Source-linked AI summary
Between Algorithm (AI) and Intuition (Human): Preserving Designer Agency in AI-Assisted Sensemaking of Qualitative UX Data
Md Haseen Akhtar
TL;DR
The paper examines whether AI can accelerate qualitative sensemaking without suppressing the designer’s interpretive agency. Through a structured comparison of AI analysis and designer close reading of 20 student responses, it finds that AI is most valuable as an instrument that amplifies attention and comprehensiveness while leaving ethical and situated judgments to designers.
Problem
AI-assisted qualitative analysis may flatten contradictory, uncertain, and value-laden user feedback, raising the question of how to gain efficiency without suppressing designer interpretation.
Method
The study compares Claude 3.5 Sonnet’s thematic analysis of 20 student responses with traditional close reading and qualitative coding, examining divergent interpretations.
Results
AI rapidly organizes frequent patterns and surfaces low-frequency suggestions, but its most valuable role is amplifying designer attention without directing ethics or situated judgment.
Takeaways & Limitations
AI should remain an instrument rather than an authority, supporting comprehensiveness and rapid organization while designers make contextual value judgments.
Abstract
from arXiv · showhide
The integration of AI into qualitative design research presents a fundamental tension: how do we leverage AI while preserving the subjective, intuitive judgments that define design expertise? This paper examines this question through a case study of analyzing 20 user responses about video conferencing platforms for educational contexts. We argue that AI sensemaking tools risk flattening the rich data patterns, amplifying contradictory textures of user feedback into sterile categories thereby transforming design research from an interpretive craft into a mechanical sorting exercise (rigid and formal). Through comparative analysis of AI-assisted sensemaking versus human-centered approaches to the same dataset, we identify when algorithmic efficiency enhances understanding and when it diminishes the designer's interpretive agency (uncovering hidden needs, critical enquiry, what if enquiries, making decisions, having trade-offs). We present a framework for augmented sensemaking that positions AI as an instrument for amplifying human judgment rather than replacing it. Our findings suggest that the most valuable role for AI in design research is not to eliminate subjectivity, but to make it more intentional, reflective, and accountable.
INTRODUCTION
The paper frames AI-assisted qualitative sensemaking as a tension between computational efficiency and the interpretive agency required to uncover uncertainty, values, and deeper user concerns. It asks whether AI can assist analysis without suppressing designers’ situated judgments.
- User statements about engagement tracking can support multiple interpretations, including discomfort with surveillance, conditional acceptance, or unstated privacy concerns.
- AI can accelerate thematic coding, pattern recognition, and cognitive-load reduction, but design judgments remain subjective and resistant to formalization.
- Design research constructs meaning through selection, emphasis, and framing rather than simply discovering objective patterns.
- Sensemaking also involves social negotiation over what counts as evidence, which interpretations are valid, and whose perspectives matter.
- AI categorization raises questions about whose categories and criteria determine which user themes become important, because technical systems encode worldviews and power relations.
METHOD
The study compares Claude 3.5 Sonnet’s thematic analysis with a designer’s close reading of 20 student responses about educational video conferencing. The dataset’s contradictions and affective nuances provide the material for examining where AI clarifies or obscures interpretation.
- METHOD: Researchers provided Claude 3.5 Sonnet with the complete dataset and context, requested themes with supporting quotes, and compared its analysis with traditional close reading and qualitative coding.
- METHOD: The comparison examined specific instances where AI-generated and designer interpretations diverged.
- METHOD: The study analyzed 20 student responses comprising approximately 4,500 words across 40 questions about video conferencing for teaching and learning.
- DATASET: The responses concerned Zoom, Google Meet, and Microsoft Teams and contained contradictions, unstated assumptions, and affective nuances that resisted straightforward categorization.
- DATASET: Contradictory preferences included dark mode alongside eye-fatigue concerns and restricted private messaging alongside its accessibility value for anxious students.
Emotional Labor and Uncertainty
Students’ softened criticism and uncertain language contain interpretive cues that simple sentiment analysis can miss. These cues may indicate concealed frustration, conflicting values, or hypotheses requiring follow-up research.
- Softened criticism can conceal frustration and suggest that students have more substantial concerns than they state directly.The example of an interface being “mostly good” with “minute tweaks” is interpreted as emotional labor that avoids sounding demanding.
- The phrase “kind of” signals uncertainty when a student recognizes tension between private chat’s social connection and its distraction from instruction.
- A simple positive sentiment label can miss distinctions between comfort, necessity, and exceptional usefulness that affect feature prioritization.
- Students’ cautious statement that privacy should not be hindered suggests anxiety about surveillance that keyword analysis may fail to capture.
COMPARATIVE SENSEMAKING: AI VERSUS DESIGNER
AI efficiently organizes frequent patterns and surfaces rare suggestions, while designer interpretation preserves differences in conviction that can shape prioritization. The comparison therefore positions AI as useful for coverage but insufficient for nuanced judgment.
- What AI does well: Within minutes, AI produced taxonomies of engagement, collaboration, and accessibility features while quantifying sentiment distributions.
- What AI does well: Chat appeared in 20 of 20 responses, screen sharing in 19 of 20, and layout customization in 18 of 20.
- What AI does well: AI surfaced a voice-typing suggestion mentioned by only one student, helping preserve rare insights that manual analysis might overlook.
