Aurkibidea: :: Library Catalog

Gorde:

Xehetasun bibliografikoak
Egile nagusia:	Liu, Ran
Formatua:	Recurso digital
Hizkuntza:
Argitaratua:	Zenodo 2026
Gaiak:	abstraction abstraction operators cognitive methodology knowledge organization learning design interdisciplinary research cross-domain transfer human-AI collaboration knowledge APIs conceptual modeling AI-assisted research
Sarrera elektronikoa:	https://doi.org/10.5281/zenodo.20259309
Etiketak:	Etiketa erantsi Etiketarik gabe, Izan zaitez lehena erregistro honi etiketa jartzen!

Aurkibidea:

The AI era has accelerated the production of information, explanations, methods, and models. The bottleneck has shifted toward cognitive organization: converting abundance into knowledge that can be navigated, reused, and transferred. Learners must decide what deserves attention. Researchers must synthesize expanding literatures. AI-assisted workflows must turn isolated outputs into reusable guidance for future reasoning and planning. Abstraction is the cognitive mechanism that makes this organization possible. It selects relevant distinctions, suppresses unnecessary complexity, and creates forms that support learning, reasoning, transfer, and action. Across disciplines, these moves have accumulated as powerful but tacit practices. They are embedded in models, interfaces, analogies, categories, mechanisms, and methods, while remaining dispersed across local vocabularies. The Abstraction Operator Framework makes these recurring moves explicit. It treats abstraction as selective re-expression of source content. Each abstraction preserves certain distinctions, hides or reduces others, and produces a new working form or abstract unit. By organizing these moves into a compact visual taxonomy, the framework provides a shared method language for learning, research, cross-domain transfer, knowledge API design, and human–AI collaboration. The framework distinguishes two broad modes. Horizontal Abstraction — Same Plane, New Working Form. Horizontal operators rework source content within its current working plane. They make content more workable through packaging, coordination, and recasting. Vertical Abstraction — New Abstract Unit, New Working Plane. Vertical operators derive abstract units that open another working plane. They enable reasoning through stabilization, formalization, rescaling, and transfer. <h3>Practical Implications</h3> 1. Abstraction Literacy: making hidden cognitive operations accountable Expertise depends on the disciplined control of abstraction. In learning and research, distinctions are selected, suppressed, recombined, named, and elevated into new working forms, often without an explicit account of the operation being performed. The framework makes these operations inspectable by separating the abstraction move from the content it acts upon. Users can identify what a move preserves, what it hides, what unit it introduces, and which conditions govern its use. This vocabulary strengthens conceptual accountability. It exposes category errors that often remain hidden beneath fluent explanation: a coarse label used as an explanation, an approximation treated as a general account, an analogy extended beyond its valid range. By making abstraction moves explicit, the framework allows users to examine the structure, function, and boundary of their own reasoning. 2. Learning and Research Navigation: entering complexity at the right level Complex domains become difficult when knowledge appears at the wrong level of abstraction. Excessive detail produces overload before structure becomes visible; excessive simplification produces fluency before understanding. A problem becomes workable only when its grain, view, mechanism, scale, or formal shape matches the cognitive task at hand. The framework supports deliberate navigation across these levels. It helps learners and researchers decide when to seek orientation, when to modularize content, when to change perspective, when to introduce an approximation, when to identify a mechanism, and when to recast a question into a standard problem form. Learning then becomes a controlled movement through abstraction levels, where each step reduces friction and increases structural understanding. 3. Knowledge Portability and Cross-Domain Transfer: turning local expertise into callable forms Knowledge gains transferable power when its internal complexity is shaped into forms that other contexts can call, adapt, inspect, and combine. Terminology can travel easily; usable structure travels only when assumptions, boundaries, and operating conditions remain attached. The framework clarifies this transformation by distinguishing labels, interfaces, parameters, mechanisms, problem forms, scale relations, analogies, and structure-preserving mappings. This distinction strengthens cross-domain work. It helps users identify the exact form in which knowledge is moving from one field to another. A metaphor can open intuition; a mapping can support reasoning; an interface can make complexity callable; a problem form can activate known solution logics. The framework therefore turns transfer from loose borrowing into a more disciplined practice of abstraction, adaptation, and boundary control. 