{"id":"315faace-8dec-41cf-8447-d5ebd0d3b3f7","arxiv_id":"2412.00330","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":3.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":8,"one_line_summary":"Seven of the nine stated hypotheses were supported, with attitude toward ethical behavior predicting AI use and social influence showing no significant effect in a PLS-SEM analysis of 237 survey responses.","lead":"This paper surveys 237 mostly IT professionals and uses structural equation modeling to test what drives people to adopt artificial intelligence, adding an ethics attitude factor to classic technology acceptance models. The authors report that ethical attitudes predict AI use, while peer influence shows no measurable effect, offering a small evidence point for ethics-aware AI adoption strategies.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"H5's ethics-to-use path is susceptible to common method bias and construct overlap; the missing measurement instrument and data prevent testing this, so the central claim is unverified.","rationale":"The paper's headline claim depends entirely on the existence of a positive, direct relationship between AFCE and Use. The most likely threat to that relationship is not sample representativeness or temporal precedence, but method-induced correlation: all variables come from the same self-report survey, and the ethical-attitude items are especially prone to social desirability and to wording that may overlap with use intention. Because the measurement instrument and item-level loadings are missing, the reader cannot even evaluate whether discriminant validity was properly established at the item level. The reader's weakest_assumption about causality from cross-sectional self-reported data is a broader statement of the same concern; our attack narrows it to the most probable spurious mechanism. The proposed CLF test would decisively show whether the H5 path survives when a shared method factor is accounted for. If the test fails, the central claim collapses; if it passes, the other reported issues (VIF contradictions, hypothesis count, overclaimed conclusions) are still problematic but less fatal to the main empirical result. Therefore, we maintain the reader's CONDITIONAL verdict, with the explicit condition that the instrument and data be released and a common-method-bias analysis be performed. This is consistent with the reader's request for publication of the measurement instrument and data, and it does not require a change in the verdict.","tokens_in":12924,"tokens_out":11126,"duration_ms":104669,"concrete_test":"Require the authors to release the full questionnaire and raw item-level data as a supplement, and conduct a common latent factor (CLF) analysis: add a single unmeasured method factor loading on all indicators and re-estimate the structural path AFCE -> Use. If the path becomes non-significant (p > 0.05) or its standardized coefficient drops below 0.10 when the CLF is controlled for, the H5 result is not robust to common method bias and the paper's central conclusion is unsupported. As a complementary check, run an independent content-validity review of the AFCE items to ensure none of them reference AI use intention or outcomes directly.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central empirical claim is that Attitude Towards Ethical Behavior (AFCE) directly influences self-reported AI Use (H5, beta = 0.207, p < 0.05, Table V), which is then used to conclude that AI can be adopted with AFCE as the 'mainstay'. All constructs are measured in one online questionnaire using 1-7 Likert scales, creating a strong common method variance threat. The authors do not report any common-method-bias diagnostic (e.g., Harman's single-factor test, common latent factor, or marker variable). Moreover, the AFCE construct is adapted from Internet ethics items [17] to AI with no content validation, and the measurement instrument, item wordings, cross-loadings, and AVE tables referenced in Annexes 2 and 4 are absent from the manuscript. If AFCE items inadvertently capture positive affect, social desirability, or a conditional-use intention (e.g., 'I would only use AI if it is ethical'), the significant path to Use may be inflated or tautological. Because the dataset and instrument are not shared, this artifact cannot be ruled out, directly undermining the headline conclusion rather than merely its generalizability.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a model of AI adoption that joins UTAUT, DeLone–McLean, and Theory of Planned Behavior constructs, adding a new dimension, Attitude Towards Ethical Behavior (AFCE), adapted from Internet ethics to AI. The model is tested with PLS-SEM on 237 online self-report responses, mainly from IT professionals. The authors report that seven of the (variously stated) eight or nine hypotheses are supported, with H5 (AFCE directly influences Use) being central to the conclusion that ethics can serve as the 'mainstay' of AI adoption. Social Influence hypotheses H3a and H3b are not supported. The paper concludes that ethical AI adoption is possible and that ethics impacts people's decisions.","tokens_in":13210,"tokens_out":3501,"duration_ms":33552,"significance":"If the empirical claims were fully supported, the main contribution would be a modest extension of technology-acceptance models by inserting an ethics attitude construct, with a concrete estimate that ethics attitudes positively predict self-reported AI use. The paper has some strengths: it uses standard PLS-SEM validation procedures (Cronbach's alpha, composite reliability, AVE, Fornell–Larcker), retains nonsignificant hypotheses as nonsignificant, and grounds the model in established theories with explicit citations. However, the central claim rests on a single cross-sectional, self-report