{"id":"f7c953c0-aca0-408a-854f-a8f0484bcc1a","arxiv_id":"2507.14961","paper_version":1,"verdict":"UNVERDICTED","confidence":"HIGH","novelty_score":3.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"An HCI researcher argues that resisting AI solutionism requires slow, deliberate use and critical thinking rather than outright bans.","lead":"This paper argues that AI hype, especially in human-computer interaction, is amplified by the field's emphasis on novelty, and that resisting AI solutionism should focus on deliberate, critical engagement rather than bans. A reader might care because it reframes the AI ethics debate from prevention to mindful, context-dependent use.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The central prescription depends on an unargued premise that human critical thinking will remain superior to AI; if that capability gap closes, the proposed resistance strategy loses its foundation.","rationale":"The reader's weakest_assumption correctly identifies the same load-bearing concern: the paper assumes AI will not achieve high-level critical thinking, and this assumption underlies the proposal's effectiveness as a form of resistance. I agree with that reading. The concern is genuine, but it does not change the verdict. This is a 5-page workshop position paper presenting a normative argument rather than a falsifiable empirical claim, so the appropriate evaluation is about clarity and plausibility, not predictive accuracy. The UNVERDICTED verdict already reflects that standard scientific verification criteria do not apply in the usual sense. My proposed test would not retroactively falsify the paper's normative stance; it would only test the empirical premise if the paper were treated as a predictive claim. Since it is not, no verdict adjustment is warranted. The paper also includes some independent support for its worries, such as citing Lee et al.'s CHI 2025 finding that higher confidence in generative AI correlates with less critical thinking, which makes the author's concern credible even if the proposed remedy remains untested.","tokens_in":5265,"tokens_out":3895,"duration_ms":49713,"concrete_test":"Construct a longitudinal, pre-registered benchmark of high-level critical-thinking tasks that are unlikely to be represented in LLM training data, such as novel ethical and methodological dilemmas requiring recognition of unstated assumptions, identification of motivated reasoning, and context-sensitive value trade-offs. Have expert human raters and the latest frontier LLMs complete the same held-out tasks annually. If any LLM performs at or above the human expert median on two consecutive annual releases, the premise of a durable human advantage is refuted and the paper's strategy would need re-grounding.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's prescriptive core in Section 2 is conditional on an unargued capability asymmetry. The author writes: 'AI technologies are getting better and better, but I have not yet seen it capable of mastering or even successfully masquerading what I would consider high-level critical thinking skills, and I remain skeptical that it will manage this in the future.' The proposed strategy—deliberation, critical engagement, and non-engagement as resistance—has force only if human critical thinking remains a capacity that AI cannot replicate. If future LLM systems reach or exceed expert-level critical thinking, understood as argument evaluation, hidden-assumption detection, and metacognitive judgment in novel contexts, then the recommended 'human advantage' ceases to be an advantage. Slow(er) research and non-engagement would no longer protect against AI solutionism, because the cheaper automated path could be equally deliberative. The paper supplies no evidence for this durability claim beyond personal skepticism, and it frames the claim as a contested forward-looking belief. This is not an internal inconsistency, but it is a load-bearing empirical premise: if it is false, the central recommendation needs re-grounding in other values, such as accountability or democratic legitimacy, which the paper does not develop.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This 5-page position paper argues that the hype surrounding AI is amplified by HCI's emphasis on novelty, and that resisting 'AI solutionism' via outright bans or abandonment is neither feasible nor desirable. Instead, it advocates slow(er), deliberate use of AI, conscientious critical engagement, and explicit non-engagement, together with redesign of HCI courses and research practices to protect critical thinking and knowledge quality. The paper supports this stance with two worries (students' critical thinking and the entrenchment of poor reporting habits), draws on a small set of recent CHI references and an informal ACM Library search, and closes with concrete personal strategies. The central claim is explicitly hedged as 'may help.'","tokens_in":5590,"tokens_out":7792,"duration_ms":87594,"significance":"For a workshop on resisting AI solutionism, the paper offers a clearly written, constructive middle path between technological solutionism and blanket rejection, and it honestly labels its empirical basis as anecdotal. The explicit hedging, the admission that skepticism about future AI capabilities is a personal belief, and the non-circular use of the author's own prior work [17] are commendable. Its broader value is limited by the absence of systematic evidence for the proposed strategies and by its reliance on one forward-looking factual assumption about human superiority in critical thinking, though the paper's transparency about these limitations makes it a defensible position statement if the requested revisions are made.","major_comments":[{"comment":"Paragraph beginning 'AI technologies are getting better and better': the prescriptive force of the paper rests on an unargued factual premise that AI will not achieve high-level critical thinking skills. The author writes that they have 'not yet seen it capable' and 'remain skeptical that it will manage this in the future,' but gives no evidence or argument for this durability. If future systems reach or exceed expert-level argument evaluation, hidden-assumption detection, and metacognitive judgment in novel contexts, then the proposed strategy (deliberation, critical engagement, and non-engagement as resistance) loses its distinguishing advantage. The paper