The malleability of memory

2.1Eyewitness testimony and the case of Steve Titus

The lecture opens with a real criminal case that illustrates why the reliability of human memory matters so much in practice: the case of Steve Titus. Titus was identified by an eyewitness and wrongfully implicated in a crime, a case that is used to introduce the theme of the lecture: eyewitness memory is not as reliable as it feels.

2.1.1The memory of eyewitnesses

• A common (and intuitively convincing) claim in court is: “He did it! I saw it with my own eyes.”

• More than 75,000 people per year in the US are accused of a crime based on eyewitness testimony.

• Memory researchers have been warning against over-reliance on eyewitness testimony since the 1970s.

• These warnings were only widely accepted from the 1990s onwards, when DNA research allowed people who had been wrongfully convicted based on mistaken identity to be acquitted (see the Innocence Project, https://www.innocenceproject.org/).

Elizabeth Loftus & Katherine Ketcham, “Witness for the Defense”, is a key work on the fallibility of eyewitness memory.

2.2The information processing model

Memory is conceptualized within a serial information processing model. This model distinguishes between different stages of processing, such as sensory memory, short-term memory, and long-term memory.

We often use a computer metaphor to describe memory, viewing the brain like a hard drive or cloud storage where information remains unchanged once saved. This is an inaccurate model for human memory: human memory is malleable and subject to change.

2.3Long-term memory (LTM)

2.3.1Characteristics of long-term memory

Long-term memory is a distinct component of the information processing model, differing from attention and short-term memory in two primary ways.

Duration. While other processes are fleeting and short-lived, LTM can preserve information for a very long time. Memories can span back to early childhood (for example, ages two or three).

Capacity. Unlike attention, which is selective and limited, LTM is believed to be extremely extensive. However, the exact capacity is difficult to determine because it is nearly impossible to test the full extent of a person's lifetime experiences.

2.3.2The structure of long-term memory

Long-term memory is conceived as a hierarchically organized system consisting of various sub-memory systems. The primary distinction is made between declarative (explicit) and procedural (implicit) memory.

Procedural memory (implicit) refers to our memory for skills and actions, the “how-to” of various tasks. It contains information on how to perform actions such as riding a bike, playing a guitar, or driving a car. Through practice, these skills become “automatized.” Once learned, they no longer require explicit intentional recall or conscious thought regarding the specific movements required. Procedural memory is resistant to forgetting because it is so deeply ingrained; this can also be observed in people with dementia. Even in individuals with advanced dementia who may lose their identity or recognition of loved ones, procedural skills (like playing an instrument or writing) are often retained.

Declarative memory (explicit) involves the conscious, intentional recollection of factual information and previous experiences. It is subdivided into two main types: semantic memory and episodic memory.

Semantic memory is your accumulated knowledge about the world, acting much like a mental dictionary. It includes the meanings of words, facts (for example, “the King of the Netherlands”), and general knowledge. It is generally devoid of context: you know the information but typically do not remember the specific time or place where you learned it.

Episodic memory consists of autobiographical memories of specific events that have happened to you as an individual, such as a specific holiday, a birthday party, or a particular summer. Unlike semantic memory, episodic memory is closely attached to a specific time and place, and that is what allows you to relive it.

2.4Semantic versus episodic memory

While both are forms of declarative memory, they differ significantly in how they are structured, experienced, and stored.

2.4.1The apple comparison

Semantic knowledge. Knowing that an apple is a piece of fruit, that it can be red, green, or brown, and knowing its taste. This is an abstract entity accumulated over many encounters.

Episodic memory. Remembering a specific moment, such as seeing a daughter wearing a skirt with an apple pattern during a specific family vacation. This allows you to “travel back in time” to that specific time and location.

The distinction between the two systems can also be understood through the fundamental questions they answer: semantic memory answers “what,” while episodic memory answers “when and where.”

2.4.2Vulnerability and robustness

Semantic memory (robust). Generally resistant to forgetting. Because semantic knowledge is the accumulation of numerous encounters over time, it is well-integrated with many other pieces of information, a stable and “well-stretched” web of knowledge, which makes it highly stable.

