Claims

While the cholera map exemplifies visualization's power to extract a single important insight from one complete picture of a dataset, in real-world applications visualization is typically only one part of a larger sense-making process that includes collecting, analyzing, and understanding information iteratively.

The 'evidence file' phase of sense-making (originated in intelligence analysis) involves highlighting and extracting specific text from collected materials that contains information relevant to the decision at hand, distinguishing important details from surrounding context.

The sense-making process is often incomplete within a single session; the searcher may spend time researching, make progress but not reach a final decision, wait several days, and then resume—at which point the challenge becomes how to efficiently reconnect with the previous exploration thread.

What appears easy to recognize in retrospect (the cholera map pattern) was extremely difficult to discern when looking only at patient records, suggesting visualization's power lies in making existing patterns cognitively accessible rather than revealing entirely new information.

While people focus on the analyzing parts (data analysis or data science), the other parts of sense-making are very important but often overlooked; the speaker claims this is something people do much more often than they realize.

The key difference between sense-making and data analysis is that in sense-making the goal is not well-defined and the process is open-ended, iterative, and exploratory, whereas data analysis has clear objectives and defined methodologies.

During the camera research process, the searcher encounters unfamiliar technical terms (mirrorless, DSLR, macro forces) and must conduct additional research to understand what these terms mean before continuing to evaluate cameras, illustrating that sense-making is non-linear and includes recursive learning phases.

A practical indicator that someone is engaged in sense-making is having 20-30 open browser tabs, which represents their accumulated exploration paths and information collection.

The sense-making process includes a 'shoebox' phase where interesting items found during research are bookmarked or saved for later review and potential use in further analysis.

During the evidence file phase, the researcher learns which information is relevant to the problem being solved; for example, in camera selection, the researcher discovers that megapixel count may not be the relevant metric, but aperture is more important when taking baby photos because it enables faster shutter speeds.

The 'hypothesis' phase of sense-making involves formulating multiple candidate solutions from the collected evidence that all appear promising within the constraints, and then conducting detailed comparisons between them to reach a final decision.

The 'presentation' phase of sense-making involves explaining both the final answer and the exploration process that led to it, rather than having others independently traverse the same sense-making journey, representing a knowledge transfer goal.

Sense-making can be understood through the analogy of navigating a maze: you have a clear destination (the goal) but no map of the path, so you must explore, feel your way through, possibly repeat paths before finding the right direction, making it fundamentally an exploratory process.

Errors in AI systems and errors in human common sense are different kinds of failures; machine learning errors often reflect lack of common sense while human errors more often reflect lapses in attention or different values

Analogical reasoning is conventionally treated as a minor component of cognition relegated to one small zone, but analogy should instead be understood as the central mechanism that connects everything—analogous to an interstate freeway system that links all of cognition together, not a peripheral Delaware.

During a year at a research institute in Trento, Italy, Hofstadter faced the problem of how to greet people of different social relationships in Italian; he learned that 'ciao' is appropriate for close friends, 'salve' for intermediate relationships, and 'buongiorno' for formal relationships, requiring on-the-spot categorization of relationship status.

Analogy is the perception of common essence between two things, where the essence is subjectively perceived based on the particular frame of mind one is in at the time, rather than an abstract philosophical property; analogies occur inside the head as connections between mental representations, which are then projected onto the external world.

While flying from California to Indiana, Hofstadter read about the Atlantic Conveyor ship avoiding Hurricane Helene by deviating south; upon walking through Denver airport, he saw a woman with a suitcase crossing the corridor and spontaneously made the analogy: 'I'm the boat. She's the hurricane or the whirly girl. This corridor is the Atlantic Ocean.'

Most analogies happen so quickly and transiently that they would pass completely unnoticed and leave no effect on the world at all unless someone is paying deliberate attention to the mechanisms of their own thinking.

A snow shadow is the absence of snow on the ground where a tree blocks falling snow, analogous to and a generalization of a regular shadow cast by an object blocking light; both are instances of the same abstract concept of an obstruction preventing something from reaching a surface.

There is no fundamental difference between a single memory trace and a category or concept; a single instance can function as a concept, and conversely, a concept with many blurred instances is structurally equivalent to what we'd call a category.

Human minds have an unlimited capacity for chunking: lower-level concepts combine into higher-level conceptual units that become semi-invisible once chunked, and multiple layers of chunking create hierarchically organized concept networks where internal components fade from awareness.

The concept of 'chair' has extreme variability; one can purchase books containing thousands of different kinds of chairs, and the variety is mind-boggling, making feature-based definitions of category membership inadequate for explaining how we recognize a chair.

Hofstadter's colleague claimed that categorization is 'a matter of feature detectors', but Hofstadter disagrees, saying they are 'not on the same planet', because categorization at the level of abstract, non-sensory concepts (like 'probably', 'mess', 'well') cannot possibly rely on feature detectors.

Roger Schank pointed out that proverbs are situation labels; sometimes proverbs are further compressed into single-word labels, as in 'sour grapes' (lower case) which evokes the situation of someone rationalizing away a desire for something unattainable.

Hofstadter's friends Kelly and Dick called his home study an 'office' because in their household, the second-floor room with books, computers, and work materials was called a 'home office', which mapped perfectly to Hofstadter's room; Hofstadter called it his 'study' because his father's prototype had been called a study, not an office.

