h t t p s : / / w w w . u k y . e d u / ~ g m s w a n 3 / 5 7 5 / A r g y r i s _ 1 9 8 2 . p d f

Ingest: Organizational Learning and Management Information Systems

Type: kb/sources/types/ingest-report.md

Classification

This is a conceptual essay presented as a theoretical journal paper: it develops a causal account from prior literature, organizational observations, and illustrative cases, but reports no systematic method or controlled evaluation. Author: Chris Argyris, identified in the reprint as a Harvard Graduate School of Education scholar, writes from his organizational research and field experience; that is a strong first-party signal for the theory's intended meaning, not independent confirmation of it. The snapshot is a Winter–Spring 1982 DATA BASE reprint of an article first published in 1977 in Accounting, Organizations and Society.

Summary

Argyris explains persistent MIS implementation failures as failures of organizational learning rather than only of technology, usability, or executive support. Organizations readily correct routine errors without questioning governing objectives and policies—single-loop learning—but struggle with double-loop inquiry when correction threatens status, competence, loyalty, or existing control. The paper attributes that struggle to Model I theories-in-use and mutually reinforcing defensive loops that hide errors, hide the inability to correct them, and sometimes camouflage the camouflage itself. It also contrasts local information systems, rich in concrete process and tacit causal detail, with distant systems, whose abstraction supports comparison and public validation but weakens access to local mechanisms. Hierarchical responsibility for information gaps can then provoke withholding, ritual compliance, and tighter control, so an apparently stronger MIS may reproduce the conditions that make its evidence unreliable. Read the paper for this conceptual mechanism and its vocabulary, not for a validated intervention: Argyris ends by calling for theory-informed case research because workable redesign recommendations had not yet been established.

Quotes

  • Source extract (verbatim): People design and guid e their behavior by the use of theories of action that they hol d in their heads . Espoused theories of action are the theories that people report are governing their actions . Theories-inuse are the theories of action that actually govern thei r actions .
  • Source location: Paragraph opening "What is a Model I theory-in-use?", organizational-learning section, DATA BASE Winter-Spring 1982 reprint of the 1977 article
  • Source extract (verbatim): rganizations learn through individuals acting a s 0 agents for them .
  • Source location: Same section, first sentence of the paragraph introducing the organizational learning system (drop-cap letter rendered separately by the PDF text capture)
  • Source extract (verbatim): An organization may be said to learn to the extent that it identifies and corrects errors . This requirement, in turn , implies that learning also requires the capacity to kno w when it is unable to identify and correct errors . There are two types of learning that are important . They are single and double-loop learning .
  • Source location: Organizational-learning framing paragraph following the MIS inner-contradictions discussion, DATA BASE Winter-Spring 1982 reprint of the 1977 article
  • Source extract (verbatim): This is singleloop learning, because the underlying program is not questioned . The overwhelming amount of learning done in a n organization is single-loop because it is designed to identify and correct errors so the job gets done and the action re mains within stated policy guidelines .
  • Source location: Thermostat paragraph introducing single-loop learning, same section (line-break hyphenation in "singleloop" and "re mains" is as captured)
  • Source extract (verbatim): The trouble arises when the technology is not effectiv e and when the underlying objectives and policies must b e questioned .
  • Source location: Opening of the paragraph after the thermostat example, same section
  • Source extract (verbatim): However, if the underlying objectives require reexamination or if someone or some department is going to get i n trouble, it will be much more difficult to identify and t o correct the errors . The former case is the equivalent of th e thermostat following its program and orders . The latter i s the equivalent of the thermostat questioning its order; that is what is meant by double-loop learning .
  • Source location: Paragraph closing the thermostat contrast and naming double-loop learning, same section
  • Source extract (verbatim): But it can be shown that Model I theories-in-use lea d to effective problem solving primarily for issues that do no t require that the underlying assumptions of Model I theories-in-use be questioned (single-loop learning) . Model I theories-in-use do not make it possible for people to hav e problem-solving skills that question the governing values o f their theory-in-use (double-loop tearing) (Argyris & Schon , 1974) .
  • Source location: Paragraph defining Model I theories-in-use, organizational learning system section ("double-loop tearing" is an OCR artifact in the capture)
  • Source extract (verbatim): If that above is true, then organizations seem to have the capacity to learn primarily those lessons that are self-sealing because they maintain the status quo .
  • Source location: End of the local-versus-distant MIS discussion, immediately before the organizational learning system section
  • Source extract (verbatim): Errors that tend not to be correctable (see bottom o f column 7) include (1) those whose discovery is a threat t o the individual's system of hiding error and his or her in ability to correct error, (2) errors that predispose primar y inhibiting loops because they are threatening Model I values (win-don't lose, suppress feelings and so on), and (3 ) errors whose correction violates organizational norms . The errors that tend to be uncorrectable are camouflaged, the primary and secondary inhibiting loops associated with them are camouflaged and the camouflage is camouflaged with the development of protective activitie s such as "j .i .c . files" (just in case the superiors ask) (8) .
  • Source location: Discussion of columns 7 and 8 of Figure 1, organizational learning systems that inhibit error detection and correction

