cortex -by reactivating the cortical parts of that memory over and over, it eventually gives this system a bunch of experiences to help it learn -only after a lot of experience builds up you will start to have these representations

Amnesia

The standard model of consolidation was based on observations of memory loss caused by trauma or injury. Retrograde amnesia: loss of memory for events that occurred before the injury Graded amnesia: the amnesia tends to be most severe for events that happened just before the injury and less severe for earlier events. -a characteristic of retrograde amnesia The gradual decrease in amnesia corresponds to changes in connections between the hippocampus and cortical areas according to the standard model. Systems consolidation can make older memories less hippocampus-dependent.

Systems Consolidation: Standard Model

The standard model of consolidation: proposes that memory unfolds according to a sequence of steps: 1) Connections between the cortex and the hippocampus are initially strong and connections between the cortical areas are weak (called reactivation). 2) As time passes, connections between the hippocampus and cortex weaken and connections between the cortical areas strengthen. 3) Eventually, only connections between cortical areas remain. (hippocampus is no longer in these memories)

Current state of knowledge on putatively recovered memories

There is not good evidence that true memories get repressed and then recovered through therapy -memory recovery therapy is unreliable What is wrong with this approach? -creates a very uncomfortable situation because they think they hold this memory -caused more harm as patients experienced more stress

3) Judgements of relative similarity experiment

There is this tendency for people to think: which one of these people's match the person in my memory the closest? The participants watched a video of the crime and had to identify the suspect from a bunch of photos Results: When they did include the suspect in the photo array: 54% got it right 21% thought that the suspect was not in there the rest was spread around other options When they don't include the suspect in the photo array: 32% said none of the above 38% of people tended to choose the person who is the next closest match

Bonnici's Multivoxel Pattern Analysis Experiment

They trained the classifier to determine the voxel patterns of participant's memories (3 were recent and 3 were remote). Results: The classifier was able to predict which recent memories and which remote memories were being recalled based on activity in the hippocampus. -evidence for the multiple trace model since the hippocampus is activated not just for new, but for remote memories too Voxels associated with the recent autobiographical memories are in the anterior hippocampus. Voxels associated with the remote autobiographical memories are in the posterior hippocampus. -more information about remote memories were in the prefrontal cortex Taken together, these results show that remote memories are richly represented in the cortex, as proposed by both the standard and multiple trace models, and that both recent and remote memories are represented in the hippocampus as well, as proposed by the multiple trace model.

Reconsolidation Rat Experiment

They used classical conditioning to create a fear response in the rat of freezing (not moving) to the presentation of a tone. This was achieved by pairing the tone with a shock. Ansiomyscin: an antibiotic that inhibits protein synthesis and so prevents changes at the synapse that are responsible for the formation of new memories. If it is injected before consolidation occurs, it eliminates memory, but if it is injected after consolidation occurs, it has no effect. Condition 1: The rat receives the pairing of the tone and the shock on day 1 and the anisomycin is injected right away, BEFORE consolidation occurs. Condition 2: The rat receives the pairing of the tone and shock on day 1. The amisomycin is injected on day 2, AFTER consolidation. Condition 3: The rat receives the pairing of the tone and shock on day 1. The amisomycin is injected on day 2 as the rat is freezing to the tone, DURING REACTIVATION. Results: Condition 1: On day 3, the rat does not freeze from the sound of the tone. This is because it has no memory of the pairing. Condition 2: On day 3, the rat freezes to the sound of the tone because it remembers the pairing. (anisomycin has no affect) Condition 3: On day 3, the rat does not freeze to the tone since it has no memory of the pairing. Explanation: This result shows that when a memory is reactivated, it becomes fragile, just as it was immediately after it was first formed, and the drug can prevent reconsolidation. Thus, just as the original memory is fragile until it is consolidated for the first time, a reactivated becomes fragile until it is reconsolidated. -memory becomes susceptible to being changed or disrupted every time it is retrieved Being able to update memory is very useful!

