KNIT: Computational Boundary Objects for Real-Time Convergence in Interdisciplinary Teams

Paper Title

KNIT: Computational Boundary Objects for Real-Time Convergence in Interdisciplinary Teams

Publication Info

• Topic area: AI-mediated tools for interdisciplinary collaboration in healthtech
• Keywords: AI-mediated collaboration, computational boundary objects, interdisciplinary teams, healthtech, knowledge boundaries, value propositions, real-time negotiation, Carlile’s 3T framework, stakeholder perspectives, co-design

Background and Problem

• Problem / challenge: Interdisciplinary teams in healthtech face persistent challenges in aligning diverse disciplinary perspectives, stakeholder priorities, and problem framings, particularly during rapid iteration. Existing tools inadequately support real-time negotiation and convergence.
• Significance: Misalignment in healthtech teams can derail product development, leading to inefficiencies, costly pivots, and products that fail to meet clinical, technical, or market needs.
• Motivation and related work: While user-centred design, Agile, and co-design methods address some aspects of interdisciplinary collaboration, they lack robust mechanisms for resolving epistemic and political tensions. Traditional boundary objects are static and fail to adapt to dynamic team needs. AI tools have primarily focused on divergence or post-hoc synthesis, leaving a gap in real-time convergence support.

Solution

• Proposed approach: KNIT (Knowledge Negotiation and Integration Tool), an AI-mediated framework that generates computational boundary objects to facilitate real-time convergence in interdisciplinary teams.
• Novelty:
  1. Introduction of computational boundary objects that dynamically reframe team inputs to support knowledge transfer, translation, and transformation.
  2. Empirical evidence of AI-mediated convergence patterns across syntactic, semantic, and pragmatic boundaries using Carlile’s 3T framework.
  3. Design principles for computational boundary objects, including regenerative plasticity, anonymisation, and stakeholder anchoring.
  4. Demonstration of AI as a cognitive mediator in interdisciplinary collaboration.
• Procedure and key techniques:
  1. Individual asynchronous and anonymised input collection, standardised by AI into draft value propositions.
  2. Team Canvas module for collaborative review, semantic clustering, and stakeholder-oriented negotiation.
  3. Perspective Reflection module using AI-generated stakeholder problem statements to prompt reframing and alignment.

Results

• Concrete findings:
  • KNIT facilitated knowledge boundary crossing with success rates of 95.0% (syntactic), 86.3% (semantic), and 84.8% (pragmatic).
  • Problem Flip component achieved a 93.2% success rate in pragmatic transformation by anchoring discussions in stakeholder perspectives.
  • Similar Expressions component supported semantic alignment with an 88.3% success rate by clustering near-synonymous phrases.
• Advantage over baselines:
  • Unlike traditional boundary objects, KNIT’s computational boundary objects dynamically adapt to team inputs, enabling real-time negotiation and alignment.
  • AI reframing tools like Problem Flip outperformed reorganisation tools by actively transforming team perspectives.
• Experiments / evaluation:
  • Conducted workshops with seven healthtech teams (28 participants) across 190 boundary-crossing episodes.
  • Evaluated using Carlile’s 3T framework and thematic analysis of team interactions.
  • Components like Problem Flip and Similar Expressions were key drivers of successful boundary crossing.
• Limitations and future work:
  • Risk of false consensus due to reliance on AI-generated artefacts.
  • Limited generalisability due to the focus on healthtech teams in the UK.
  • Future work should explore longitudinal effects, domain-specific adaptations, and cultural variations in AI receptivity.

Summary

KNIT introduces computational boundary objects as AI-generated artefacts that dynamically support interdisciplinary teams in real-time convergence. Empirical evaluation with healthtech teams demonstrated high success rates in crossing syntactic, semantic, and pragmatic knowledge boundaries. Key components like Problem Flip and Similar Expressions facilitated stakeholder-centred reframing and semantic alignment. KNIT’s anonymisation and structured workflows reduced interpersonal tensions and enabled collective stance formation. These findings advance boundary theory and provide actionable design principles for AI-mediated collaboration in high-stakes interdisciplinary settings.