should either support this premise with empirical or analytic argument, or explicitly conditionalize the recommendation on it and discuss the implications if the premise fails; re-grounding in values such as accountability or democratic legitimacy, which are only mentioned in passing, would be one way to make the recommendation robust.","section":"Section 2"},{"comment":"Paragraphs from 'Pushing back more firmly...' through the numbered strategy list: the move from the two harms (student critical thinking, knowledge/reporting quality) to the proposed strategies ('think before we use or design AI,' use freed time for quality or social good, redesign courses to assess critical thinking) is not accompanied by an account of the mechanism by which these actions will reduce harmful impacts or build a solid knowledge foundation. Because the abstract's 'may help' is a predictive claim, the manuscript should distinguish value-based commitments from evidence-based interventions and, ideally, identify observable indicators or a research agenda that could test the effectiveness of slow research and non-engagement. Without this, the central recommendation remains an untested opinion, which is acceptable for a position paper only if explicitly framed as such.","section":"Section 2"},{"comment":"The paper acknowledges that AI tools can be genuinely beneficial, e.g., for non-native English speakers or dyslexic researchers (Section 1, 'There are good reasons for excitement'), yet later recommends non-engagement as part of the resistance strategy (Section 2, 'Overall, for me, resisting AI solutionism means...'). The manuscript does not explain how non-engagement can be encouraged without inequitably affecting those who rely on AI for accessibility or language support. Please address this tension, for instance by clarifying that non-engagement is a context-specific individual choice and discussing how community advocacy would handle such dependencies.","section":"Section 1 and Section 2"}],"minor_comments":[{"comment":"Figure 1 is presented without a methods note: provide the date of the search, the searched fields (e.g., title, abstract, or full text), and whether results were deduplicated or normalized; currently the comparison is an anecdote rather than a reproducible query.","section":"Figure 1"},{"comment":"The phrase 'AI solutionism' is central to the argument but is not defined; please define it on first use, for example by explicitly engaging with the term as used in the cited workshop [16].","section":"Section 1"},{"comment":"In the enumeration '1) ... 2) ... 3) ...' the formatting and punctuation are inconsistent; use a consistent list style with complete sentences for each item.","section":"Section 2"},{"comment":"Reference [16] lists 'CHI EA ’24' while the paper's context is a CHI ’25 workshop; please verify the correct year and proceedings series.","section":"References"},{"comment":"The informal phrase 'doesn’t need much belabouring' may be better phrased as 'scarcely needs elaboration' to match the register of the rest of the paper.","section":"Section 1"}],"recommendation":"major_revision","confidential_remarks":"This is a workshop-style position paper rather than an empirical study, so the lack of systematic evidence is understandable; however, the unargued capability assumption and the ambiguity about the status of the strategies are load-bearing for the central recommendation and should be addressed before archival publication. The author's transparent hedging suggests the revisions are feasible within the scope of the manuscript."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a five-page workshop position paper, not a research paper. It argues for slow(er) research, deliberation, critical thinking, and non-engagement as a way to resist AI solutionism, instead of bans or abandonment. It doesn't present new data or a new framework, and it doesn't need to—for a workshop position paper, the bar is whether the argument is clear and worth discussing. It is.\n\nWhat it does well: it names a plausible middle path between hype and Luddism, and it gets concrete: think before using AI, convert time saved into meaningful work or social good, and redesign courses to assess critical thinking. The author is honest about relying on anecdote, and the citation set is relevant, including the recent CHI paper on genAI confidence and reduced critical thinking. The little corpus count of \"novel\" vs \"human-computer interaction\" in CHI proceedings is a nice illustrative touch, though it's not a systematic analysis.\n\nThe soft spot the stress test flags is real: the argument leans on the premise that human critical thinking will remain something AI can't replicate. The author states that as a personal bet, not a defended claim. If LLMs reach expert-level argument evaluation and metacognitive judgment in novel contexts, the \"human advantage\" framing loses force. This is not fatal for a position paper, but it's the place where a sharper version would spend more time—either defending the bet or re-grounding the recommendation in accountability and democratic legitimacy rather than capability.\n\nTwo more minor things. The paper's recommendations are plausible but unmeasured; there's no evidence that \"slower research\" or course redesigns move the needle on AI solutionism. And the novelty is modest—similar calls exist, as the citations show. But the paper doesn't overclaim; it says \"may help.\"\n\nWho is this for: HCI researchers who teach, and anyone in a research group feeling institutional pressure to adopt AI. It would be a good reading-group piece.