Paper deep dive

Drag or Traction: Understanding How Designers Appropriate Friction in AI Ideation Outputs

A. Baki Kocaballi, Joseph Kizana, Sharon Stein, Simon Buckingham Shum

Year: 2026Venue: arXiv preprintArea: cs.HCType: PreprintEmbeddings: 31

Intelligence

Status: succeeded | Model: google/gemini-3.1-flash-lite-preview | Prompt: intel-v1 | Confidence: 95%

Last extracted: 3/31/2026, 2:06:54 AM

Summary

This paper introduces 'Generative Friction' as a design strategy for AI tools to counter design fixation by intentionally introducing disruptions (fragmentation, delay, ambiguity) into AI outputs. Through a qualitative study with six designers, the authors demonstrate that while friction can transform AI output from a 'finished product' into 'semi-finished material' that invites human contribution, its effectiveness is moderated by 'Friction Disposition'—a user's propensity to interpret resistance as either an invitation or an obstruction.

Entities (5)

A. Baki Kocaballi · researcher · 100%Friction Disposition · concept · 100%Generative Friction · concept · 100%Design Fixation · phenomenon · 95%SPARK v1 · software · 95%

Relation Signals (3)

Friction Disposition → moderates → Generative Friction

confidence 95% · Friction Disposition emerged as a potential moderator: high-disposition users treated friction as 'liberating,' while low-disposition users experienced drag.

SPARK v1 → implements → Generative Friction

confidence 90% · We developed SPARK v1, a custom AI ideation tool... the presentation of output varied across four conditions.

Generative Friction → mitigates → Design Fixation

confidence 90% · We propose Generative Friction... designed to transform it from finished product into semi-finished material, inviting human contribution rather than passive acceptance.

Cypher Suggestions (0)

No Cypher suggestions yet.

Abstract

Abstract:Seamless AI presents output as a finished, polished product that users consume rather than shape. This risks design fixation: users anchor on AI suggestions rather than generating their own ideas. We propose Generative Friction, which introduces intentional disruptions to AI output (fragmentation, delay, ambiguity) designed to transform it from finished product into semi-finished material, inviting human contribution rather than passive acceptance. In a qualitative study with six designers, we identified the different ways in which designers appropriated the different types of friction: users mined keywords from broken text, used delays as workspace for independent thought, and solved metaphors as creative puzzles. However, this transformation was not universal, motivating the concept of Friction Disposition, a user's propensity to interpret resistance as invitation rather than obstruction. Grounded in tolerance for ambiguity and pre-existing workflow orientation, Friction Disposition emerged as a potential moderator: high-disposition users treated friction as "liberating," while low-disposition users experienced drag. We contribute the concept of Generative Friction as distinct from Protective Friction, with design implications for AI tools that counter fixation while preserving agency.

PDF

Open source PDF →Open local PDF →

Trouble viewing inline? Open PDF directly →

Full Text

30,890 characters extracted from source content.

