Critically Engaged Pragmatism in Practice
- Critically Engaged Pragmatism is a reflective, purpose-sensitive framework that merges iterative inquiry with critical power analysis and discursive accountability.
- It guides AI evaluation, algorithmic fairness, and sustainable education by aligning purpose-specific reliability with participatory critique and ethical validation.
- The approach mandates rigorous validation methodologies and adaptive practices to ensure tools and assessments remain contextually relevant and socially just.
Searching arXiv for papers on Critically Engaged Pragmatism and closely related "critical pragmatism" formulations. Critically Engaged Pragmatism is a practice-centered, reflective, and deliberative orientation that combines pragmatist commitments to experimental inquiry, iteration, and purpose sensitivity with critical commitments to power analysis, legitimacy, social justice, and discursive scrutiny. In the literature, the exact label is used unevenly. "Critically Engaged Pragmatism" is explicitly articulated as a scientific norm for AI science evaluation tools, where it requires vigorous scrutiny of the purposes and purpose-specific reliability of such tools (Lee, 13 Jan 2026). In adjacent work on algorithmic fairness, the operative term is "critical pragmatism," rooted in John Forester’s urban planning theory and applied to fairness as a class of wicked problems; that work does not define "critically engaged pragmatism" as a separate term, but its practice orientation, deliberative stance, and attention to conflict and power substantially align with the label (Gosciak et al., 4 May 2026). A related paper in sustainable mathematics education likewise does not explicitly use the phrase, but describes a "critical pragmatic bridge" joining critique, engagement, and pragmatist implementation (Müller, 24 Apr 2025).
1. Definition and domain of application
In its most explicit formulation, Critically Engaged Pragmatism (CEP) is “a scientific norm articulated to guide the design, interpretation, use, and evaluation of AI science assessment tools.” Its central requirement is that scientific communities “vigorously scrutinize the purposes to which such tools are put and their purpose-specific reliability.” Under this norm, AI tools are not treated as objective arbiters of scientific credibility; rather, they are themselves subject to the critical discursive practices through which scientific credibility is established (Lee, 13 Jan 2026).
In the algorithmic fairness literature, the corresponding formulation is a “reflective, deliberative, power-aware practice for designing fairer algorithms under wickedness.” It foregrounds what practitioners actually do in the face of conflict and power, and it explicitly “considers both process and outcomes.” The problem setting is not one in which a definitive technical rule can settle disagreement; instead, fairness work is situated in contexts marked by plural values, contested frames, irreversible stakes, and structural asymmetries (Gosciak et al., 4 May 2026).
In sustainable mathematics education, critical pragmatism is defined as an overview of “pragmatism's focus on experience and practically minded action” with “critical theory's commitment to analyzing power, promoting social justice, and pursuing emancipation.” There, the concept is organized around existential sustainability, ethical practice, and a sequenced implementation strategy designed to move from classroom culture to epistemic diversification and then to socio-critical modeling (Müller, 24 Apr 2025).
A plausible implication is that Critically Engaged Pragmatism is best understood not as a single fixed doctrine, but as a family of related approaches unified by four recurring features: purpose sensitivity, iterative inquiry, discursive accountability, and sustained attention to power.
2. Intellectual lineage
The pragmatist component is anchored in James, Peirce, and especially Dewey. The relevant emphasis is anti-absolutism: inquiry proceeds experimentally, under uncertainty, and amid durable disagreement. In the educational formulation, this appears as a rejection of “Either-Ors” and of uncontextualized claims; in the AI science evaluation formulation, it appears as reliability-for-purpose and epistemic fit rather than abstract construct matching (Lee, 13 Jan 2026, Müller, 24 Apr 2025).
A second line of inheritance comes from Donald Schön’s reflective practice. In the fairness formulation, this appears as “reflection-in-action”: the practitioner frames a problem, acts, encounters “backtalk” from the situation, and revises the framing. This model dissolves a strict theory/practice divide and treats learning as situated, improvisational, and repertoire-building rather than as the application of universal rules (Gosciak et al., 4 May 2026).
The critical dimension is drawn from several sources. In AI science evaluation, CEP integrates pragmatism with Longino’s procedural objectivity. Four social norms are central: public venues for critique, uptake of criticism, shared standards for reasoning, and critical engagement with diverse perspectives. Objectivity is thus procedural and social, not guaranteed by metricized outputs or predictive accuracy alone (Lee, 13 Jan 2026).
In algorithmic fairness, the concept is explicitly rooted in Forester’s critical pragmatism in urban planning. This lineage adds an emphasis on listening, mediation, communication, and the ways relations of power and authority shape which frames and knowledge claims appear plausible. It also imports the planning literature’s treatment of wicked problems: problems with no definitive formulation or stopping rule, no exhaustive solution set, and no neutral or final test of success (Gosciak et al., 4 May 2026).
