AI Expert Twin: Capturing Expert Cognition for Human-Centred, Practice-Based Learning

Annie Yuan; Judy Kay; Kalina Yacef; Xiaohua Chen

arxiv: 2605.01401 · v2 · submitted 2026-05-02 · 💻 cs.HC · cs.AI

AI Expert Twin: Capturing Expert Cognition for Human-Centred, Practice-Based Learning

Annie Yuan , Xiaohua Chen , Kalina Yacef , Judy Kay This is my paper

Pith reviewed 2026-05-12 01:32 UTC · model grok-4.3

classification 💻 cs.HC cs.AI

keywords expert cognitiontacit knowledgeAI in educationpractice-based learninglearner modellinghuman-centred AIdecision processes

0 comments

The pith

Expert cognition can be captured as a three-layer computable model for use in AI-driven practice-based learning.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces the AI Expert Twin framework to make tacit expert knowledge explicit and usable in educational AI systems. It formalizes this knowledge as a three-layer structure covering procedural actions, semantic concepts, and decision processes, while also tracking how value-laden preferences, trade-offs, and uncertainty influence real judgments. A cultural heritage workshop case study shows the approach can be applied with actual experts to produce structured representations. The framework is presented as transferable to vocational education and creative industries. Its purpose is to support scalable practice-based learning that keeps expert heuristics transparent and preserves learner agency.

Core claim

The central claim is that expert cognition can be formalized as a three-layer representation of procedural actions, semantic concepts, and decision processes that accounts for value-laden preferences, trade-offs, and uncertainty, allowing this knowledge to be captured from experts and integrated into AI-powered educational systems, as shown feasible in a cultural heritage workshop.

What carries the argument

The three-layer representation of expert cognition that structures procedural actions, semantic concepts, and decision processes while incorporating value-laden preferences, trade-offs, and uncertainty.

If this is right

The captured representations can be embedded into AI systems to provide learners with access to expert heuristics.
The approach supports practice-based learning across multiple domains including vocational education and creative industries.
Integration maintains transparency in the AI and preserves learner agency during the learning process.
The framework provides a foundation for further development of human-centred AI applications in education.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The three-layer structure could be tested for compatibility with existing learner models to see if it improves personalization in practice-based settings.
If the layers prove stable, the same capture process might be applied in high-stakes domains such as clinical training to simulate expert judgment under time pressure.
Transfer to digital creative tools could allow AI systems to replicate not just steps but also the trade-off reasoning of master practitioners.

Load-bearing premise

That tacit expert knowledge and context-sensitive judgement can be adequately captured and structured into computable three-layer representations without substantial loss of nuance or real-world applicability.

What would settle it

A follow-up session in which the same experts review the captured three-layer models and identify major mismatches with how they actually decide in new, unscripted situations, or where the models cannot be applied in a second domain without extensive re-engineering.

Figures

Figures reproduced from arXiv: 2605.01401 by Annie Yuan, Judy Kay, Kalina Yacef, Xiaohua Chen.

**Figure 1.** Figure 1: AI Expert Twin conceptual framework. Expert practice is captured as multimodal data and transformed into a cognitive model of expert cognition. This Expert Twin informs an AI tutor that supports learners through adaptive guidance and feedback. The framework is organised around five interacting layers: the data layer, cognitive layer, AI tutor layer, user layer, and governance layer. It addresses transpa… view at source ↗

**Figure 2.** Figure 2: Three-layer representation of expert cognition in the AI Expert Twin framework, with a personality/tension engine modulating the decision layer through tension vector T. cus, and C orientation toward tradition or change. These weights influence how experts select among candidate actions and explain decisions under uncertainty. Together, these three layers form a computable model of expert cognition. The p… view at source ↗

**Figure 3.** Figure 3: Jade carving workshop and multimodal data capture. Left: jade carving master Tiecheng Zhang and selected signature works, including the Beijing Olympics jadeinlaid gold medal. Right: interviews conducted by the research team led by Professor Xiaohua Chen, and multimodal records of the master’s workflow from design and carving to apprenticeship-based instruction. 4 Initial Instantiation: Cultural Heritage… view at source ↗

read the original abstract

Tacit knowledge embedded in expert practice remains difficult to capture, formalise, and scale. While AI-driven educational systems have advanced personalisation, learner modelling, affective support, and self-regulated learning, they less often model the tacit reasoning and context-sensitive judgement that underpin expert practice in practice-based domains. This paper introduces the AI Expert Twin, a cognition-centric framework that models expert knowledge as structured, computable representations of procedural actions, semantic concepts, and decision processes. The framework also considers how value-laden preferences, trade-offs, and uncertainty shape expert judgement in practice. We formalise expert cognition as a three-layer representation and capture knowledge from experts under this model, laying the groundwork for integration into AI-powered educational system. A case study in a cultural heritage workshop demonstrates the feasibility of the approach in a real-world setting. The framework is designed to be transferable across domains such as vocational education and creative industries. By embedding expert heuristics into AI while maintaining transparency and learner agency, the AI Expert Twin offers a novel path towards scalable, practice-based learning and invites further research on ethical, human-centred applications of AI in education.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper offers a three-layer framework for modeling expert tacit knowledge including values and uncertainty, but the case study provides no concrete details on elicitation or validation.