- Interpretive differences: AI classified “I think it will make a difference” and “It should definitely be a must have” as positive sentiment despite their different levels of conviction.
- Interpretive differences: Those differences can inform prioritization because designers may weigh emphatic certainty more heavily or investigate uncertainty through follow-up questions.
2. Missing Situated Enquiry
AI often converts situated uncertainty, institutional critique, and ethical tension into feature categories, while designers can interrogate the assumptions and contexts behind responses.
- A comment about teachers not encouraging group discussions can indicate institutional critique or pedagogical choice rather than a request for more prominent breakout rooms.
- Users’ doubts about whether features will be adopted reveal uncertainty that may make education and onboarding more relevant than additional functionality.
- AI did not challenge survey framing that treated surveillance as desirable, whereas designers could question its implications for teacher-student power relations.
COMPARATIVE SYNTHESIS: WHEN AI HELPS AND WHEN IT HINDERS
AI assists sensemaking through speed, coverage, organization, and quantification, but human interpretation remains necessary for uncertainty, absences, critical framing, contradictions, and domain-specific meaning.
- AI provides breadth and consistency through rapid coding, frequency counting, preliminary taxonomies, and quantification across responses.
- 95 per cent of students (19/20) reported frequently switching tabs, while 85 per cent found this distracting, surfacing a high-priority pain point quickly.
- AI can amplify low-frequency suggestions, such as one student’s voice-typing idea addressing accessibility and faster input.
- AI grouped questions into thematic clusters, creating preliminary structure that revealed greater discussion of engagement than collaboration.
- AI identified a 75 per cent dark-mode preference while preserving varied rationales, supporting decisions about default or customizable themes.
- Human reading adds interpretive depth by distinguishing uncertainty and necessity, noticing meaningful silences, and questioning surveillance assumptions.
- Contradictory needs around private chat require ethical prioritization between focus and accessibility rather than averaging or majority-rule resolution.
- Comments about breakout rooms can point to lecture-based teaching norms, requiring interventions beyond interface changes.
FRAMEWORK TOWARDS AUGMENTED SENSEMAKING
The proposed augmented-sensemaking framework positions AI as an instrument for amplifying human judgment rather than replacing it.
- The work-in-progress framework treats AI as an instrument that amplifies human judgment rather than replacing it.
1. Distributed Labor between AI and Designer
The framework divides sensemaking labor by assigning mechanical analysis to AI while preserving designers’ authority over interpretive decisions and requiring human review.
- AI should handle initial coding, frequency counting, and quote extraction, while designers retain authority over prioritization, contradiction resolution, and reading silences.
- Human review should cover ambiguous cases and outliers, with raw quotes preserved alongside AI summaries for contextual verification.
2. Contradiction as Resource and not Errors
The paper treats contradictory user feedback as a productive design resource rather than an error to eliminate. AI should surface tensions for human discussion and value judgments, not resolve them through consensus.
- Contradiction as Resource and not Errors: Contradictions between minimal interfaces and requests for more features reveal design spaces where stakeholder needs may diverge.These tensions can also expose underlying power dynamics.
- Contradiction as Resource and not Errors: Prompt AI to identify conflicting responses and contradictory themes instead of averaging them into consensus.
- Contradiction as Resource and not Errors: Use contradictions as workshop discussion prompts that require teams to make explicit judgments about which user needs to privilege.
- Contradiction as Resource and not Errors: Treat AI-generated analyses as provocations by testing which perspectives support its categories and which alternative readings they foreclose.
- Contradiction as Resource and not Errors: Compare AI categories with human-generated codes, using disagreements as sites for methodological reflection and further investigation.
IMPLICATIONS FOR DESIGN PRACTICE
The implications for design practice center on balancing AI’s speed with the interpretive value of ambiguity and preserving accountable human judgment. Practitioners should retain original responses because summaries can lose contextual texture.
- IMPLICATIONS FOR DESIGN PRACTICE: AI can analyze qualitative data in minutes rather than days, but rapid analysis may produce efficient designs lacking insight.Sensemaking also involves dwelling with ambiguity, noticing unexpected patterns, and developing tacit knowledge about users.
- IMPLICATIONS FOR DESIGN PRACTICE: Designers should make interpretive choices explicit and accountable because selecting analytical criteria remains subjective.Documenting judgments such as prioritizing accessibility over efficiency creates a design rationale.
- IMPLICATIONS FOR DESIGN PRACTICE: Maintain access to original user responses because AI summaries can lose contextual texture essential to interpretation.Examples include personal opinion signaled by wording and uncertainty indicated by question marks.
Develop Critical AI Literacy
Critical AI literacy requires design teams to understand AI’s capabilities and limits, recognize problematic assumptions, and retain authority to override its suggestions. AI is most useful as an instrument that organizes and flags data without determining ethical or situated design judgments.
- Develop Critical AI Literacy: Design teams need training to critically evaluate AI outputs, including what pattern matching can and cannot establish.
- Develop Critical AI Literacy: Teams should recognize when AI perpetuates problematic assumptions and override its suggestions using domain expertise.The paper specifically connects this need to its engagement-tracking example.
- Develop Critical AI Literacy: AI can rapidly organize qualitative data, flag low-frequency pain points, and support comprehensiveness without deciding direction, ethics, or situated knowledge.
- Develop Critical AI Literacy: Augmented sensemaking should make subjectivity intentional, reflective, and accountable rather than attempt to eliminate it.