Expand or collapse full text

Drag or Traction: Understanding How Designers Appropriate Friction in AI Ideation Outputs A. Baki Kocaballi University of Technology Sydney, Australia Joseph Kizana University of Technology Sydney, Australia Sharon Stein University of British Columbia, Canada Simon Buckingham Shum University of Technology Sydney, Australia Abstract Seamless AI presents output as a !nished, polished product that users consume rather than shape. This risks design !xation: users anchor on AI suggestions rather than generating their own ideas. We proposeGenerative Friction—intentional disruptions to AI out- put (fragmentation, delay, ambiguity) designed to transform it from !nished product into semi-!nished material, inviting human con- tribution rather than passive acceptance. In a qualitative study with six designers, we identi!ed the di"erent ways in which de- signers appropriated the di"erent types of friction: users mined keywords from broken text, used delays as workspace for indepen- dent thought, and solved metaphors as creative puzzles. However, this transformation was not universal, motivating the concept of Friction Disposition—a user’s propensity to interpret resistance as invitation rather than obstruction. Grounded in tolerance for ambi- guity and pre-existing work#ow orientation, Friction Disposition emerged as a potential moderator: high-disposition users treated friction as “liberating,” while low-disposition users experienced drag. We contribute the concept of Generative Friction as distinct from Protective Friction, with design implications for AI tools that counter !xation while preserving agency. CCS Concepts •Human-centered computing→Interactive systems and tools;Natural language interfaces. Keywords generative AI, friction, ideation, human-AI collaboration, design !xation, appropriation ACM Reference Format: A. Baki Kocaballi, Joseph Kizana, Sharon Stein, and Simon Buckingham Shum. 2026. Drag or Traction: Understanding How Designers Appropriate Friction in AI Ideation Outputs. InProceedings of CHI Conference on Hu- man Factors in Computing Systems Workshop on Tools for Thought (CHI’26 Workshop on Tools for Thought).ACM, New York, NY, USA, 6 pages. 1 Introduction Modern generative AI tools are optimized forseamlessness—delivering polished, #uent responses instantly. While e$cient in some con- texts, this presents AI output as a!nished productto be consumed rather than material to be shaped, risking design !xation [25] and reduced ownership [15]. In design, friction is often treated as drag to be eliminated. But constraints can also be generative: Manning’s concept of enabling CHI’26 Workshop on Tools for Thought, Barcelona, Spain 2026. constraints describes limitations that create conditions for novelty to emerge [20]. Friction in human-AI collaboration can operate similarly—it can be drag (wasted e"ort) or traction (productive resistance that enables agency). We propose the concept ofGenerative Friction: intentional dis- ruptions that degrade the seamlessness of AI output to invite human contribution. By fragmenting text (Physical friction), delaying deliv- ery (Temporal friction), or obscuring meaning (Semantic friction), we aim to break the illusion of completeness. WhileProtective Fric- tionadds barriers to verify accuracy in high-stakes tasks [4]—asking “Should I accept this?”—Generative Friction disrupts presentation to ask “What can Imakefrom this?” This distinction is critical: creative ideation is low-stakes, where the risk is not accepting a wrong answer but!xatingon an idea prematurely. We draw on Dourish’s [13] concept of appropriation, the ac- tive process of adapting technology to one’s own purposes, and Schön’s [23] notion of design materials that “talk back.” Where seamless AI o"ers no resistance to prompt re#ection, generative friction transforms AI output into material that invites appropria- tion. We conducted a qualitative study with six designers using an AI ideation tool under seamless and frictional conditions, asking: How does intentional friction transform the way designers appropriate AI-generated outputs?We found that friction transformed the status of AI output from a !nished product to accept into semi-!nished material requiring further human engagement. Participantsmined keywords from broken text,front-loadedtheir own ideas during delays, andtranslatedmetaphors into features. However, this trans- formation was not universal. We conceptualiseFriction Disposi- tion—a user’s propensity to interpret resistance as invitation rather than obstruction—as a potential moderator. We contribute: (1) the concept of Generative Friction for ideation; (2) empirical evidence of appropriation strategies; and (3) a preliminary model of Friction Disposition. 