4. Human–AI Collaboration: turning task performance into methodological guidance AI systems can produce plans, summaries, explanations, and solutions at scale. These outputs become more valuable when their underlying abstraction strategies can be named, compared, reused, and refined. An explicit abstraction layer makes this possible. It records the method behind the output: which distinctions were selected, which complexity was hidden, which working form was created, and which boundary shaped the result. The framework turns human–AI interaction into a cumulative methodological process. Humans can externalize tacit strategies as abstraction moves; AI systems can apply, test, and refine those moves across cases. Successful task execution then becomes methodological experience: evidence about which abstraction worked, under which conditions, and how it should guide future planning. This creates reusable soft priors for reasoning, transfer, and action. <h1 class="p2">Framework architecture</h1> The sixteen abstraction operators are organized into seven functional families. <h2>Horizontal Abstraction: Same Plane, New Working Form</h2> Horizontal operators rework source content within its current working plane. They create a more usable working form of the source. <h3>1. Package</h3> Chunking groups tightly coupled elements into one workable block. It is useful when multiple items are repeatedly used, learned, or operated together. Examples include memory chunks, workflow blocks, lesson units, skill packages, and set-meal bundling. Chunking differs from aggregation because the parts remain distinct but are handled as a unit. Encapsulation exposes a stable interface while hiding internal complexity. It is central to software, systems design, product design, organizational workflows, and knowledge packaging. APIs, interface contracts, black-box modules, information hiding, and protocol endpoints are examples from different domains. Encapsulation differs from Chunking because its key move is not grouping but controlled access. Resolution Coarsening replaces fine distinctions with coarser labels or partitions of the same descriptive space. It appears in binning, age bands, map zoning, time-windowing, coarse labels, and taxonomy broadening. It differs from Aggregate / Emergent Variable because it creates a coarser description, not a new higher-level variable or distribution. <h3>2. Coordinate</h3> Parameterization turns implicit variation into explicit adjustable parameters. It helps users navigate a family of possibilities: design knobs, model parameters, policy levers, simulation settings, and personalization controls. It differs from Axis Compression because it exposes variation rather than reducing the number of dimensions. Axis Compression condenses dimensions into fewer informative axes. It appears in latent dimensions, principal axes, strategic trade-off maps, perceptual maps, and design spaces. It is useful when many variables can be organized through a smaller number of explanatory directions. It differs from Resolution Coarsening because it reduces dimensional structure rather than lowering descriptive resolution. <h3>3. Recast</h3> View Switching changes the view while preserving relevant distinctions. It includes shifts such as local-to-global, static-to-dynamic, individual-to-system, object-to-process, function-to-mechanism, graph-to-matrix, or narrative-to-diagram. It is useful when the content is difficult because the current view makes the wrong relations salient. Idealization suppresses secondary effects to obtain a tractable first-order approximation. It appears in frictionless planes, rational-agent assumptions, simplified pathways, baseline cases, and controlled scenarios. Idealization differs from mere error because it is deliberate: it removes complications to reveal a core structure, while retaining a path for later refinement. <h2>Vertical Abstraction: New Abstract Unit, New Working Plane</h2> Vertical operators derive a new abstract unit that opens another working plane. They create a new unit through which reasoning, evaluation, scaling, solving, or transfer can proceed. <h3>1. Stabilize</h3> Invariant Quantity extracts a quantity that remains stable, conserved, or consistently trackable across contexts. It provides a stable meter for reasoning across change. Examples include conserved quantities, budgets, capacities, probability mass, information measures, and balances. It differs from Parameterization because the quantity is not merely a knob; it constrains or tracks behavior across contexts. Transformation Invariance identifies transformations that preserve relevant structure and group equivalent cases. It is useful when many cases look different but are equivalent under rotation, reflection, permutation, coordinate change, relabeling, or other transformations. It reduces unnecessary case-by-case reasoning by revealing what stays the same under change. <h3>2. Formalize</h3> Mechanistic Template extracts a reusable causal or process pattern that links inputs, states, roles, feedback, and outcomes. It appears in causal mechanisms, feedback loops, intervention pathways, workflow protocols, and agent skill routines. It is especially useful when the user needs to understand how something works. Problem Canonicalization recasts messy tasks as standard problem forms for known solver families. It turns an unclear task into a recognizable form such as search, optimization, classification, inference, control, allocation, scheduling, or constraint satisfaction. Its value is that once a problem form is recognized, known methods become available. Normative Framing sets values, duties, constraints, and priorities that guide judgment, choice, or action. It is broader than value framing alone. It can include ethical values, legal constraints, rights, responsibilities, duties, prohibitions, permissions, legitimacy conditions, and priority rules. It is useful whenever the key question is “what should count, guide, constrain, or decide?” <h3>3. Rescale</h3> Scale Renormalization shifts scale while preserving an effective description or relation. It appears when users zoom in or out and ask which variables, relations, or parameters remain useful. Examples include biological allometry, ecological scaling relations, urban scaling, and effective descriptors in physics or complex systems. It differs from Resolution Coarsening because it is not merely lowering resolution; it studies what remains meaningful across scale changes. Aggregate / Emergent Variable derives a higher-level variable, distribution, or order parameter from lower-level units, samples, or interactions. It includes aggregate variables such as income distributions, prevalence rates, and vote shares, as well as emergent variables such as temperature, volatility, magnetization, or collective oscillation. It differs from category coarsening because it creates a variable for measurement, modeling, or