{"id":"5cfcbdca-adc3-46d7-8898-c967672afdd9","arxiv_id":"2501.16518","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A qualitative study identifies six playful and five concerning qualities of generative-AI urban play, based on workshops with 14 designers and 14 citizens using an image-to-image tool.","lead":"This paper built a phone tool that uses generative AI to turn photos of city streets into playful new scenes, then had 14 designers and 14 citizens try it and talk about what is fun and what is worrying. It names six playful features and five citizen concerns, and suggests a 'tourist metaphor' for how AI interacts with cities.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The playful feature–concern pairings in Figure 2 may be partly produced by the confirmatory thematic loop in §4.4.1 and §6.4, not discovered from the data.","rationale":"The reader's weakest assumption was that iWonder, a single image-to-image tool, may not represent GAI as a class. That is a legitimate scope concern, but it is not the most load-bearing issue for the central claim. The paper's central contribution is the tension mapping between playful features and citizen concerns, culminating in the tourist metaphor. The analysis sections explicitly describe an iterative procedure in which themes were revised after the citizen workshop precisely so that playful features would carry associations with risks and concerns. This means the six-feature/five-concern structure and the pairwise associations in Figure 2 are partly an output of the method, not an independent finding from the data. That is a direct threat to internal validity, whereas the single-tool issue only threatens external scope. I therefore partially agree with the reader: the tool-representativeness concern is real, but the confirmatory thematic loop is more fundamental. I do not call for a different verdict because the paper is transparent about the procedure, provides rich quotes, and frames the analysis as reflexive thematic analysis, which is interpretive by design. These features make the study a plausible conditional contribution but do not eliminate the need for an independent audit of the mapping. The proposed blinded re-analysis would settle whether the tension narrative is data-grounded. Until then, CONDITIONAL remains appropriate; no verdict change is warranted.","tokens_in":28164,"tokens_out":5327,"duration_ms":55605,"concrete_test":"Commission an independent, blinded re-analysis of the raw interview transcripts, photo logs, and generated images from §4 and §6 using the same reflexive thematic analysis protocol but without the instruction to 'keep tensions alive' or to link codes to concerns. Compare the emergent themes and any feature–concern pairings to Figure 2. If the blind analysis does not reproduce the six-feature, five-concern structure and the pairwise links, the central tension mapping is not robust to the confirmatory refinement step.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that six Playful Features, five Citizen Concerns, and their paired relationships in Figure 2 reveal a genuine dynamic interplay between users, GAI, and urban contexts. That claim requires the categories and their pairwise links to be grounded in participant experiences. The analysis procedure, however, includes a confirmatory step that selects for exactly those links. In §4.4.1, after the citizen workshop the authors write that tentative themes 'lacked connections to citizen concerns' and that this 'prompted us to emphasize more on keeping the tensions alive within the themes, retaining underlying associations of playful features with risks.' In §6.4, the citizen-analysis step similarly 'led us to revisit the codes representing positive and negative citizen comments and examine how they were (not) connected to specific Playful Features, followed by theme reorganization and refinement.' A procedure that reorganizes themes until connections appear cannot, by itself, establish that the connections are empirical. The same pressure applies to the 'offensive qualities' (dissonant agency, insensitive erosion, creative vandalism) and the tourist metaphor, which depend on the existence of these pairings. A second, milder threat compounds this: the six features are not independent—F1–F3 are all framed as 'empowerment of visualization'—so the later pairing of each to distinct concerns may reflect thematic carving rather than separable phenomena. The reader's single-tool concern is valid but secondary; even if iWonder perfectly represented GAI, the mapping could still be an analytic construction. The paper is transparent about the loop, which helps, but transparency does not remove the threat to the central claim.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports a two-workshop qualitative study of iWonder, a mobile image-to-image GAI tool for urban play. Fourteen designers explored city spaces with the tool, identified six Playful Features, and produced six urban-play design ideas; fourteen citizens then evaluated the ideas and raised five Citizen Concerns. The authors present a tension map (Figure 2) linking features to concerns, distill three playful qualities (generative agency, meaningful unpredictability, social performativity) and three offensive qualities (dissonant agency, insensitive erosion, creative vandalism), and propose the metaphor of 'GAI as a playful yet offensive tourist.' The central claim is that these features and concerns reveal a dynamic interplay between users, GAI, and urban contexts, offering design considerations for GAI-enabled urban play.","tokens_in":28402,"tokens_out":4387,"duration_ms":44703,"significance":"If the findings hold, this is a useful empirical contribution to HCI and urban-play