2 Related Work 2.1 Friction in HCI and AI HCI has moved from treating friction as drag to be minimised [21] to recognising its productive potential through Slow Technology’s reframing of delay as re#ective time [17], Seamful Design’s argu- ment that exposing system seams empowers users [7], and Cox et al.’s operationalisation of “microboundaries” for mindfulness [10]. In the AI domain, this lineage continues with Cabitza et al.’s pro- grammed ine$ciencies that stimulate cognitive engagement [6], and Chen & Schmidt’s [8] behavioral model of Positive Friction. We build on this trajectory, introducing Generative Friction as friction CHI’26 Workshop on Tools for Thought, April 2026, Barcelona, SpainA. Baki Kocaballi, Joseph Kizana, Sharon Stein, and Simon Buckingham Shum designed to degrade AI output seamlessness and stimulate user contribution. 2.2 Design Fixation and Appropriation Seamless AI risks automation complacency [22] and design !xation: users anchor on AI-generated ideas, constraining divergent think- ing. Wadinambiarachchi et al. [25] found that AI during ideation increased !xation and proposed “partially completed or blurred outputs” as countermeasure. Prior work supports this direction: ambiguous, incomplete or misaligned AI outputs can work as gener- ative resources [12], and partial photographs of products can reduce design !xation [9]. Cognitive Forcing Functions [4] interrupt fast, intuitive acceptance (System 1) and trigger deliberate evaluation (System 2) [18]. Seamful XAI [14] extends this to explainability, arguing that revealing AI seams increases agency. However, this work predominantly focuses on high-stakes do- mains such as aviation, medicine, decision-making, where the goal is preventing costly errors. Creative ideation islow-stakes: the risk is not accepting a wrong answer, but!xatingon an idea prema- turely without exploring a larger space of possibilities. Friction in ideation should therefore stimulate elaboration rather than verify accuracy. As introduced earlier, Dourish’s [13] appropriation framework and Schön’s [23] concept of re#ective resistance provide our theo- retical lens. Where seamless AI presents output as !nished—leaving little “back-talk”—our study examines how generative friction in- creases appropriability by inviting users to workwiththe output rather than accept it. 3 Method We conducted a qualitative, within-subjects study using think-aloud protocols to compare unrestricted (seamless) and restricted (friction) AI-assisted ideation tasks. Each session began with a short pre-task interview, followed by the ideation tasks. Low risk ethics approval was received, and sessions lasted 45–60 minutes. 3.1 Participants We recruited 6 participants (3 female, 3 male; ages 22–30) with design backgrounds (Table 1). Three were students studying Master of Design and three were working professionals (Table 2). Table 1: Participant Demographics ID Age/Gender RoleGenAI Use P1 22MPostgrad StudentOccasional P2 23MPostgrad StudentOccasional P3 23FPostgrad StudentRegular P4 24FProduct designer (2 yrs)Regular P5 27FProduct designer (5 yrs)Regular P6 30MUX/UI designer (7 yrs)Regular Because our friction conditions deliberately introduce ambigu- ity into AI output, we anticipated that participants’ tolerance for ambiguity—their capacity to engage productively with uncertain or incomplete information [5], a trait positively associated with cre- ative behaviour [26]—could in#uence their responses. We therefore assessed each participant’s orientation toward AI output and ambi- guity qualitatively, based on pre-task interview statements prior to any exposure to friction conditions. Future work will complement this qualitative coding with a standardised ambiguity tolerance scale (Table 2). Table 2: Participant Predispositions toward AI and Ambiguity ID OrientationRepresentative Quote P4 Low Tolerance“I just don’t see the point of staring at a blank page anymore.” P6 Low Tolerance“I use it when I just really can’t be bothered.” P1 High Tolerance “It shouldn’t