higher-level reasoning, not just a broader label. <h3>4. Transfer</h3> Structure-Preserving Translation maps objects, relations, and operations across domains so structure-dependent reasoning can transfer. It appears in algorithm transfer, model transfer, domain adapters, conceptual isomorphisms, and functor-like mappings. It is stronger than analogy because it attempts to preserve structure for reasoning patterns to remain valid. Analogical Mapping borrows a partial relational pattern from a familiar domain to guide intuition or hypotheses. It is useful for first contact, explanation, hypothesis generation, and creative transfer. It differs from Structure-Preserving Translation because it is partial, heuristic, and boundary-sensitive. A good analogy helps thinking begin; it does not by itself prove that the structure fully transfers. <h2>How to Use This Framework</h2> For learners, the framework provides a way to enter unfamiliar knowledge without being overwhelmed by surface detail. It helps learners recognize when they need a rough map, a modular breakdown, a different view, a first-order approximation, a mechanism, or a transferable analogy. Learning becomes less dependent on passively following a fixed sequence of topics and more dependent on choosing the abstraction move that makes the next layer of understanding available. For researchers, the framework supports conceptual diagnosis and theory construction. It helps distinguish a label from an explanatory variable, a heuristic analogy from a structure-preserving transfer, and an approximation from a robust model. This distinction is especially useful when building frameworks, reviewing literature, designing methods, or translating ideas across disciplines. The framework makes it easier to identify which conceptual operation is doing the real work in an argument. For educators and curriculum designers, the framework offers a vocabulary for designing learning paths as sequences of abstraction moves. A course or explanation can begin with orientation and coarser partitions, then move through modules, mechanisms, problem forms, scale shifts, and transfer structures. This supports instruction that stages complexity deliberately rather than presenting knowledge as a flat list of concepts. For interdisciplinary teams, the framework clarifies how knowledge travels across fields. It helps teams separate terminology transfer from method transfer, metaphor from structural mapping, and domain-specific detail from portable abstraction. This is valuable in collaborations where different disciplines use similar words for different structures, or different words for similar abstraction moves. For AI system designers and AI-assisted workflows, the framework can serve as a method layer above concrete tasks. It can guide how agents plan, decompose, reframe, simplify, compare, transfer, and diagnose their own outputs. Instead of recording only task results, an AI system can record which abstraction move shaped the result, under what conditions it worked, and how that experience should inform future planning. This supports cumulative methodological experience rather than isolated task execution. The Abstraction Operator Framework positions abstraction as a practical infrastructure for knowledge work. It makes conceptual operations visible, selectable, and reusable, while clarifying how complexity is reorganized, how new working planes are opened, and how knowledge becomes portable across contexts. This release provides the stable visual taxonomy of the framework as a top-level map for learning-path design, cross-disciplinary research, knowledge API design, AI-assisted reasoning, and future zoom-in modules. Each operator can be expanded into subtypes, domain examples, boundary tests, failure modes, and prompting patterns. In this sense, the framework functions both as a visual taxonomy and as a generative scaffold for developing new methods of learning, research, transfer, and human–AI collaboration. <h3>Selected Anchor Literature (Non-Exhaustive)</h3> This framework is anchored in several traditions where abstraction has been studied under different names and functions. In artificial intelligence, abstraction appears through representation mapping, reformulation, approximation, and problem-space transformation. In computer science, it appears through data abstraction, information hiding, modularity, and interface design. In cognitive science, it appears through concept learning, analogy, relational transfer, and sample-efficient generalization. In mathematics, physics, and complex systems, it appears through invariance, symmetry, conserved quantities, scale transformation, coarse-graining, and emergent variables. The contribution of this framework is to synthesize these traditions into a compact operator vocabulary for learning, research, cross-domain transfer, knowledge API design, and human–AI collaboration. <ul> <li> Gentner, D. (1983). Structure-mapping: A theoretical framework for analogy. Cognitive Science, 7(2), 155–170. </li> <li> Giunchiglia, F., & Walsh, T. (1992). A theory of abstraction. Artificial Intelligence, 57(2–3), 323–389.  </li> <li> Holte, R. C., & Choueiry, B. Y. (2003). Abstraction and reformulation in artificial intelligence. Philosophical Transactions of the Royal Society of London. Series B: Biological Sciences, 358(1435), 1197–1204.  </li> <li> Lake, B. M., Salakhutdinov, R., & Tenenbaum, J. B. (2015). Human-level concept learning through probabilistic program induction. Science, 350(6266), 1332–1338.  </li> <li> Liskov, B. (1988). Data abstraction and hierarchy. ACM SIGPLAN Notices, 23(5), 17–34.  </li> <li> Noether, E. (1918). Invariante Variationsprobleme. Nachrichten von der Gesellschaft der Wissenschaften zu Göttingen, Mathematisch-Physikalische Klasse, 235–257. </li> <li> Saitta, L., & Zucker, J.-D. (2013). Abstraction in artificial intelligence and complex systems. Springer. </li> <li> Wilson, K. G., & Kogut, J. (1974). The renormalization group and the ε expansion. Physics Reports, 12(2), 75–199.  </li> </ul>

Antzeko izenburuak