research. The study is methodologically transparent in several respects: it reports participant recruitment and demographics, gives a positionality statement, includes extensive participant quotes and image examples, and explicitly seeks both positive and critical citizen perspectives. The two-workshop structure, with designers generating ideas and citizens evaluating them, is appropriate for surfacing tensions rather than only celebrations of GAI. The proposed 'tourist metaphor' is a thought-provoking interpretive synthesis that could help designers and city stakeholders think about the dual nature of GAI. The paper also documents the tool implementation and study procedures in enough detail to be replicable. The main risk is that the central pairing of features and concerns rests on an analysis step that selected for exactly those pairings, so the empirical grounding of Figure 2 needs to be demonstrated more directly.","major_comments":[{"comment":"The analysis procedure contains a confirmatory loop that materially weakens the evidential status of Figure 2. The authors state in §4.4.1 that after the citizen workshop tentative themes 'lacked connections to citizen concerns' and this 'prompted us to emphasize more on keeping the tensions alive within the themes, retaining underlying associations of playful features with risks.' Section 6.4 similarly reports that the authors 'realized that the candidate themes lacked links to the Playful Features' and therefore 'revisit[ed] the codes representing positive and negative citizen comments and examine how they were (not) connected to specific Playful Features, followed by theme reorganization and refinement.' A procedure that reorganizes themes until connections appear can illustrate tensions, but it cannot by itself establish that those pairings are grounded in participant experiences. Because the playful qualities, the offensive qualities, and the tourist metaphor are all built on these pairings, the paper should either report the pre-refinement themes and the specific code-to-feature links that survived the refinement, or validate the pairings through independent coding or member checking.","section":"§4.4.1 and §6.4"},{"comment":"The central claims are phrased about 'GAI' in the abstract and throughout the discussion, but all empirical material comes from a single image-to-image tool implemented with the Stability AI REST API. Several features, especially 'whimsical urban montage' (F5), depend on the specific failure modes, randomness, and training data of Stable Diffusion and may not extend to LLM-based or text-to-image GAI. To support the general claim, the authors should either reframe the findings as being about image-to-image GAI tools, or add comparative observations with at least one other GAI modality and explicitly state how the playful features and concerns vary across modalities. As written, the scope of the generalization is overstated.","section":"§3.1 and §5.1"},{"comment":"The first three playful features all share the title stem 'Empowerment of Visualization,' and the distinctions among them—sharing desired neighborhoods, re-discovering people-place relationships, and shuttling between multiverse and reality—appear to be analytical subdivisions of one capability rather than independent features. This becomes load-bearing in Figure 2, where these three features are paired with different citizen concerns; if the features are not empirically separable, those pairings may be artifacts of thematic boundary-drawing rather than genuine relationships in the data. The authors should show the participant quotes and codes that distinguish F1 from F2 and F3, or collapse the features and adjust Figure 2 accordingly.","section":"§5.1, F1–F3"}],"minor_comments":[{"comment":"Participant identifiers are inconsistent: §5.1.5 uses 'P3, P14' and §5.1.6 uses 'P9' instead of the DP-prefixed identifiers used elsewhere; please standardize.","section":"§5.1.5 and §5.1.6"},{"comment":"There is a typo in the phrase 'experiment wtih' which should read 'experiment with.'","section":"§7.1"},{"comment":"The identifier 'C10' should be 'CP10' for consistency with the other citizen-participant labels.","section":"§6.5.2"},{"comment":"The participant quotation contains 'Walk to Pain,' which is presumably a typo for 'Walk to Paint'; please verify against the transcript or indicate if this is an intentional participant wording.","section":"§6.5.2"}],"recommendation":"major_revision","confidential_remarks":"The paper is within scope for CHI and the qualitative methods are broadly appropriate. The main risk is not the choice of method but the confirmatory theme-refinement step, which selects for the exact feature–concern pairings presented as findings. If the authors can document the pre-refinement themes, show the stability of the pairings, or validate them with an independent coder, I would be comfortable supporting acceptance. The tourist metaphor is interpretive and should be clearly labeled as such, but that is not a blocking issue."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First thing to know: this is a competent, transparent qualitative HCI paper, and the central claim is plausible but softer than the framing suggests. The six playful features and five citizen concerns are a useful map for anyone designing GAI urban play, and the tourist metaphor is a nice lens. The paper is not a rigorous empirical test of those categories; it's a design exploration with reflexive thematic analysis, and the authors mostly say so.