\n\nPeer review: I'd accept it for a workshop or a position-paper track, and for a venue that invites critical discussion. It doesn't have the empirical or theoretical weight for a main-track CHI paper, but it's a serious, honest contribution that should be engaged with rather than desk rejected.","headline":"A clear, honest position paper that makes a modest but sensible case for deliberation and critical thinking as a middle path against AI solutionism; the main soft spot is its reliance on the unargued premise that human critical thinking stays ahead of AI.","tokens_in":5915,"tokens_out":2801,"would_cite":false,"duration_ms":30214,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"To counter AI solutionism, the paper argues for deliberate, slow research and critical engagement—including non-engagement—rather than bans.","keywords":["AI solutionism","critical thinking","deliberation","non-engagement","generative AI hype","human-computer interaction","slow research","novelty"],"falsifier":"A controlled longitudinal study in which students who rely on generative AI from early on score equal to or better on critical-thinking assessments than students who are trained to deliberate without it would undercut the claim that early reliance stymies critical thinking.","tokens_in":5052,"feed_emoji":"🧠","tokens_out":3329,"duration_ms":33545,"temperature":0.7,"pith_summary":"The paper argues that resisting AI solutionism should not mean trying to ban or abandon the technology, which is unlikely to work and probably not beneficial. Instead, the author proposes resistance through slower, more deliberate research, conscientious critical engagement with AI outputs, and the legitimate choice of non-engagement. The reason this matters is that early and unthinking reliance on generative AI may stymie critical thinking in students and amplify existing poor research practices. If true, the path forward is not policy prohibition but a change in individual and community norms about when, where, and why to use AI.","feed_headline":"Don't ban AI—resist it with deliberation","feed_subtitle":"A CHI position paper argues that careful use and non-use of AI, plus critical thinking, beats blanket bans.","key_machinery":"The central mechanism is the 'when, where, and why' questioning heuristic, borrowed from discussions of conference travel, applied to AI use. It turns resistance from a blanket rejection into a case-by-case weighing of trade-offs, and it anchors that weighing in critical thinking as a skill that must be maintained through education and practice.","core_discovery":"The central claim is that human judgement and critical thinking remain a crucial component in human-AI interaction, and that deliberately exercising them is a form of resistance to AI solutionism. The author contends that banning AI is neither feasible nor desirable, and that the better strategy is conscientious engagement: think before designing or using AI, use saved time for social or quality-improving activities, and redesign education to teach and assess critical thinking rather than generating ChatGPT-gradable output. This stance rests on the belief that AI, while improving, has not mastered and may never master high-level critical thinking, so human deliberation retains durable value.","pith_inferences":["The author's stance implies that AI literacy education should focus on critical evaluation of AI output rather than tool proficiency alone.","A testable extension would be comparing research or educational outcomes when non-engagement is explicitly framed as a valid choice.","The argument could also reshape review criteria, for example by requiring authors to state how they deliberated over AI use."],"forward_implications":["Bans and abandonment are not the goal; instead, individuals and communities should ask when, where, and why AI is appropriate before use or design.","Time saved by AI should be redirected toward social or interpersonal good, or toward improving the quality and meaningfulness of work, not just more output.","Courses should be redesigned so learning objectives include and assess critical thinking, for example through exam formats that avoid grading ChatGPT output.","Research communities should push back against incentives for quantity and speed and reward quality and future quality improvements."],"supporting_citations":[{"why":"Defines AI solutionism, the phenomenon the paper positions itself against.","marker":"[16]"},{"why":"Documents the economic, representational, misinformation, and other harms of LLMs that motivate the need for resistance.","marker":"[15]"},{"why":"Provides evidence that higher confidence in generative AI correlates with less critical thinking, supporting the education worry.","marker":"[8]"},{"why":"Supplies the principle of thinking before designing or deploying technology of any kind.","marker":"[1]"},{"why":"Offers the 'when, where, and why' heuristic for asking about conferencing that the paper adapts to AI use.","marker":"[6]"},{"why":"Argues that LLM-generated content may amplify poor reporting habits, supporting the knowledge-quality risk.","marker":"[17]"},{"why":"Backs the call to push back against quantity and speed in research output.","marker":"[11]"}],"fun_headline_variants":["Deliberate resistance beats AI bans","Critical thinking is your AI resistance","Slow down, think critically, resist AI hype","Deliberation and non-use counter AI solutionism"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument relies on the assumption that AI systems will not achieve high-level critical thinking skills, so that human deliberation remains a durable advantage and an effective form of resistance.","fun_headline_variants_meta":{"raw":{"variants":["Deliberate resistance beats AI bans","Critical thinking is your AI resistance","Slow down, think critically, resist AI hype","Deliberation and non-use counter AI solutionism"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000147,"raw_usage":{"total_tokens":1070,"prompt_tokens":713,"completion_tokens":357,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":329,"completion_tokens_details":{"reasoning_tokens":314}},"tokens_in":329,"tokens_out":357,"duration_ms":4122,"temperature":1.0,"reasoning_tokens":314,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T15:42:49.068462+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A controlled longitudinal study in which students who rely on generative AI from early on score equal to or better on critical-thinking assessments than students who are trained to deliberate without it would undercut the claim that early reliance stymies critical thinking.","supporting_citations":[{"cited_title":"Williams, Karin Hansson, and Ivana Feldfeber","cited_arxiv_id":null,"evidence_quote":"Defines AI solutionism, the phenomenon the paper positions itself against."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Offers the 'when, where, and why' heuristic for asking about conferencing that the paper adapts to AI use."},{"cited_title":"The Shiny Scary Future of Automated Research Synthesis in HCI","cited_arxiv_id":"2501.16084","evidence_quote":"Argues that LLM-generated content may amplify poor reporting habits, supporting the knowledge-quality risk."}],"review_version":1}