do everything for you.” P2 High Tolerance“There’s nothing challenging your thinking [with AI].” P3 High Tolerance“I avoid using it before I’ve sketched something myself.” P5 High Tolerance“Creativity comes from thinking out- side the box.” 3.2 System and Conditions We developedSPARK v1, a custom AI ideation tool (GPT-4o). The underlying model and system prompt were constant; thepresenta- tionof output varied across four conditions: (1)Seamless (Baseline):Complete AI output displayed nor- mally without any intervention. (2)Physical Friction (Fragmentation):AI Output visually broken— every second word of the full idea text is hidden, causing fragmented text blocks. Designed to thedis!uency e"ect[1]. (3)Temporal Friction (Delay):The words of the full AI output are displayed one by one gradually over time. Designed to theincubation e"ect[24]. (4)Semantic Friction (Ambiguity):AI Output was cryptic involving a metaphorical description of the idea. Designed toabstract stimuli[16] andsemantic ambiguity[12] 3.3 Procedure and Analysis The study consisted of pre-task interviews (assessing AI experience and attitudes), four 7-minute ideation tasks responding to design briefs, and post-task interviews. The four briefs were selected to minimise variation while maintaining a similar cognitive task pro- !le: all were student-support product design problems in a shared domain: designing a product, app, or service to help students (1) manage productivity, (2) collaborate as a group, (3) plan for the future, and (4) re#ect on what they had learnt, requiring compara- ble levels of problem framing, constraint satisfaction, and feature ideation. We set the ideation window to 7 minutes, informed by our pilot study, to provide su$cient time for concept generation while maintaining task focus. Conditions were presented in !xed order (Seamless→Physical→Temporal→Semantic). We acknowledge this confounds friction type with order and fatigue e"ects; we return to this limitation in Section 6. We used a process-oriented, re#exive thematic analysis on transcribed data, focusing on “breakdowns” Drag or Traction: Understanding How Designers Appropriate Friction in AI Ideation OutputsCHI’26 Workshop on Tools for Thought, April 2026, Barcelona, Spain (moments where friction halted progress) and “repair strategies” (how users overcame the halt) to surface speci!c appropriation moves [2]. 4 Findings We found that intentional friction did not simply “break” the user’s work#ow; participants actively appropriated the disruption, trans- forming obstacles into creative tools. However, this appropriation appeared to have depended on the user’s underlying orientation toward ambiguity and constraint. 4.1 Strategies of Appropriation Participants developed three primary strategies to appropriate de- graded AI output. Keyword Mining (Physical Friction).When fragmented out- put prevented sentence-level reading, participants shifted tokey- word mining—scanning broken text for actionable nouns and verbs. P3 described this asaccelerated reading:“It was pulling out keywords for me. So I didn’t have to go through the text and high- light”(P3). The broken text became a “tag cloud” for immediate concept extraction. P6 similarly noted:“I’m seeing bits and pieces now. It’s just triggering a di"erent thought.”Notably, P3’s case was distinctive: keyword mining was their pre-existing AI work#ow practice, meaning physical fragmentation was functionally invisible as friction—a point we return to in Section 4.3. Parallel Processing & Front-Loading (Temporal Friction). Under delays, participants !lled the void with independent ideation rather than waiting passively. P5 describedfront-loading:“I went to there [my own mind], came up with a few ideas while I was waiting... and then you can come back and tweak”(P5). This shifted the usual dynamic: participants reported leading with their own ideas before consulting AI output, though we acknowledge the study design cannot fully distinguish deliberate front-loading from idle time- !lling during the enforced wait. P2 similarly noted:“I like to think I had some good ideas in the time that it took.” Abstract Interpretation (Semantic Friction).Semantic fric- tion, the cryptic, metaphorical outputs, forcedinterpretive labor. Unlike physical friction (which obscured form) or temporal friction (which delayed delivery), semantic friction obscuredmeaning itself. For