read the letter

The main takeaway is that this is a framework proposal for an AI Expert Twin that structures expert cognition into procedural actions, semantic concepts, and decision processes while adding layers for value-laden preferences, trade-offs, and uncertainty. It targets a real gap in AI education systems that focus more on personalization than on capturing context-sensitive judgment in practice-based domains like vocational training or creative fields. The cultural heritage workshop is presented as a feasibility demonstration, which is a fair starting move for this kind of work. What stands out as new is the explicit inclusion of uncertainty and values in the model rather than treating expert reasoning as purely procedural or semantic. That extension feels like a legitimate step beyond standard learner modeling approaches. The paper does a decent job laying out why current AI tutoring tools fall short on tacit knowledge and sketching how the three-layer structure could support more transparent, human-centered systems. The soft spot is the lack of substance around the case study. The description claims feasibility but supplies no specifics on how experts were interviewed, how the layers were mapped, what artifacts were produced, or how the model was checked against actual practice. Without those details the central claim that the representation preserves nuance and transfers across domains stays untested. This is the kind of paper that would interest researchers in AI in education who are already working on learner models or vocational training tools. Readers looking for concrete implementations or validated results will come away wanting more, but the framework itself could serve as a useful reference point for designing future systems. It deserves peer review because the idea is coherent and the gap it identifies is genuine, even though the current version would need expansion on the empirical side before publication.

Referee Report

2 major / 2 minor

Summary. The paper introduces the AI Expert Twin framework, which models expert cognition as a structured three-layer representation of procedural actions, semantic concepts, and decision processes, while incorporating value-laden preferences, trade-offs, and uncertainty. It claims to capture this knowledge from experts in a computable form suitable for integration into AI educational systems, demonstrates feasibility via a case study in a cultural heritage workshop, and positions the approach as transferable to domains such as vocational education and creative industries for human-centred, practice-based learning.

Significance. If the three-layer model can be shown to capture tacit expert knowledge with minimal loss of context-sensitive judgement, the framework would offer a transparent, cognition-centric alternative to existing AI tools in education that often overlook tacit reasoning. This could meaningfully advance scalable practice-based learning while preserving learner agency, particularly in domains where expert heuristics are central.

major comments (2)

Case study description: The feasibility claim rests on a cultural heritage workshop but provides no details on the elicitation protocol for deriving the three-layer representations, the specific data or artefacts collected from experts, the mapping of value-laden preferences and uncertainty into the model, or any validation against real expert practice or outcomes. This is load-bearing for the central claim that the representation preserves nuance and is transferable.
Framework formalization: The description of the three-layer model (procedural actions, semantic concepts, decision processes) does not specify how the layers interact, how trade-offs and uncertainty are encoded as computable elements, or provide even a schematic example of the representation for a concrete task. Without this, it is difficult to assess whether the model is genuinely structured and computable as claimed.

minor comments (2)

Abstract: The phrasing 'capture knowledge from experts under this model' is vague; specifying the knowledge elicitation methods or tools used would improve clarity for readers.
Terminology: The repeated use of 'AI Expert Twin' as an invented entity would benefit from a brief definition or diagram early in the paper to distinguish it from related concepts like digital twins or expert systems.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and insightful comments, which highlight important areas where the manuscript can be strengthened to better support its central claims. We address each major comment below and will revise the manuscript to incorporate the requested clarifications and details.

read point-by-point responses

Referee: Case study description: The feasibility claim rests on a cultural heritage workshop but provides no details on the elicitation protocol for deriving the three-layer representations, the specific data or artefacts collected from experts, the mapping of value-laden preferences and uncertainty into the model, or any validation against real expert practice or outcomes. This is load-bearing for the central claim that the representation preserves nuance and is transferable.