Repressed memories

This experiment was based on a misguided approach to psychotherapy from the 1990's Experiment: told patients their symptoms were indicative of repressed memories Told to use: 1) Guided imagery: thinking about the adults in your life and picture if they would do something bad to you 2) Stream-of-consciouness journaling 3) Use of old family photos as cues 4) Used confirmatory feedback to have them continue thinking of these repressed memories

General knowledge: Calibration of confidence

To be good at metacognition, it does not have to equate to having high confidence about something Calibrated confidence: being confident when you should be Experiment: gave people a set of trivia questions ex: Is absinthe a liquor or a precious stone? How certain are you on a scale from 50-100%? guessing: 50% super sure: 100% For items receiving each level of confidence, what was the actual rate of accuracy?

Metacognition has limits

We don't have complete introspective access to our own cognitive processes Sometimes we don't know what we don't know ex: change detection failures Other times we use information without realizing it ex: the light-from above assumption This means it's worth systematically studying people's metacognitive abilities and limitations

Schemas

a concept of how a scene or situation should work ex: you have a schema of what belongs in a kitchen and what doesn't Experiment: people were waiting in an office and then they were asked to say what they saw in the office Results: people obtained false reports of schema-congruent items ex: people said there were books or a phone because those are items that are usually in an office but they were not in this office

Semantic dementia

a general loss of knowledge for all concepts -equally deficient in identifying living things and artifacts -the anterior temporal lobe is damaged Experiment: Showed a picture of an item to a patient with semantic dementia and they had to say what it was -sometimes got it wrong sometimes got it right A lot of them are shown a picture of a type of bird and the patient calls all of them a bird -not wrong necessarily, but shows that this person lost higher level concepts such as the type of bird As time passed the condition gets worse, and the names of the items became more generalized or just wrong ex: type of bird -> bird -> animal Semantic dementia is not as much about episodic memory, but more about loosing semantic memory functions

Consolidation

a process that transforms new memories from a fragile state, in which they can be disrupted, into a more permanent, durable state 1) Synaptic consolidation: involves the structural changes at synapses -a faster process (mins-hours) 2) Systems consolidation: involves the gradual reorganization of neural circuits within the brain -a slower process (months-years) Both occur simultaneously, but at different speeds and at different levels of the nervous system

Temporal Context Model

another explanation for Hupbach's experiment that does not involve reconsolidation -they say that it is because of the association, participants incorrectly recall some list B items Temporal context model focuses on the context within which learning and retrieval occur and assumes old contexts can become associated with new memories, without changing the content of existing memories.

Reactivation

helps form direct connections between the various cortical areas The hippocampus is crucial during the early stages of memory, as it is replaying the neural activity associated with a memory and sending this information to the cortex. -hippocampus acts as glue that binds together the representations of memory from different cortical areas -becomes unnecessary once these representations are formed

Prototype approach

membership in a category is determined by comparing the object to a prototype or "typical" member that represents the category ex: sparrow is a prototype for the category birds People abstract out a prototype organizing concepts based on similarity between exemplars and prototype

Multivoxel Pattern Analysis

standard fMRI procedure is to present a task to the participant and determine the activity of voxels in the brain MVPA determines the pattern of voxel activation within various structures Experiment: A classifier, a computer program designed to recognize patterns of voxel activity, is trained. Participants look at pictures of an apple and pear, feeding the voxel patterns for each object into the classifier. This is repeated many trials so the classifier can learn which patterns go to what object. The classifier attempts to determine which object is presented based on the voxel activation pattern.

Reconsolidation

the idea that when a memory is retrieved (remembered), it becomes fragile, like it was when it was originally formed, and thus, needs to be consolidated again. -reactivating memories can open it to being changed Newly formed memories are fragile. Newly retrieved memories are also fragile. Even if a memory has already undergone synaptic consolidation, it might need to be reconsolidated after being activated.

Exemplar approach

the standard for determining whether an object is similar to other objects involves many exemplars -exemplars are actual members of the category that a person has encountered in the past You have item to item similarities -instead of having an idealized version in your head ex: if you have encountered sparrows, robins, and blu jays in the past, these would be exemplars for the category birds organizing concepts based on similarity between exemplars and all the other exemplars