ambiguity-tolerant participants, this became puzzle-solving:“I feel like you’ve got given a di"erent medium, like jelly. But it allowed you to move more #uid within it”(P5). P1 described the experience as a “mini-game,” actively seeking connections between abstract metaphors and concrete design problems:“Thinking of a labyrinth makes me think of like a maze... so maybe a game app.” 4.2 “Friction Disposition” Appropriation and appreciation of friction were not universal; it depended on users’ tolerance for ambiguity—their readiness to engage with uncertain, incomplete, or metaphorical content. High- tolerance users treated friction as invitation; low-tolerance users experienced it as obstruction (Table 3). We propose that Friction Disposition relates to at least two mea- surable factors: (a)tolerance for ambiguity—a well-studied personal- ity construct [5]—and (b)pre-existing work#ow orientationtoward AI output. P3, who already parsed AI text by keywords rather than sentences, experienced physical friction as invisible. P4, who re- lied on copy-paste work#ows optimised for speed, experienced all friction as obstruction. In our data, these pre-existing orientations appeared to align with friction responses at least as strongly as friction type alone. 4.3 Preliminary Dispositional Pro#les Though preliminary, our data with six designers reveals three dis- tinct archetypes of what we term Friction Disposition. The Reframers(P1, P5) interpreted friction as creative con- straint. P1 exhibited playful engagement—treating friction as a puzzle:“I could genuinely feel my head throbbing to think... usually when I’m using AI I’m not thinking.”P5 described semantic friction as liberating:“I could go wherever I wanted, like a constellation, a forest. It didn’t give me the answer, so I had to make it”(P5). The Fluent Appropriator(P3) represents users for whom friction becomesinvisible. As noted in Section 4.1, P3’s baseline keyword-processing practice meant physical fragmentation was not experienced as friction at all. Furthermore, P3 treated semantic metaphors not as riddles but as “role reversal”:“It’s forcing me to have to use my own brain... It’s almost like a role reversal”while producing the highest idea count under this condition Table 4. The Resisters(P4, P6) perceived friction as obstruction. P4:“I don’t like it at all. It’s just annoying. I want it to give me the ideas so I can move on”(P4). P6 exhibited friction fatigue: they adapted to physical and temporal friction early but collapsed under semantic friction:“This is where I would tap out... This hurts my brain to try to interpret.”(We note that semantic friction was always the !nal condition, so cumulative fatigue may have contributed to P6’s collapse; see Section 6.) 4.4 A Dispositional Contrast: P4 vs P3 The P3–P4 contrast most clearly illustrates Friction Disposition as a moderator. Both experienced identical conditions yet diverged dramatically. P3 entered the study with a keyword-processing men- tal model (“I never copy-paste whole ideas”), and this pre-existing orientation meant friction was absorbed into work#ow rather than fought against—physical fragmentation was frictionless, temporal delays became background processing time, and semantic ambigu- ity produced the highest engagement. P4, by contrast, prioritised speed and e$ciency (“e$ciency is a big part of my process”), experi- encing each friction type as obstruction to be endured or rejected. This divergence under identical experimental conditions suggests that friction appropriation isrelational, depending not only the fric- tion design but also on the user’s pre-existing readiness to engage with resistance—their Friction Disposition (see also Table 3). 5 Discussion Our !ndings extend the discourse on friction in human-AI collabo- ration for AI-assisted creative ideation, a low-stakes domain where the cost of a “bad” idea is tolerable compared to a high-stakes do- main with strict requirements of safety and reliability. We see three paradoxes playing out, which merit further investigation: Friction: Drag or Traction?In physics, friction providestrac- tion—the grip that enables controlled motion. Our !ndings reveal the same duality. P1’s disposition converted friction intotraction: CHI’26 Workshop on Tools for Thought, April 2026, Barcelona, SpainA. Baki Kocaballi, Joseph Kizana, Sharon Stein, and Simon Buckingham Shum