Episodic memory (vulnerable). Highly susceptible to forgetting. These memories are more isolated in the mind and lack the dense web of connections found in semantic memory.

The role of context. Time and location act as “entry tickets” to an episodic memory. If the contextual information (the “when” and “where”) fades or is lost, it becomes significantly more difficult to retrieve the specific episode. Once these “entry tickets” are lost, retrieval becomes extremely difficult.

2.5The three stages of memory formation

Remembering an episode is not a single event but a process involving three distinct, interacting stages: encoding, storage, and retrieval.

2.5.1Encoding

Encoding is the initial stage where information is perceived and processed to be converted into a mental representation: the initial learning of information through the process of perceiving information in the world and relating it to past knowledge.

• Continuous. Encoding is happening all the time, though with varying levels of success.

• Selective (~ attention). We do not encode everything. Distinctiveness: highly unusual or unique events (for example, an ambulance on campus or seeing a famous person) are encoded more strongly because they stand out from the mundane. Emotional content: information with high emotional value is more easily and strongly committed to memory.

• Recoding. Converting information in a way that makes sense, that is, connecting new information to existing knowledge to facilitate better learning.

An example of recoding: using a funny sentence or an emotionally resonant word as a password is easier to remember than a random string of digits, because it can be connected to existing mental structures.

2.5.2Memory enhancement techniques

Chunking. Breaking down long, meaningless sequences (like a 20-digit number) into smaller, more manageable “chunks” to aid short-term memory.

Storytelling and mnemonic association. To transfer information from short-term to long-term memory, one can create a narrative or story linking the pieces of information together (for example, imagining a specific number represents a certain age or a specific number of steps in a tower).

2.5.3Storage

Storage is the phase where a person attempts to maintain the encoded information over time. Once information is encoded, a mental representation known as an engram (or memory trace) is laid down. Engrams are subject to two competing processes.

Consolidation. The process of strengthening the memory trace, the neural changes that occur after learning, over time, to imprint the engram. This often happens during sleep, which is highly beneficial for memory stabilization; active rehearsal can also aid consolidation.

Decay. The process where memory traces weaken and eventually “crumble” or fade over time if not maintained.

2.5.4Retrieval

Retrieval is the process of accessing and bringing stored information back into conscious awareness for use. It involves pulling information out of memory with little to no external guidance, and this is distinct from recognition.

Recall. Actively searching the memory to produce information (for example, answering an open-ended exam question).

Recognition. Identifying something as familiar when encountering it again (for example, selecting the correct answer in a multiple-choice exam where the correct option stands out due to familiarity).

An example of the whole process is meeting someone at a party. You encode the link between their face and their name, you store that link in your long-term memory, and when you see them again, you retrieve that information to address them correctly.

2.5.5Interdependence of the three stages

These three stages are not independent; they interact constantly. The quality of encoding directly impacts how well information is maintained (storage) and how easily it can be accessed (retrieval). Frequent retrieval of a specific memory can enhance its consolidation.

2.6Reliability of memories and biases

Memory is not a perfect recording; each stage of the memory process, encoding, storage, and retrieval, is susceptible to various biases. Memories are not always factual reflections of experience; they can be distorted during any of the three stages.

2.6.1Biases in encoding

Biases at this stage occur during the initial perception of an event. Distortions during encoding often stem from the specific circumstances or characteristics of the observation.

Characteristics of the observed (Clifford & Scott, 1978). This research highlights how the nature of an event can affect how it is encoded by witnesses.

• Study design: a between-subjects manipulation using a movie in which two police officers find a suspect using information obtained from a third person.

• Non-violent condition: officers persuade the third person, the informant, to provide the information.

• Violent condition: officers extract the information from the informant using violence.

• Procedure: participants watched the movie, performed a distractor task (to ensure recall was based on long-term rather than short-term memory), and then received an unexpected recall task.