Authors: We acknowledge that the current version of the manuscript presents the cultural heritage workshop case study at a high level and does not include sufficient methodological specifics. This is a fair observation that weakens the support for the feasibility and transferability claims. In the revised manuscript, we will expand the case study section to describe the elicitation protocol in detail, including the structured interview and observation methods used to derive the three-layer representations. We will provide concrete examples of the data and artefacts collected (such as expert verbal protocols, workshop artefacts, and decision logs), explain the mapping process for value-laden preferences, trade-offs, and uncertainty (e.g., via explicit elicitation of expert criteria and confidence assessments), and report on validation steps such as expert feedback on the fidelity of the resulting representations to their actual practice. These additions will make the demonstration more robust. revision: yes
Referee: Framework formalization: The description of the three-layer model (procedural actions, semantic concepts, decision processes) does not specify how the layers interact, how trade-offs and uncertainty are encoded as computable elements, or provide even a schematic example of the representation for a concrete task. Without this, it is difficult to assess whether the model is genuinely structured and computable as claimed.

Authors: We agree that the framework section would benefit from greater precision and illustration to demonstrate its structured, computable character. In the revision, we will add a schematic diagram and worked example of the three-layer model applied to a specific task from the case study (e.g., a heritage documentation procedure). This will show the content of each layer, the interactions between layers (such as how semantic concepts constrain procedural actions and how decision processes incorporate values and uncertainty), and the encoding of trade-offs and uncertainty as computable elements (for instance, through weighted preference structures or probabilistic qualifiers). These changes will allow readers to evaluate the model's suitability for AI integration more directly. revision: yes

Circularity Check

0 steps flagged

No circularity: purely conceptual framework proposal with no derivations or predictions

full rationale

The paper introduces a cognition-centric framework by defining a three-layer representation of expert knowledge (procedural actions, semantic concepts, decision processes) plus value-laden elements, then illustrates feasibility via a cultural heritage workshop case study. No equations, fitted parameters, predictions, or first-principles derivations exist that could reduce to inputs by construction. The framework is presented as an author-defined structure for future integration into AI systems, with the case study serving as an existence demonstration rather than a statistical or self-referential validation. No load-bearing self-citations, uniqueness theorems, or ansatzes are invoked. The work is self-contained as a conceptual proposal; any limitations concern evidence strength or transferability, not circular reduction of claims.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests primarily on a domain assumption that expert tacit knowledge is formalizable in the described structure; no free parameters or invented entities with independent evidence are introduced beyond the framework name.

axioms (1)

domain assumption Expert cognition, including tacit knowledge and context-sensitive judgement, can be captured as structured, computable representations in three layers (procedural actions, semantic concepts, decision processes) that also incorporate value-laden preferences, trade-offs, and uncertainty.
This premise underpins the entire framework and is stated directly in the abstract as the basis for the model.

invented entities (1)

AI Expert Twin no independent evidence
purpose: A cognition-centric framework for modeling and capturing expert knowledge to integrate into AI-powered educational systems.
The framework is the core proposed contribution; no external falsifiable evidence for its components is provided beyond the high-level case study mention.

pith-pipeline@v0.9.0 · 5502 in / 1308 out tokens · 45703 ms · 2026-05-12T01:32:05.125033+00:00 · methodology

Review history (2 revisions) →

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

We formalise expert cognition as a three-layer representation... procedural layer, semantic layer, and decision layer... personality/tension vector T={R, I, G, E, C}

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

13 extracted references · 13 canonical work pages

[1]

Handbook of Artificial Intelligence in Education pp

Aleven, V., Rowe, J., Huang, Y., Mitrovic, A.: Domain modeling for aied systems with connections to modeling student knowledge: A review. Handbook of Artificial Intelligence in Education pp. 127–169 (2023)

work page 2023
[2]

The journal of the learning sciences4(2), 167–207 (1995)

Anderson, J.R., Corbett, A.T., Koedinger, K.R., Pelletier, R.: Cognitive tutors: Lessons learned. The journal of the learning sciences4(2), 167–207 (1995)

work page 1995
[3]

In: Advances in intelligent tutoring systems, pp

Bull, S., Kay, J.: Open learner models. In: Advances in intelligent tutoring systems, pp. 301–322. Springer (2010)

work page 2010
[4]

Research & Practice in Technology Enhanced Learning20(2025)

Chou, C.Y., Chan, T.W., Chen, Z.H., Liao, C.Y., Shih, J.L., Wu, Y.T., Hung, H.C., et al.: Defining ai companions: a research agenda–from artificial companions for learning to general artificial companions for global harwell. Research & Practice in Technology Enhanced Learning20(2025)

work page 2025
[5]

Computers & Education40(3), 255–269 (2003)

Chou, C.Y., Chan, T.W., Lin, C.J.: Redefining the learning companion: the past, present, and future of educational agents. Computers & Education40(3), 255–269 (2003)

work page 2003
[6]

In: Handbook of human-computer interaction, pp

Corbett, A.T., Koedinger, K.R., Anderson, J.R.: Intelligent tutoring systems. In: Handbook of human-computer interaction, pp. 849–874. Elsevier (1997)

work page 1997
[7]