Table 3: Participant appropriation trajectories and strategies across friction conditions. ID ArchetypePhysical FrictionTemporal FrictionSemantic Friction P1 ReframerBreakdown→Repair:decod- ing/puzzle solving. Breakdown→Repair:front- loading (sketching while waiting). Breakdown→Repair:abstract interpretation (“mini-game”). P2 ReframerBreakdown→Repair:keyword mining/extraction. Breakdown→Repair:anticipa- tory ideation during delay. Endured/Worked-around: mixed uptake; limited interpreta- tion. P3 Fluent AppropriatorAbsorbed:keyword mining as baseline practice. Absorbed:parallel processing; AI queued in background. Breakdown→Repair:abstract interpretation (“role reversal”). P4 ResisterRejected:abandoned interpreta- tion. Rejected:impatient wait- ing/bypassing. Rejected:re-rolled for concrete- ness. P5 ReframerBreakdown→Repair:keyword mining (slower but functional). Breakdown→Repair:front- loading own ideas !rst. Breakdown→Repair:abstract interpretation (“like jelly”). P6 ResisterRejected:incomplete ideas un- helpful. Rejected:delays as failure.Collapse:tapped out under inter- pretive load. productive resistance that enabled agency. P4’s disposition experi- enced onlydrag: wasted e"ort that blocked work#ow. The same design produced opposite valences depending on disposition. The Seamlessness Trap.A central tension emerged: users who mostneed friction (over-trusters like P4) areleastlikely to accept it. P4 copy-pasted AI output verbatim in the seamless condition— demonstrating precisely the over-reliance that friction targets—yet rejected friction outright. Buçinca et al. [4] found e"ective inter- ventions received the worst ratings. Unlike their uniform !nding, we observed variance: P1 called friction “like a game”; P5 found it “liberating.” This suggests the e"ectiveness-acceptance trade-o" depends onwhothe user is. This pattern resonates with psycholog- ical reactance theory: users who perceive friction as threatening their e$ciency, a valued behavioral freedom, resist the intervention most strongly, regardless of its potential bene!t [3]. Friction Fatigue.P6’s collapse in the !nal condition suggests friction tolerance may be a function ofcumulative cognitive load. Sustained interpretive e"ort under friction depletes attentional re- sources, leading to breakdown—a framing grounded in cognitive load theory rather than the contested ego depletion model. Fu- ture designs should consider fatigue-aware adaptation by reducing friction intensity as sessions progress or cognitive load exceeds a threshold. Prior work on friction has focused onwhetherto intro- duce friction, notwho bears the cost. However, friction could be shareable: future designs might distribute the burden through AI self-interpretation (o"ering multiple readings of its own metaphor) or progressive disclosure (cryptic output with an “explain” tog- gle). Table 5 in Appendix shows the improved version of Spark implementing tuneable frictions. Design Implications.Based on our !ndings, we propose four principles for designing Generative Friction: •Mode-selectable:O"er exploration, e$ciency, and validation modes rather than a single intensity dial. As P3 noted, some users already “parse text by keywords,” making Physical friction invisible for some and obstructive for others. •Legible:Explicitly communicate friction’s purpose. Instead of “You may experience incomplete suggestions,” try “This tool shows fragments to help you build ideas, not copy them”— reframing friction from “broken feature” to “designed a"or- dance.” •Burden-shared:Distribute cognitive work between user and AI through mechanisms like progressive disclosure or AI self-interpretation. •Escapable:Preserve agency through opt-out mechanisms, since inescapable friction drives abandonment [19]. Friction should be anudge, not atrap. We note a productive tension between legibility and escapability: making friction’s purpose visible may reduce the need for escape, while easy escape may undermine the generative intent. Future work should explore how these principles interact in practice. We note also that Friction Disposition’s focus on readiness to engage with ambiguity and uncertainty resonates directly with professional and student "Learning Dispositions" [11] that slows learners down to re#ect on deep-seated assumptions. We see a key opportunity to investigate friction design for deeper learning. Our sample (N=6) allowed