Addressing Trust in AI Systems through Education: A Didactic Perspective
Pierre Haritz, Hendrik Krone, Thomas Liebig
cs.CY · cs.AI
Quick verdict
Presents ICE-T, a didactic framework making trust calibration an explicit ML-education goal; coherent and literature-based, yet an untested hypothesis.
Summary
This concept paper proposes ICE-T, a didactic framework for machine learning education that treats trust calibration as an explicit learning objective. ICE-T integrates three facets: intermodal transfer via Bruner's enactive-iconic-symbolic modes, computational thinking operationalized as the Use-Modify-Create progression, and explanatory thinking supported by PETSP-ML, a four-stage adaptation of CRISP-ML that elevates model understanding to a distinct pre-training stage. The authors combine a qualitative evaluation of existing educational tools (Sect. 4) with two systematic reviews of K-12 ML activities (126 and 133 activities) documenting gaps: supervised-learning activities cluster at low abstraction levels (~60%), open-ended creation appears only twice, and technical and societal perspectives co-occur in roughly 3% of 133 activities (four combine both). They synthesize trust research - calibrated trust (Lee and See), algorithm aversion (Dietvorst et al.), and literacy effects showing recalibration rather than uniform increase in trust - to argue that ICE-T's three facets supply the cognitive mechanisms of calibrated reliance: rich mental models, graduated process control, and error contextualization. The paper candidly labels the predicted UMC-trust gradient a testable hypothesis, and the conclusion concedes the connection is theoretical, with no classroom evaluation. Main limitations: absent empirical validation; a tool evaluation too thinly reported to audit; and an evidence base drawn largely from the authors' own research program.
Review complete
Assessment at a glance
5
risk flags identified
● high 0
● medium 3
● low 2
4 strengths · 4 weaknesses · 5 author questions
AI text signal = Binoculars zero-shot detector score over 20 sampled passages (Fast-DetectGPT co-reported) — absence does not prove human authorship, and a flag does not imply misconduct.
Levels reflect verification complexity — not a quality score. Evolve never rates your paper.
Key claims under review
- Treating ML as an opaque black box simultaneously causes superficial conceptual learning and uncalibrated trust in AI systems (Sects. 1, 7.2).
- The three ICE-T facets supply the cognitive mechanisms the trust-calibration literature identifies as drivers of appropriate reliance: representational richness, graduated process control, and error contextualization (Sect. 7.4).
- Current K-12 practice systematically undersupports these mechanisms: 69 of 89 supervised-learning activities keep learners in the User role, and technical and societal perspectives co-occur in ~3% of 133 activities (Sect. 7.3).
- Trust calibration should be adopted as an explicit educational objective of ML education, with ICE-T as the means of achieving it (Sects. 7.5, 8).
Strengths
Honest scoping of the central claim
Section 7.4 explicitly labels the predicted UMC trust gradient 'a testable hypothesis derived from the aversion literature rather than an established result', and Sect. 8 concedes the ICE-T-trust connection is 'theoretical' and untested - rare candor for a framework paper.
Mechanism-to-driver mapping is precise and falsifiable
Section 7.4 maps each facet to a specific literature-backed trust driver: intermodal transfer to mental-model richness [26,14,39], UMC to graduated process control (Dietvorst 2018 slight-modification effect; Cheng & Chouldechova 2023 process vs outcome control), explanatory thinking to error contextualization (Dietvorst 2015). Each mapping yields a checkable prediction.
Gap claims anchored in published systematic reviews
Synthesis contradicts the intuition that education simply raises trust
Weaknesses
Zero empirical evaluation of the proposed framework
Section 4 tool evaluation is unauditable
Key evidence generalizes from adult decision-support settings
Evidence base is internally concentrated
Full Review Report
$1.99 · one-timeUnlock the complete preliminary analysis Evolve produced for this paper.
- All 4 strengths & 4 weaknesses, analyzed in detail
- 5 risk flags with severity, section & reviewer questions
- Methods & statistical evidence assessment
- 5 questions a peer reviewer would ask
- Full six-dimension evaluation report & recommendations
One-time purchase · this paper only
One-time purchase, this paper only. AI-generated preliminary analysis — not peer review.
Frequently asked questions
What is the ICE-T framework for teaching machine learning?
ICE-T is a didactic framework from TU Dortmund researchers that integrates three facets of ML teaching: intermodal transfer based on Bruner's enactive-iconic-symbolic modes of representation, computational thinking operationalized through the Use-Modify-Create progression, and explanatory thinking supported by PETSP-ML, a four-stage classroom adaptation of the CRISP-DM data-mining process. The framework is described in a position paper (arXiv:2609.02453) arguing that trust calibration should be an explicit learning objective.
Does AI literacy increase trust in AI systems?
According to the evidence the paper synthesizes, not uniformly. Literacy interventions tend to recalibrate reliance rather than raise trust: users learn when to rely on a model and when to withhold trust. Some studies even find AI literacy reduces trust in specific domains (e.g., medical AI) by producing 'informed skepticism', while very high familiarity can lead to over-reliance.
Is there empirical evidence that the ICE-T framework improves trust calibration?
No. The paper itself states that the connection between ICE-T and trust calibration is theoretical at this stage: it derives testable predictions from the trust and algorithm-aversion literature but has not been evaluated in any classroom study. The evidence cited concerns adult decision-support users, not K-12 learners.
Who is ICE-T intended for, and what age groups does it target?
ICE-T is grounded in K-12 machine learning education and the systematic reviews it draws on cover primary and secondary school activities. The authors note that transfer across age groups and to ML paradigms beyond supervised learning remains an open question for future work.
About this review
This is an AI-generated preliminary review by Evolve.Pub, produced from the openly available text of the paper. It is not peer review and does not represent the authors' views. The canonical record of the paper is at the publisher link above.
- Figures 1 and 2 (Trust Matrix; PETSP-ML process diagram) are referenced but their visual content is not recoverable from text extraction; assessments of them rely on captions only.
Go deeper on this paper
Wrote a paper? Stress-test it before reviewers do.
Upload your manuscript for the same structured scrutiny — reasoning, statistics, evidence, figures, and reproducibility.