Intro & History of Cognitive Science
What were two main issues with introspectionism?
Not all phenomena are accessible to self-report. There is no way to resolve disagreements that are not observable by more than one person.
What are the two-levels of processing?
What is retinal disparity? And name three monocular depth cues.
Retinal disparity is the difference between the images projected onto each eye due to their slightly different positions, which the brain uses to compute depth.
Monocular cues: atmospheric blur, motion parallax, relative height, relative size, superposition, linear perspective, texture gradient, and shading and contours.
What is change blindness?
The tendency to fail to recognize continuous changes in a scene, especially when attention is on something else.
What is chunking?
Chunking is the process of combining several pieces of information into a single unit or code, which effectively increases the amount of information that can be held in short term memory. For example, the string 1 1 2 2 5 5 3 1 4 1 5 is difficult to remember as 11 separate digits but easy to remember as 3 chunks: 112, 255, and 31415.
What is the difference between classical conditioning and operant conditioning?
Classical conditioning involves passive, innate responses, whereas operant conditioning involves intentional actions.
What is Spacing effect / distributed practice?
Study for 30 minutes today, 30 minutes tomorrow, 30 minutes Thursday > studying for 90 minutes straight the night before. Sleep between study sessions helps consolidate memories. Cramming feels like it works (fluency!) but spaced practice actually sticks.
Explain what top-down vs. bottom-up processing means, and give one example for each.
Bottom-up processing means that perception is driven entirely by the properties of the stimulus itself, such as when a low-contrast image is harder to recognize simply because the sensory signal is weak. Top-down processing means that your prior knowledge and expectations actively shape what you perceive, such as when you recognize a blurry object more easily because the surrounding context gives you a clue about what it is likely to be.
How does the concept of perceptual load help resolve between early and late models of attention?
Perceptual mode may *moderate* attention. Under high perceptual load, we may experience early bottlenecking or attenuation; and under low perceptual load, we may experience late bottlenecking or attenuation.
What is the modal model of memory (Atkinson & Shiffrin)
The modal model of memory is the classic idea that information flows from sensory memory to short term memory to long term memory, with rehearsal serving as a key control process that maintains information in short term memory and can transfer it to long term storage.
What is an example of a phenomenon that could not be explain by a behaviorist perspective, and why was behaviorism unable to explain this phenomenon?
Using a mental map of a maze to take a new route (through planning) is an *unobservable* process that is neither an input nor an output. It is a "latent" cognitive representation that explains how rats can learn to solve mazes better without direct experience.
What did Roediger and Karpicke (2006) find evidence for?
The found evidence for the testng effect and that ppl who were tested repeatedly did better than those whi just studied
Can you expain what a stereogram is? And the mechanism behind it?
A stereogram is an image that appears flat but produces a vivid perception of depth when viewed correctly. The mechanism relies on retinal disparity. Because our two eyes are positioned slightly apart, each eye receives a slightly different image of the world, and the brain uses the difference between these two images to compute depth. To do this, the visual system must first solve the correspondence problem, which is the challenge of figuring out which point in the left eye's image matches which point in the right eye's image. A stereogram encodes disparity artificially by presenting slightly offset patterns to each eye
What is one of the key predictions made by feature integration theory?
Either (1) Visual search: Conjunction searches should require more attention than feature searches; or (2) Illusory conjunctions: In the absence of sufficent attention, binding should sometimes occur incorrectly
What is 4±2 chunks?
Miller originally claimed short-term memory holds about 7±2 items, but this was likely an overestimate because participants were grouping individual items into meaningful units. When this grouping strategy is controlled for, true short-term memory capacity is closer to this number
What are savings in relearning? How can these (and reaction times) be used to draw inferences about mental representations and processing?
Memory is not directly observable, and so the performance of retrieving information is a way to get at memory. Savings in relearning refers to improvements in memorization (in time or trials) to reach a certain performance threshold. Similarly, RT can be used to measure ease of retrieval or how well something has been learned.
What is the illusion of explanatory depth caused by internet searching
Fisher et al. (2015) found that searching for explanations online made people rate their own explanatory ability as higher — even for topics they never searched
What did Gauthier and Tarr (1997) find using Greebles, and what does it tell us about holistic face processing?
After training participants to become experts with novel "Greeble" objects, participants showed inversion effects similar to those seen with faces. This suggests holistic processing can arise from perceptual expertise rather than being a hardwired, face-specific mechanism
Name two pieces of evidence in favor of late selection models of attention.
(1)If an ambiguous word is presented to an attended message in the left ear, its interpretation can depend on an unattended message to the right ear / (2) If you hear 9 Aunt 6 in the unattended ear and Dear 7 Jane in the attended ear, participants will often report hearing Dear Aunt Jane
Miller observed that both the span of absolute judgment and the span of immediate memory converge on roughly 7, but these two cases are actually measuring very different things. What is the key difference between them?
The span of absolute judgment refers to how many distinct stimulus magnitudes people can reliably distinguish.
The span of immediate memory refers to how many items people can recall in sequence. They are not directly comparable because memory span can be boosted through chunking, which makes the two phenomena converge on the same number for very different reasons.
Are Marr's levels independent and mutually exclusive? Explain with an example.
No. The same phenomenon can be explained at multiple levels (they are not exclusive). They are also not independent. A computational level problem can constrain what the relevant algorithms are. (e.g., the computational problem of doing addition has many algorithms, but it also excludes some algorithms that do not do addition)
Melumad and Yun (2025) argue that LLM search (compared to web search) results in lower depth of learning. Why, and what is their evidence?
Melumad and Yun (2025) found that LLM search leads to shallower learning because it removes the "desirable difficulty" of navigating, interpreting, and synthesizing multiple sources yourself. Across seven experiments, LLM users spent less time with results, reported learning less, and wrote advice that was shorter, less factual, and more generic even when the exact same facts were presented in both conditions (Experiment 2). Independent raters also found LLM-based advice less informative and were less willing to adopt it.
Question: What is the desirable difficulty hypothesis that the effortful process of self-directed synthesis in web search builds deeper knowledge than passively receiving LLM summaries?
What do the Gestalt principles of good continuation, simplicity, and similarity have in common at a deeper level, and why does this allow them to work so reliably?
Each principle is a heuristic that reflects a genuine statistical regularity in natural scenes. The visual system applies a default interpretation when input is ambiguous, and these defaults work because they capture how objects and scenes are actually structured in the world. In other words, all three are examples of the visual system using prior knowledge to select the most likely interpretation.
Is feature integration theory right? Name at least one way that this original theory has held up and one way that it has not.
Yes and no. The key predictions are extremely robust and well-replicated (that conjunction search is harder than feature search and that you get some illusory conjunction mistakes). However, we discussed two contemporary challenges to this view in class. (1) Attention seems to be based on objecthood boundaries as well as low-level perceptual features like spatial distance. For two points equidistant from the initial fixation point, reaction time is *faster* for a point that appears to be part of the same object as the fixation point. I2) Attention seems to also vary based on on the agent's action intentions. If objects vary in color and orientation, participants make *more* intitial eye-movement errors with regards to orientation when their task is to point to the object rather than to grasp it.
How can expertise affect chunking?
Expertise allows people to recognize meaningful patterns as single chunks, effectively increasing how much information they can hold in short term memory. For example, when playing chess, experts can recall the positions of pieces on a board much better than novices, but only when the pieces are arranged in a realistic game configuration. When the pieces are placed randomly, the expert advantage disappears, which shows that experts are not simply better at memorizing individual locations but have learned to recognize familiar configurations as single units.