Thought is the highest level of abstraction-seeking, defined as putting one's finger on the essence of a situation and then bouncing back and forth between the actual situation and the perceived essence found in one's memories.

The standard proportional analogy format 'A is to B as C is to D' (e.g., 'shoe is to foot as glove is to hand') is commonly used on IQ tests and SATs but represents a narrow, overly formal definition of analogy that misses the true nature of how analogies function in cognition; restricting analogy to this formula puts analogy 'into a little tiny box.'

The phrase 'hope the package got there in one shape' (instead of 'in one piece' or 'intact') was produced by a native English speaker in a typed email and reveals competition between two idiomatic expressions, demonstrating that blending occurs even in formal written communication.

Two native English speakers conversing in a restaurant neither noticed when one said 'easy-go-lucky' (blending 'easy-going' and 'happy-go-lucky'), demonstrating that most blending events pass completely undetected by both speaker and listener.

Many analogies are reminding events that happen continuously and involuntarily; analogies are not there to serve purposes but simply happen, with evolution being the only purposeful driver behind them.

Hofstadter lists dozens of complex concepts we use fluently—'solar system,' 'slam dunk,' 'beltway,' 'plagiarism,' 'Wikipedia,' 'Bose-Einstein condensation,' 'quantum cryptography,' 'baby boomer,' 'the Fed,' 'pork belly futures,' 'affirmative action,' 'presidential lecture'—showing the remarkable range and depth of human conceptual systems.

If strokes of genius, personal insights, political decisions, dinner table conversations, me-too comments, random remindings, instantaneous categorization, and blends of all sorts are all made of analogies, then might not all of cognition also be made of analogies?—the cognition core hypothesis.

Categorization is the fundamental game of cognition, and analogy is the mechanism that creates and enables categorization by allowing the mind to perceive what something is and grasp its essence.

Repeated analogies expand and generalize concepts over time; by seeing new instances that share essence with existing concepts, we broaden the meaning and applicability of those concepts, both for individuals and for culture as a whole.

Hofstadter as an 8-year-old was fascinated by exponents and subscripts; upon seeing his physicist father's papers with notation like 'X sub 1, X sub 2', he hypothesized that subscripting might be an operation on numbers analogous to superscripting's exponentiation, but was disappointed when his father explained subscripts were merely a naming convention for variables.

Categories are not restricted to nouns and visual objects; words of all parts of speech—adverbs, prepositions, pronouns, verbs—are categories evoked by appropriate circumstances; most categories are not visual or sensory at all, and theories of categorization that focus on visual objects miss where the real action is.

Hofstadter's assistant Helga said 'Every morning I took a capsy ride,' blending 'cab' and 'taxi' into 'capsy,' which reveals subterranean competition between two synonymous words for the same vehicle.

The concept of 'mess' (as in a visually chaotic or disorganized state) cannot be explained by feature detectors; there are no specific neurons or feature detectors that fire when seeing a mess.

Word choice involves competition between multiple candidate words in 'semantic space'; when one word is overwhelmingly activated (like 'dog' when seeing a dog), no error occurs; but when multiple words are nearly equally activated, audible competition manifests as speech errors, hesitations, and word blends.

When Hofstadter said 'finding finding' and prolonged the F sound, he was experiencing subterranean competition between 'find' and 'figure out,' which became audible only because he hesitated to articulate a sound that he consciously registered as slightly too lengthy.

The concept of 'airline hub' is built from many nested concepts including: ball→roll→wheel→spokes→hub→centrality; vehicle concepts; company concepts; trip→legs→routes; time→appointment→timetable→network; network→node; get; buy→money→economizing→downsizing; city; airport.

For 50 years I was working on artificial neural networks assuming the real brain was better and artificial neural nets were worse, but a few months ago I suddenly changed my mind because digital systems have better learning algorithms and knowledge-sharing capabilities than biological systems.

Human teaching via language is extremely inefficient. When I produce strings of words to communicate with you, you have to figure out how to change trillions of connection strengths in your brain so you would have said the same thing. This is distillation—a very different and much slower process than how digital computers can directly share weight changes.

Spreading activation theory explains forgetting as a functional feature—if you didn't forget, you'd have a cacophony of noise in memory retrieval; forgetting helps you focus on what's salient and relevant to the current situation by tuning weights on memory connections.

Memories are not actually lost completely—decades later, properly cued, people can retrieve things they thought were forgotten, though you need to be careful about false memories created by suggestive cueing.

Memory champions and people using explicit memory systems don't lose other normal human knowledge—there's no resource management constraint; they simply know more. This is supported by experimental literature and the fact that multilingual people have absurdly high numbers of memorized things.

Predictive utility theory suggests the probability of retrieving a memory depends on how much the brain predicts it will be useful—repeated practice increases the predicted utility, and practicing in varied situations increases utility prediction across broader distributions.

Humans and animals learn through the development of mental models of the world—abstract causal models that allow them to understand object permanence, solidity, intuitive physics (gravity, inertia), and natural categories before or independently of language learning, and these models are formed through observation and minimal interaction.

There is no general Master algorithm and no special intelligence fluid; instead there are a tremendous number of special cases that we learn and then code into our brains.