rich qualitative insight but limits gen- eralizability. Conditions were presented in !xed order (Seamless→ Physical→Temporal→Semantic), so we cannot fully disentangle friction e"ects from order e"ects or fatigue. Future work should counterbalance conditions, develop a validated Friction Disposition scale, and explore adaptive friction systems that respond to user state in real time. 6 Conclusion We investigated whether intentional friction can transform AI out- put from !nished product into semi-!nished material requiring further human engagement. We found that friction enabled di"er- ent appropriation methods including disappropriation. Our !ndings suggest user appropriation is moderated by Friction Disposition. High-disposition users described friction as “liberating” and “like a game”; low-disposition users experienced only drag. We propose two conceptual contributions: the concept of Generative Friction to stimulate human engagement and Friction Disposition as a po- tential moderating factor. Drag or Traction: Understanding How Designers Appropriate Friction in AI Ideation OutputsCHI’26 Workshop on Tools for Thought, April 2026, Barcelona, Spain References [1]Adam L. Alter, Daniel M. Oppenheimer, Nicholas Epley, and Rebecca N. Eyre. 2007. Overcoming Intuition: Metacognitive Di$culty Activates Analytic Reasoning. Journal of Experimental Psychology: General136, 4 (2007), 569–576. doi:10.1037/ 0096-3445.136.4.569 [2]Virginia Braun, Victoria Clarke, and Nicola Rance. 2014. How to Use Thematic Analysis with Interview Data for Process Research. InThe Counselling & Psy- chotherapy Research Handbook. SAGE Publications. [3]Jack W. Brehm. 1966.A Theory of Psychological Reactance. Academic Press. [4]Zana Buçinca, Maja Barbara Malaya, and Krzysztof Z. Gajos. 2021. To Trust or to Think: Cognitive Forcing Functions Can Reduce Overreliance on AI in AI-assisted Decision-making. InProceedings of the ACM on Human-Computer Interaction, Vol. 5. Article 188. doi:10.1145/3449287 [5]Stanley Budner. 1962. Intolerance of Ambiguity as a Personality Variable.Journal of Personality30, 1 (1962), 29–50. doi:10.1111/j.1467-6494.1962.tb02303.x [6]Federico Cabitza, Andrea Campagner, Davide Ciucci, and Andrea Seveso. 2019. Programmed ine$ciencies in DSS-supported human decision making. InInter- national Conference on Modeling Decisions for Arti!cial Intelligence. Springer, 201–212. [7]Matthew Chalmers and Areti Galani. 2004. Seamful Interweaving: Heterogeneity in the Theory and Design of Interactive Systems. InProceedings of the 5th Con- ference on Designing Interactive Systems (DIS ’04). 243–252. doi:10.1145/1013115. 1013149 [8]Zeya Chen and Ruth Schmidt. 2024. Exploring anbsp;Behavioral Model ofnbsp;“Positive Friction” innbsp;Human-AI Interaction. InDesign, User Expe- rience, and Usability: 13th International Conference, DUXU 2024, Held as Part of the 26th HCI International Conference, HCII 2024, Washington, DC, USA, June 29–July 4, 2024, Proceedings, Part I(Washington DC, USA). Springer-Verlag, Berlin, Heidelberg, 3–22. doi:10.1007/978-3-031-61353-1_1 [9]Peiyao Cheng, Ruth Mugge, and Jan P. L. Schoormans. 2014. A New Strategy to Reduce Design Fixation: Presenting Partial Photographs to Designers.Design Studies35, 4 (2014), 374–391. doi:10.1016/j.destud.2014.02.004 [10]Anna L. Cox, Sandy J. J. Gould, Marta E. Cecchinato, Ioanna Iacovides, and Ian Ren- free. 2016. Design Frictions for Mindful Interactions: The Case for Microbound- aries. InProceedings of the 2016 CHI Conference Extended Abstracts on Human Fac- tors in Computing Systems (CHI EA ’16). 1389–1397. doi:10.1145/2851581.2892410 [11]Ruth Deakin Crick and Chris Goldspink. 2014. Learner Dispositions, Self-Theories and Student Engagement.British Journal of Educational Studies62, 1 (2014), 19–35. doi:10.1080/00071005.2014.904038 [12]Peter Dalsgaard. 2025. Creative Ambiguity and Cognitive Tension in Generative AI Tools. InProceedings of the 36th Annual Conference of the European Association of Cognitive Ergonomics (ECCE ’25). Association for Computing Machinery, New York, NY, USA, Article 4, 10 pages. doi:10.1145/3746175.3746183 [13]Paul Dourish. 2003. The Appropriation of Interactive Technologies: Some Lessons from Placeless Documents.Computer Supported Cooperative Work12, 4 (2003), 465–490. doi:10.1023/A:1026149119426 [14]Upol Ehsan, Q. Vera Liao, Samir Passi, Mark O. Riedl, and I Daumé, Hal. 2024. Seamful XAI: Operationalizing Seamful Design in Explainable AI.Proc. ACM Hum.