• Findings: participants in the violent condition showed poorer recall of the movie's entirety compared to the non-violent condition. This impairment was not limited to the violent scene itself, but affected recall of events both before and after it.

Characteristics of the observer (Hastorf & Cantril, 1954). This study demonstrates how an observer's own expectations and biases (such as loyalty to a specific sports team) can influence their perception, and subsequently their memory, of an event.

• Study: observations of a Dartmouth versus Princeton football game.

• Method: participants, students from both universities, were asked to count fouls and rate the seriousness of the play.

• Findings: expectations influenced how fouls were perceived. Princeton students rated fouls committed by the Dartmouth team much more severely than Dartmouth students did for the same plays.

People bring preconceived expectations into an observation. In this case, students viewed members of the opposing team as “aggressors” and members of their own team as “victims.” This difference in perception leads to different subjective “realities” and, subsequently, different memories of the event. This is a form of confirmation bias: an individual brings their own expectations to a situation, perceives events in a way that aligns with those expectations, and subsequently recalls those events based on those same biases.

2.6.2Biases in storage

Once an event has transpired and been encoded into long-term memory, the storage stage is not static. Information processed after the fact can strengthen, distort, or even create memories.

Strengthening memories (Loftus, 1975). Learning new information after an event can reinforce existing memories if the new information is consistent with what was originally perceived.

• Study: participants watched a movie of a driver ignoring a stop sign, turning right, and causing an accident; afterwards they answered questions about the movie.

• Between-subjects manipulation, Q1: “How fast was the car going when the driver ignored the stop sign?” versus “How fast was the car going when the driver turned right?”

• Dependent variable (Q10): “Do you remember seeing a stop sign that was ignored by the driver?” 53% (mention condition) versus 35% (no-mention condition) answered yes.

Conclusion: merely mentioning something increases the likelihood that it will be remembered. Post-event information that is consistent with the original event can strengthen the memory of that event. Note: this requires the post-event information to be consistent with what was actually observed.

Distorting memories (Loftus, 1977). Information presented after an event can also distort memory, particularly when the new information is inconsistent with the original event, sometimes referred to as “compromising” the memory.

• Study: participants watched a movie of a driver turning right and hitting a pedestrian; a green car is seen passing the scene.

• Between-subjects manipulation, Q10: “Did the blue car that drove past the accident have a ski rack on the roof?” (inconsistent and leading) versus “Did the car that drove past the accident have a ski rack on the roof?” (neutral).

• Dependent variable: colour indication of objects using a colour wheel, more than 20 minutes later.

Findings: participants who were asked the question suggesting the car was blue showed a memory distortion. Their recollection of the car's colour shifted from the actual green towards the suggested blue. Conclusion: providing inconsistent information after the fact may distort memories (“compromise memories”).

Creating new memories: the post-event misinformation effect (Loftus et al., 1978; Loftus, 1979). Post-event information can lead to the implantation of entirely new, false memories through the integration of verbal information into visual memory.

Laboratory evidence (Loftus et al., 1978).

• Scenario: a movie featuring a driver ignoring a traffic sign (either a stop sign or a yield sign), causing an accident.

• Experimental design, a 2 × 3 manipulation. Visual stimulus: the movie contained either a stop sign or a yield sign. Question type (Q17): “Did another car pass the red Datsun while it was stopped?” neutral (no sign mentioned), consistent (mentions the sign actually shown), or inconsistent (mentions the other sign).

• Dependent variable: forced choice after various retention intervals (20 minutes, 1 day, 2 days, 1 week).

Results:

• Neutral condition: accuracy was high immediately after the movie (about 80%) but decayed naturally as time progressed.

• Consistent condition: participants performed better than in the neutral condition. Aligning post-event information with actual observations helps strengthen the memory.

• Inconsistent condition: this led to the most significant distortion. Participants often “remembered” seeing the incorrect sign; for example, if they saw a stop sign but were asked about a yield sign, they were highly likely to later report seeing a yield sign.

Field evidence (Loftus, 1979). This effect is not restricted to the laboratory.