In sustainable mathematics education, critical pragmatism also draws on critical theory, critical pedagogy, decolonial and Indigenous perspectives, and the ethics turn in mathematics education. The bridge mechanism is explicit: existential sustainability supplies breadth and moral significance, while ethics frameworks provide actionable norms and workflows (Müller, 24 Apr 2025).
3. Core commitments and methodological structure
Across the literature, Critically Engaged Pragmatism is defined less by a closed formalism than by a set of operational commitments. In AI science evaluation, these commitments are named directly: purpose-indexed evaluation, epistemic fit, social adjudication, and anti-abstraction vigilance. Reliability is assessed relative to a specific, articulated purpose; measures, labels, data, and modeling choices must fit that purpose; communities must determine and contest purposes through discursive practices; and practitioners must resist “inference by false ascent,” the decontextualization and repurposing of credibility markers into abstract, portable labels that invite epistemically inapt reuse (Lee, 13 Jan 2026).
The same paper gives a concrete process model. It begins with purpose articulation and proxy analysis, proceeds through participatory co-design, transparency practices, and validation aligned with the stated purpose, and then adds portability governance, uptake and revision, and ongoing audit. The operative question is not whether a tool is reliable in the abstract, but whether its measures, data, model, and validation protocols directly track the evidential needs and inferential standards appropriate to the use case (Lee, 13 Jan 2026).
In algorithmic fairness, the process is framed through four practice heuristics rather than a checklist. The first is reflect-in-action: iterative prototyping, scenario or sensitivity analysis, simulation as “sketching,” and decision logs that record framing shifts. The second is public deliberation through recurring forums such as listening sessions, workshops, and town halls that are designed for trust-building and genuine dialogue rather than one-way consultation. The third is critical listening and creative negotiation: a hermeneutic stance attentive to information distortions, legitimacy claims, and conflicts that require mediated compromise. The fourth is practice-oriented storytelling, through which rich first-person accounts become a repertoire for transfer across unique contexts (Gosciak et al., 4 May 2026).
In sustainable mathematics education, the structure is staged rather than heuristic. Stage 1 cultivates an ethical classroom culture through communicative norms, psychological safety, and care. Stage 2 engages ethnomathematics, understood as embedded in ethics and focused on the recovery of cultural dignity. Stage 3 addresses complex sustainability problems through socio-critical modeling. A parallel workflow is provided by ten ethical pillars ranging from “Deciding whether to begin” to “Emergency response strategies,” together with levels of teacher ethical awareness from Level to Level $4$ (Müller, 24 Apr 2025).
This suggests that Critically Engaged Pragmatism is methodologically plural but architecturally consistent: it seeks operational forms that preserve critique while remaining action-guiding.
4. Major articulations across research areas
Three recent arXiv papers present the clearest formulations of this orientation in distinct domains.
| Domain | Formulation | Operational emphasis |
|---|---|---|
| AI science evaluation tools | CEP as a scientific norm | Purpose-specific reliability, epistemic fit, social adjudication, portability guardrails |
| Algorithmic fairness | Foresterian critical pragmatism applied to wicked problems | Reflect-in-action, public deliberation, critical listening, creative negotiation, practice stories |
| Sustainable mathematics education | Critical pragmatic bridge | Ethical classroom culture, ethnomathematics, socio-critical modeling |
In AI science evaluation, the central problem is that crises in peer review capacity, replication, and AI-fabricated science create pressure to automate evaluation. CEP responds by arguing that tools should not be treated as portable indicators of “rigor,” “replicability,” “impact,” or “confidence” unless their proxies and validations are shown to fit the specific decision they are meant to support. The paper’s case study of ML models predicting “replicability” illustrates the danger: a model trained against one operationalization can be redescribed at a more abstract level and then repurposed for “confidence in findings,” even when that reuse lacks epistemic fit (Lee, 13 Jan 2026).
In algorithmic fairness, the orientation is motivated by the claim that fairness problems resemble urban planning’s wicked problems. Mortgage lending, school choice, public safety, and hiring involve allocative decisions that affect life chances, deep value conflicts, strong power asymmetries, and no consensus definition of fairness. Formal fairness methods are treated as necessary but insufficient: they provide precision and tractability, yet they do not resolve governance, resource allocation, participation, or conflict mediation. The case studies show how deliberation, critical listening, iterative piloting, and narrative documentation alter what counts as fair practice in mortgage underwriting, school choice design, and feminicide counterdata collection (Gosciak et al., 4 May 2026).