Connections Found

This source is a primary conceptual anchor for the organizational-learning lineage. It is evidence for Learning is not only about generality, but only for Argyris's distinction between correction within governing commitments and inquiry into those commitments; the note's mapping from loop type to a generality axis remains Commonplace's interpretation. It is also evidence for An action model matters only through its consumption path: an espoused theory can be legible yet behaviorally inert, whereas a theory-in-use is identified by its role in governing conduct.

As a boundary-setting counterpoint, the paper compares with Theory-mediated learning may improve sample efficiency under structured shifts. Argyris supplies an earlier account of action-guiding theory and revision of governing commitments, but not explicit theory search, reach assessment, addressable retention, or sample-efficiency evidence. It also compares with Design for a Brain: Ashby's second feedback changes supplied parameters when a viability bound is breached, while Argyris's double loop questions objectives, policies, or values through organizational inquiry. The shared loop vocabulary does not establish a shared mechanism.

Extractable Value

  1. An operativity test for stated action theories — the espoused-theory/theory-in-use distinction supports requiring an action model to participate in behavior selection rather than treating declaration, storage, or legibility as evidence that it governs conduct. [quick-win]
  2. An update-target account of learning loops — single-loop learning changes action while holding objectives and policies fixed; double-loop learning makes those governing commitments objects of inquiry. This is more precise than treating the distinction as merely narrow versus broad learning. [quick-win]
  3. A self-sealing failure mechanism — unilateral-control theories-in-use can amplify ambiguous or inaccessible information, generate defensive group dynamics, conceal threatening errors, and then reinforce the original control response. This supplies a mechanism for why adding monitoring may reduce rather than improve error visibility. [deep-dive]
  4. A representation tradeoff across organizational distance — local information preserves process detail and situated causal clues but resists comparison and independent validation; distant information gains comparability by abstracting away those mechanisms. KB routing and review should preserve both views when decisions require local diagnosis and cross-case oversight. [deep-dive]
  5. Control can corrupt the evidence it is meant to improve — foolproofing, tighter intervention, and reduced local discretion can signal distrust and induce withholding, doctored reports, withdrawal, or literal rule-following. An information-control intervention therefore changes the behavior that produces its own inputs. [deep-dive]
  6. A boundary for computational transfer — the paper offers no addressable retained theory, theory-search procedure, controlled reflection operation, benchmark, or sample-efficiency result. It should ground the historical concepts and organizational failure mechanism, not claims about an agent architecture's learning performance. [just-a-reference]

Limitations (our opinion)

The paper's central account is a theory-backed interpretation of selected organizational observations, not a controlled study with explicit sampling, measures, comparison conditions, or outcome data. Model I and the primary and secondary inhibiting loops can explain many forms of resistance after the fact, but the paper does not state observations that would discriminate that account from simpler explanations such as poor system design, conflicting incentives, or ordinary information loss. Its local-versus-distant MIS contrast is deliberately stylized, and the paper tests neither a hybrid information design nor the proposed alliance among line managers, MIS specialists, and behavioral scientists. The conclusions should therefore guide diagnosis and hypothesis formation, not establish that defensive learning is the dominant cause in a new organization.

Transfer to agent-operated KBs needs a further boundary. The source does not establish the explicit search, addressable representation, retention, or reach-assessment pathway conjectured in Theory-mediated learning may improve sample efficiency under structured shifts, and its double-loop mechanism should not be conflated with Ashby's direct parameter-reorganization mechanism. Finally, the retained pdftotext snapshot has pervasive spacing artifacts and apparent OCR errors; any future verbatim support should be checked against the printed page before it is added to the quote pool.

Update Learning is not only about generality to replace its secondary Infed citation with this ingest using the exact marker (snapshot required) and to label the generality-axis mapping as Commonplace's interpretation rather than Argyris's claim.