Intelligent Decision Technologies p

Guo, G., Hu, J.: Making tacit knowledge explicit: Generative ai’s role in enhancing apprenticeship systems. Intelligent Decision Technologies p. 18724981251397523 (2025)

work page 2025
[8]

Jackson, P.: Introduction to expert systems (1985)

work page 1985
[9]

Knowledge6(1), 1 (2025)

Khalili, N., Jahanbakht, M.: Unveiling the unspoken: A conceptual framework for ai-enabled tacit knowledge co-evolution. Knowledge6(1), 1 (2025)

work page 2025
[10]

Saracco, R.: Applying cognitive digital twins to professional education. Tech. rep., IEEE Digital Reality Initiative (September 2019), https://ieee.org, co-chair, IEEE Digital Reality Initiative

work page 2019
[11]

Journal of Manufacturing Systems64, 372–389 (2022)

Tao, F., Xiao, B., Qi, Q., Cheng, J., Ji, P.: Digital twin modeling. Journal of Manufacturing Systems64, 372–389 (2022)

work page 2022
[12]

Psychology Press (2013)

VanLehn, K.: Student modeling. Psychology Press (2013)

work page 2013
[13]

International Journal of EBusiness and EGovernment Studies16(3), 355–384 (2024)

Yang, F.: Exploring the role of information systems in documenting tacit knowl- edge and driving transformation in traditional handicrafts. International Journal of EBusiness and EGovernment Studies16(3), 355–384 (2024)

work page 2024

[1] [1]

Handbook of Artificial Intelligence in Education pp

Aleven, V., Rowe, J., Huang, Y., Mitrovic, A.: Domain modeling for aied systems with connections to modeling student knowledge: A review. Handbook of Artificial Intelligence in Education pp. 127–169 (2023)

work page 2023

[2] [2]

The journal of the learning sciences4(2), 167–207 (1995)

Anderson, J.R., Corbett, A.T., Koedinger, K.R., Pelletier, R.: Cognitive tutors: Lessons learned. The journal of the learning sciences4(2), 167–207 (1995)

work page 1995

[3] [3]

In: Advances in intelligent tutoring systems, pp

Bull, S., Kay, J.: Open learner models. In: Advances in intelligent tutoring systems, pp. 301–322. Springer (2010)

work page 2010

[4] [4]

Research & Practice in Technology Enhanced Learning20(2025)

Chou, C.Y., Chan, T.W., Chen, Z.H., Liao, C.Y., Shih, J.L., Wu, Y.T., Hung, H.C., et al.: Defining ai companions: a research agenda–from artificial companions for learning to general artificial companions for global harwell. Research & Practice in Technology Enhanced Learning20(2025)

work page 2025

[5] [5]

Computers & Education40(3), 255–269 (2003)

Chou, C.Y., Chan, T.W., Lin, C.J.: Redefining the learning companion: the past, present, and future of educational agents. Computers & Education40(3), 255–269 (2003)

work page 2003

[6] [6]

In: Handbook of human-computer interaction, pp

Corbett, A.T., Koedinger, K.R., Anderson, J.R.: Intelligent tutoring systems. In: Handbook of human-computer interaction, pp. 849–874. Elsevier (1997)

work page 1997

[7] [7]

Intelligent Decision Technologies p

Guo, G., Hu, J.: Making tacit knowledge explicit: Generative ai’s role in enhancing apprenticeship systems. Intelligent Decision Technologies p. 18724981251397523 (2025)

work page 2025

[8] [8]

Jackson, P.: Introduction to expert systems (1985)

work page 1985

[9] [9]

Knowledge6(1), 1 (2025)

Khalili, N., Jahanbakht, M.: Unveiling the unspoken: A conceptual framework for ai-enabled tacit knowledge co-evolution. Knowledge6(1), 1 (2025)

work page 2025

[10] [10]

Saracco, R.: Applying cognitive digital twins to professional education. Tech. rep., IEEE Digital Reality Initiative (September 2019), https://ieee.org, co-chair, IEEE Digital Reality Initiative

work page 2019

[11] [11]

Journal of Manufacturing Systems64, 372–389 (2022)

Tao, F., Xiao, B., Qi, Q., Cheng, J., Ji, P.: Digital twin modeling. Journal of Manufacturing Systems64, 372–389 (2022)

work page 2022

[12] [12]

Psychology Press (2013)

VanLehn, K.: Student modeling. Psychology Press (2013)

work page 2013

[13] [13]

International Journal of EBusiness and EGovernment Studies16(3), 355–384 (2024)

Yang, F.: Exploring the role of information systems in documenting tacit knowl- edge and driving transformation in traditional handicrafts. International Journal of EBusiness and EGovernment Studies16(3), 355–384 (2024)

work page 2024