-Comput. Interact.8, CSCW1, Article 119 (April 2024), 29 pages. doi:10.1145/ 3637396 [15]Katy Ilonka Gero, Zahra Ashktorab, Casey Dugan, Qian Pan, James Johnson, Raj Kuber, and Werner Werner. 2022. Sparks: Inspiration for Science Writing using Language Models. InProceedings of the 2022 ACM Designing Interactive Systems Conference (DIS ’22). 1002–1019. doi:10.1145/3532106.3533533 [16]M Gonalves, Carlos Cardoso, and Petra Badke-Schaub. 2012. Find your inspiration: exploring di"erent levels of abstraction in textual stimuli. InDS 73-1 Proceedings of the 2nd International Conference on Design Creativity Volume 1. [17]Lars Hallnäs and Johan Redström. 2001. Slow Technology – Designing for Re#ection.Personal and Ubiquitous Computing5, 3 (2001), 201–212. doi:10.1007/ PL00000019 [18]Daniel Kahneman. 2011.Thinking, Fast and Slow. Farrar, Straus and Giroux. [19]Ulrik Lyngs, Kai Luko", Petr Slovak, Reuben Binns, Adam Slack, Michael Inzlicht, Jelle Van Dijk, and Nigel Shadbolt. 2020. I Just Want to Hack Myself to Not Get Distracted: Evaluating Design Interventions for Self-Control on Facebook. In Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems (CHI ’20). 1–15. doi:10.1145/3313831.3376672 [20]Erin Manning. 2016.The minor gesture. Duke university press. [21]Jakob Nielsen. 1994.Usability Engineering. Morgan Kaufmann. [22]Raja Parasuraman and Victor Riley. 1997. Humans and Automation: Use, Misuse, Disuse, Abuse.Human Factors39, 2 (1997), 230–253. doi:10.1518/ 001872097778543886 [23]Donald A. Schön. 1983.The Re#ective Practitioner: How Professionals Think in Action. Basic Books. [24]Ut Na Sio and Thomas C. Ormerod. 2009. Does Incubation Enhance Problem Solving? A Meta-Analytic Review.Psychological Bulletin135, 1 (2009), 94–120. doi:10.1037/a0014212 [25]Sandun Wadinambiarachchi, Ryan M. Kelly, Saumya Pareek, Qiushi Zhou, and Eduardo Velloso. 2024. The E"ects of Generative AI on Design Fixation and Divergent Thinking. InProceedings of the 2024 CHI Conference on Human Factors in Computing Systems (CHI ’24). doi:10.1145/3613904.3642919 [26]Franck Zenasni, Maud Besançon, and Todd Lubart. 2008. Creativity and Tolerance of Ambiguity: An Empirical Study.The Journal of Creative Behavior42, 1 (2008), 61–73. doi:10.1002/j.2162-6057.2008.tb01080.x A Appendix: Idea Counts across Sessions and SPARK v1 vs SPARK v2 (With Tuneable Friction) While Table 4 provides a summary of the idea counts across all the sessions, Table 5 summarises the key di"erences between SPARK v1 (used in the current study) and the redesigned SPARK v2, which incorporates tuneable friction controls informed by our !ndings. SPARK v2’s design directly implements two of our proposed design principles:Mode-selectable(users choose their friction level) and Escapable(progressive disclosure allows opting out of high friction without abandoning the tool entirely). The tuneable controls also operationalise theBurden-sharedprinciple by distributing inter- pretive e"ort between the system’s default friction and the user’s chosen level of engagement. Table 4: Idea counts per participant across seamless and fric- tion conditions. Participant Seamless Physical Temporal Semantic P16566 P25111 P3971114 P45354 P54355 P65452 CHI’26 Workshop on Tools for Thought, April 2026, Barcelona, SpainA. Baki Kocaballi, Joseph Kizana, Sharon Stein, and Simon Buckingham Shum Table 5: Comparison of SPARK v1 and SPARK v2 friction conditions. SPARK v2 introduces user-controllable mechanisms that allow participants to modulate friction intensity, addressing the Friction Disposition di"erences observed in the study. Friction Type SPARK v1SPARK v2 (Tuneable) Physical Every second words are hidden, creating a fragmented presentation with ellipses between words. Fragmented display is retained, but users can click on the ellipsisto reveal hidden words, progressively reducing friction on demand. Temporal AI output text is displayed gradually over time.Users canplay, pause, and speed upthe pre- sentation of ideas using playback controls, giving them agency over pacing. Semantic Output is presented as cryptic metaphors and riddle-like text. Users must interpret abstract lan- guage to extract design ideas. Metaphorical output is retained, but users can click on the idea boxto reveal a more literal description, providing an “explain” toggle.

Instant research discovery