\n\nWhat's new: they built iWonder, a mobile image-to-image tool, and ran two workshops with designers and citizens. That combination—situated designer exploration plus bottom-up citizen evaluation—isn't in the cited literature. The quotes and images ground the features and concerns. The analysis is honestly reported: open coding, affinity diagrams, positionality, iterative theme refinement. The 'playful yet offensive tourist' metaphor is generative and captures the tension between generative agency and insensitive erosion.\n\nSoft spots: two. First, the confirmatory loop. In §4.4.1 and §6.4 the authors write that after the citizen workshop they reorganized themes because tentative themes 'lacked connections to citizen concerns' and needed to 'keep the tensions alive.' That means the feature–concern pairings in Figure 2 are partly products of the analytic procedure, not discoveries from the data. The authors are transparent about this, and it's a normal part of reflexive thematic analysis, but it weakens the claim that the pairings reflect a genuine empirical dynamic. Second, the generalization from one tool (Stability AI image-to-image API) to 'GAI' as a class. Some features like whimsical montage may be specific to image-to-image models; text-based GAI might behave differently. The citizen sample is small and convenience-based, but for qualitative work that's acceptable.\n\nThe paper would benefit from framing the map as design considerations rather than findings, and from a limitation section that states outright that the pairings were refined through iteration. Neither issue is fatal. The evidence is what it is: 14+14 participants, rich quotes, no code or data released, but the methods section is detailed enough to trust the process.\n\nBottom line: worth a serious referee, and likely a solid CHI-level contribution to the urban play and GAI ethics community. I'd bring it to reading group and would cite it for the feature-concern map and the tourist metaphor.","headline":"A transparent qualitative study whose playful feature–concern map is useful but partly shaped by the authors' confirmatory theme refinement; worth engaging, not overclaimed.","tokens_in":28976,"tokens_out":1751,"would_cite":true,"duration_ms":18721,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Generative AI can make cities more playful, but the same features that delight can also offend, and this paper maps that tension through two workshops with a situated image-to-image tool.","keywords":["generative AI","urban play","playable cities","human-AI interaction","image-to-image generation","citizen concerns","tourist metaphor","qualitative user study"],"falsifier":"Run the same two-workshop protocol with a different GAI modality, such as text-to-image or video generation, in the same urban settings; if designers and citizens report a substantially different set of playful features and concerns, the paper's claim that these are GAI-level qualities is falsified. A second check is whether the five concerns still arise when prompts are given in the participant's first language with spelling or voice support.","tokens_in":27950,"feed_emoji":"🏙️","tokens_out":4978,"duration_ms":44320,"temperature":0.7,"pith_summary":"This paper argues that generative AI (GAI) can enrich urban life with playful experiences, but only if designers face the fact that the same features that delight can also offend. The authors built a mobile image-to-image tool called iWonder and ran two workshops: fourteen designers wandered through streets in Taiwan and the Netherlands to surface GAI's playful features and generate design ideas, then fourteen citizens critiqued and prototyped those ideas. The result is a map of six playful features, five citizen concerns, and three paired playful and offensive qualities. The paper's central proposal is the tourist metaphor: GAI behaves like an experienced but careless visitor who brings fresh perspectives while remaining unaware of local nuances, so cities should design with both sides in mind. If this is right, designers gain a structured vocabulary for making public AI interventions playful without eroding safety, cultural meaning, or trust.","feed_headline":"Generative AI is a playful yet offensive tourist, study finds","feed_subtitle":"Two workshops with designers and citizens yield six playful features, five concerns, and design guidance for urban play.","key_machinery":"The carrying mechanism is a paired mapping: each playful quality is tied to an offensive counterpart. Generative agency, meaningful unpredictability, and social performativity are paired with dissonant agency, insensitive erosion, and creative vandalism; the tourist metaphor names the pair as a single character, a traveler who brings curiosity and fresh perspective while lacking local attunement. The instrument that produces the evidence is iWonder, a mobile image-to-image GAI tool with Reimage and Inpaint functions, backed by a commercially hosted generation API, that lets people photograph, prompt, and regenerate urban scenes in situ.","core_discovery":"The central claim is that GAI-enabled urban play is defined by a tension: the same generative capacities that produce agency, meaningful unpredictability, and social performativity also produce dissonant agency, insensitive erosion, and creative vandalism. Concretely, the paper identifies six playful features: empowerment of visualization to share desired neighborhoods, empowerment to re-discover people-place relationships, empowerment to shuttle between multiverse and reality, encounter of defamiliarized places, whimsical urban montage, and pass-down co-improvisation. It also identifies five citizen concerns: fatigue and marginalization from misaligned content creation, navigational unsafety from scene rendering and defamiliarization, tensions between preserving and transforming socio-cultural significance, dissemination of disturbing modifications, and general bias, misinformation, and