In sustainable mathematics education, the same broad orientation becomes pedagogical. The paper argues that critical scholarship has often exposed problems of neutrality, power, and unsustainability without providing sufficiently actionable classroom pathways. Its response is a sequenced implementation strategy: first establish a sustainable classroom culture approximating inclusive speech; then engage ethnomathematics to challenge neutrality, purity, and universality; then tackle climate change, biodiversity loss, resource allocation, and environmental justice through modeling, data handling, scrutiny, explainability, and reflection on the politics of mathematical artefacts (Müller, 24 Apr 2025).
5. Relation to formalism, objectivity, and participation
A central theme across the literature is that Critically Engaged Pragmatism does not reject formal methods; it refuses their sufficiency. In the fairness literature, purely formal frameworks are credited with precision, tractability, and tool support, and are explicitly retained as instruments within reflective inquiry. Their limitation is narrowness: impossibility results, simplifying assumptions, and an inability to address governance, participation, and structural power mean that formal criteria alone cannot resolve wickedness (Gosciak et al., 4 May 2026).
The AI science evaluation literature makes an analogous argument against “metric-driven objectivism” and “algorithmic neutrality.” Predictive accuracy against a given ground truth does not settle whether the ground truth is appropriate to the purpose. CEP therefore reframes objectivity as procedural and social, produced by public critique, uptake of criticism, shared standards, and diversity of perspectives. A numeric output may be useful, but it does not become authoritative independently of those community practices (Lee, 13 Jan 2026).
The participatory dimension is similarly qualified. In algorithmic fairness, the paper warns against merely “consultative” participation that ignores power. It urges movement beyond interviews toward recurring public deliberation, mixed-format dialogue, inclusive logistics, and explicit confrontation with legitimacy claims and information distortions. The approach aligns with participatory and co-design traditions, data feminism, algorithmic realism, and pipeline-aware fairness, but adds conflict mediation and power analysis as non-optional elements (Gosciak et al., 4 May 2026).
In mathematics education, comparable concerns appear in the rejection of myths of mathematical neutrality, purity, and universality and in the insistence that all mathematical artefacts have politics. Classroom communication is therefore not a neutral delivery channel; it is part of the ethical and epistemic substance of the work. This gives the approach an explicitly engaged character, even where the exact term “Critically Engaged Pragmatism” is not used (Müller, 24 Apr 2025).
6. Critiques, limitations, and open questions
The literature presents Critically Engaged Pragmatism as action-guiding but not as a catch-all solution. In the fairness context, wicked problems are not “solvable,” and communicative planning traditions face familiar critiques: they may underplay structural power, rely too heavily on planner neutrality, exhibit consensus bias, or remain Western-centric. There are also domain-specific limits: urban planning’s public-sector orientation and local scale differ from scaled private-sector technology settings, and effectiveness depends on at least some institutional commitment to fairness. Risks include tokenistic deliberation, capture by powerful stakeholders, burnout of marginalized participants, and reversion to formal metrics without process (Gosciak et al., 4 May 2026).
In AI science evaluation, practical constraints include closed-source tools, meager labeled datasets, persistent normative disagreement, and the possibility that standardization will marginalize diverse epistemic communities or reproduce inequities. CEP can still require documentation, disclaimers, external evaluation, and community oversight under such conditions, but full adherence may be constrained when transparency and modifiability are limited (Lee, 13 Jan 2026).
In mathematics education, existential sustainability is explicitly described as underdetermined, and the implementation of ethnomathematics and discourse ethics faces curricular, institutional, and cultural constraints. The paper cautions against tokenism, notes that interest in other cultures’ ethnomathematics may vary, and emphasizes the need for careful framing, local collaboration, and institutional support, especially when work moves beyond technical modeling into the political dimensions of mathematical practice (Müller, 24 Apr 2025).
The open questions are correspondingly structural. The fairness literature asks how deliberation can redistribute power without entrenching consensus, how critically engaged pragmatism can scale in large private platforms, how iterative experimentation can be balanced against one-shot stakes and accountability, and how practice stories can be codified while protecting sensitive information (Gosciak et al., 4 May 2026). The AI science evaluation literature asks related questions in another register: how to govern portability across purposes, how to secure purpose-appropriate validation amid pressure for large datasets and rapid deployment, and how to maintain public critique and documented uptake when evaluation systems are embedded in crises of scale and trust (Lee, 13 Jan 2026).
Taken together, these works present Critically Engaged Pragmatism as a norm of inquiry and intervention for domains where stakes are high, purposes are contested, and formal abstraction is both necessary and dangerous. Its distinctive claim is not that conflict can be eliminated, but that disciplined, inclusive, power-aware, purpose-sensitive practice can make action more legitimate, more accountable, and more epistemically fit under conditions of irreducible complexity.