• Field experiment: two students at a train station step away from an empty bag to check the train schedule.

• An accomplice pretends to steal something from the bag; the student then calls out that her tape recorder has been stolen.

• Bystanders are asked for a description and phone number, and are called one week later by a supposed insurance agent.

• More than 50% of bystanders indicated having seen a tape recorder, an object that had never actually been present, and many could provide specific details about it.

Explanations of the post-event misinformation effect.

a. Social pressure. Participants may provide the information they think the researcher wants to hear. However, providing incentives or rewards does not make the effect disappear, so social pressure is not the primary driver of the distortion. This explanation has been ruled out.

b. Memory replacement theory. The original memory is completely overwritten or replaced by the new, suggested information. This theory has also largely been ruled out, because researchers found that the original memory is still available and accessible if participants are specifically probed for it.

c. Memory coexistence theory (most supported). The original memory and the suggested misinformation coexist in the individual's mind. The suggestion given during questioning occurs after the original memory was formed, making the misinformation feel more recent and more familiar. When asked to recall the event, people use recency and familiarity as cues, and mistakenly infer that the option that feels most familiar must be the correct one. The core issue is faulty source monitoring: a failure of source memory, in which people struggle to reliably distinguish the original source of a memory (what they actually saw) from a secondary source (what they were told later).

2.6.3Biases in retrieval

The process of accessing memory can be influenced by how questions are framed, or by the context in which retrieval occurs.

Leading questions (Loftus & Palmer, 1974). Participants watched a movie of two cars in an accident. They were then asked: “How fast were the cars going when they … ?”, with the verb varied between conditions.

The wording of a question at retrieval alone shifted participants' estimated speed of the cars by roughly 10 mph, showing how leading questions can bias what is retrieved from memory.

Circumstances and context of retrieval (Godden & Baddeley, 1975). Context-dependent memory: the environment in which information is retrieved can affect accuracy. In this study, divers learned word lists either underwater or on land, and were later asked to recall them either underwater or on land.

Recall was substantially better when the environment during retrieval matched the environment during learning, illustrating that retrieval, too, can be biased by context.

2.7Implanted and false memories

Beyond simply distorting existing details (like colour or signs), research has investigated whether entirely new, even traumatic, memories can be implanted. So far, the studies above concerned the introduction of new objects into existing memories. But is it also possible to introduce entirely new traumatic memories?

2.7.1The case of Jean Piaget

Piaget recounted a story from his own childhood regarding a kidnapping attempt in the Champs Elysees. Piaget eventually realized the memory was false: he had heard the story as a child (told by his former nurse, who later confessed she had invented it), and had unconsciously projected it into his own past in the form of a vivid, detailed visual memory.

2.7.2Implantation in adults (Loftus & Pickrell, 1995)

Can memories be implanted in adults? In a study about childhood memories:

• Participants were asked to recall details about four childhood events, if they could remember them.

• Three events were real; one event was made up, but included authentic details (for example, a 5-year-old lost in a large mall and crying, then approached by an elderly person who reunited them with their parents).

• 25% of adults (aged 18 to 53) came to “remember” the fabricated event that had never happened, compared with 68% of the true events being remembered.

2.7.3Advertising influence (Braun et al., 2002)

Marketing and advertising can also influence memory. In a study where participants evaluated advertisements for Disneyland, approximately 16% of participants later reported a false memory of meeting the cartoon character Bugs Bunny while leaving Disneyland, despite the fact that Bugs Bunny is a Warner Bros. character and could never actually appear at a Disney park.

2.8Example exam questions

The following example questions were provided by the lecturer as practice material.

Which memory system stores the knowledge about multiplication (e.g., 8 x 5 = 40)? a. episodic memory b. autobiographical memory c. semantic memory d. procedural memory

(Answer: knowledge of facts such as multiplication tables is stored in semantic memory, option c.)

(Answer: varying how long the culprit was visible manipulates how thoroughly the information could be perceived and processed initially, that is encoding, option c.)