privacy issues. The argument is grounded in the situated experience of designers and citizens using iWonder in real urban contexts, and it culminates in the tourist metaphor as a design lens.","pith_inferences":["If the six playful features are truly GAI-level rather than iWonder-specific, the same two-workshop protocol should reproduce them with text-to-image, video, or multimodal GAI tools; that is a testable extension the paper does not run.","The tourist metaphor may transfer beyond cities to any domain where an AI system offers outsider novelty with shallow local understanding, such as museum interpretation, community co-design, or organizational creativity tools.","The meaningful unpredictability finding suggests a design principle: build interfaces that let users hand control to the model while still giving citizens an easy stop or opt-out, an untested balance.","Prompt-language barriers point to an accessible fix: pairing an image generator with a large language model that refines imperfect prompts could reduce the fatigue and marginalization concern documented in the citizen workshop."],"forward_implications":["Designers can use the six playful features as a building-block checklist and the five citizen concerns as a pre-deployment risk screen for GAI-enabled urban play.","Design ideas such as City Renovation Solitaire and Mirrorverse Switch become concrete prototypes for testing civic participation, storytelling, and cultural exchange.","The tourist metaphor gives urban stakeholders a common language for discussing GAI's double edge before committing to public deployments.","Citizen-suggested safeguards, including face blurring, immediate photo deletion, opt-out mechanisms, clear AI-generated labels, and prompt-language switching, become concrete design requirements.","Field-testing and multi-stakeholder workshops are recommended before deployment, since the paper finds that contextual tensions are difficult to capture in indoor settings."],"supporting_citations":[{"why":"Supplies the urban play qualities and the bottom-up, multi-stakeholder framework that structure the designer and citizen workshops.","marker":"[2]"},{"why":"Provides the reflexive thematic analysis method used to identify the six playful features and five citizen concerns.","marker":"[13]"},{"why":"Gives the worked example and criteria for reflexive thematic analysis that the authors apply in their final theme iterations.","marker":"[15]"},{"why":"Establishes GAI's design opportunities and risks in cultural-heritage settings, which the paper transfers to urban play contexts.","marker":"[27]"},{"why":"Prior study of city wandering with image-to-image AI whose playful perspective and situated findings iWonder directly builds on.","marker":"[33]"},{"why":"The Google Pixel Reimagine example that motivates the tension between playful image reimagination and civic disturbance.","marker":"[37]"},{"why":"Supplies principles for co-creative AI in public spaces and frames how GAI can act as a design material for urban play.","marker":"[48]"},{"why":"Defines urban play and playfulness, the conceptual basis for the study's focus on playable cities.","marker":"[56]"}],"fun_headline_variants":["GAI: playful yet offensive tourist in urban play","Generative AI's dual role: playful and offensive in city design","Tourist metaphor: GAI's playful-offensive tension in urban spaces","iWonder: GAI as playful yet offensive urban tourist","Urban play: GAI's playful features and citizen concerns"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that one image-to-image tool, iWonder, stands for generative AI as a class in urban spaces: if its playful features and citizen concerns are specific to this tool rather than to GAI generally, the generalized findings and the tourist metaphor lose their scope.","fun_headline_variants_meta":{"raw":{"variants":["GAI: playful yet offensive tourist in urban play","Generative AI's dual role: playful and offensive in city design","Tourist metaphor: GAI's playful-offensive tension in urban spaces","iWonder: GAI as playful yet offensive urban tourist","Urban play: GAI's playful features and citizen concerns"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000582,"raw_usage":{"total_tokens":2719,"prompt_tokens":905,"completion_tokens":1814,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":521,"completion_tokens_details":{"reasoning_tokens":1729}},"tokens_in":521,"tokens_out":1814,"duration_ms":13201,"temperature":1.0,"reasoning_tokens":1729,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T12:39:42.512172+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same two-workshop protocol with a different GAI modality, such as text-to-image or video generation, in the same urban settings; if designers and citizens report a substantially different set of playful features and concerns, the paper's claim that these are GAI-level qualities is falsified. A second check is whether the five concerns still arise when prompts are given in the participant's first language with spelling or voice support.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Prior study of city wandering with image-to-image AI whose playful perspective and situated findings iWonder directly builds on."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"The Google Pixel Reimagine example that motivates the tension between playful image reimagination and civic disturbance."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies principles for co-creative AI in public spaces and frames how GAI can act as a design material for urban play."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines urban play and playfulness, the conceptual basis for the study's focus on playable cities."}],"review_version":1}