survey, and the manuscript omits the measurement instrument, the common method bias diagnostics, and the collinearity table that would be required to substantiate the path estimates. The significance is therefore conditional on additional evidence that is not currently in the paper.","major_comments":[{"comment":"The collinearity assessment is contradicted by the reported numbers. The text states that 'All Inner VIF values vary between 0.768 and 0.921,' but Table II lists values such as 1.000, 1.284, 1.429, 1.369, 1.320, 1.185, and 1.482. The stated range matches the diagonal of the Fornell–Larcker table (Table III), not the Inner VIF values. As a result, the paper does not actually report the Inner VIF results, and the claim that there are no collinearity problems is unsupported. This matters because collinearity is a precondition for interpreting the individual path coefficients, including the central H5 estimate.","section":"Section V, Table II and accompanying text"},{"comment":"No common method bias diagnostic is reported, yet all constructs were measured with the same 1–7 Likert-scale online questionnaire. The path from AFCE to Use (beta = 0.207, p reported as 0) is the key evidence for the conclusion that ethics 'impacts' AI adoption. Without a test such as Harman's single-factor, a common latent factor, or a marker variable, the observed correlation could be inflated by response style, social desirability, or item wording. The absence of this test is a load-bearing gap for the central claim, not a mere reporting detail.","section":"Section V, H5 and Section VI"},{"comment":"The measurement instrument for AFCE is not available. The paper cites reference [17] for Internet ethics and states that AFCE is an adaptation to AI, but the actual item wordings, cross-loadings, and AVE tables are referenced as 'Annex 2' and 'Annex 4' that are not included in the manuscript. This omission prevents the reader from checking whether the AFCE items inadvertently capture conditional use intention (for example, willingness to use AI only if it is ethical) or positive affect, either of which would make the significant H5 path tautological. It also prevents verification of the claimed content validity of the adaptation. This is directly relevant to the paper's headline result.","section":"Section III, AFCE construct and Annexes 2 and 4"},{"comment":"The paper uses causal language ('directly affects', 'positively impacts', 'Ethics affects people's decisions') while the data are a single cross-sectional, self-report survey from a convenience sample of 237 respondents, mainly from the Information Technology area. Temporal precedence is not established, and the sample is not representative of the 'society' to which the conclusion generalizes. At minimum, the discussion should reframe the findings as associations and acknowledge the limitation that causality cannot be inferred from these data; otherwise the central conclusion is overreaching.","section":"Section V and Section VI, causal language"},{"comment":"The paper states repeatedly that 'seven of the nine investigation hypotheses' were validated, but only eight hypotheses are actually listed (H1, H2a, H2b, H3a, H3b, H4, H5, H6). Section VI even says 'seven research hypotheses' while describing the model. This inconsistency affects the headline quantitative claim and indicates that the hypothesis list or the summary count is in error. The authors should correct the count and ensure all hypotheses are enumerated consistently.","section":"Abstract and Section V, hypothesis count"}],"minor_comments":[{"comment":"The p-values for supported hypotheses are reported as '0'; considering the typical reporting conventions, the authors should report exact values (e.g., p < 0.001) rather than 0, since p is never exactly zero.","section":"Section V, Table V"},{"comment":"Hypothesis H6 is stated as 'There is a correlation between the Use of Artificial Intelligence mechanisms and the Net Benefits dimension,' which is a correlational statement, while H1–H5 use 'directly affects' or 'positively impacts.' The model results are then interpreted causally in the conclusion; aligning the hypothesis wording with the analysis would clarify the intended claim.","section":"Section III, H6 wording"},{"comment":"The method section does not describe how the questionnaire was distributed, the sampling procedure, the response rate, or how missing data were handled. Adding these details would improve reproducibility and help readers assess sample representativeness.","section":"Section IV, method description"},{"comment":"There are several typographical and formatting errors, for example 'recen t years' in the abstract, 'Hipothesys' in Table V, mixed use of 'Behavioral Intent' and 'Behavioral Intention', and inconsistent use of commas and periods as decimal separators (e.g., Table II uses '1,284' while the text uses '0.768'). A careful proofreading pass is needed.","section":"Throughout"},{"comment":"Several references are to the authors' own prior work (e.g., [1], [2], [12], [13], [20], [22], [23]). While self-citation is not inappropriate, the literature review could benefit from a broader set of sources on AI ethics and technology adoption to ensure the framing is not overly reliant on the authors' own papers.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The manuscript reads as an incomplete conference-paper version: the annexes with the measurement model (cross-loadings, AVE, item wordings) are referenced but absent, and the inconsistent hypothesis counts and Inner VIF/text mismatch suggest the text has not been carefully checked. The central claim about the ethics-to-use path cannot be verified from the materials provided. I would encourage the editor to require the authors to supply the full measurement instrument, the missing tables, and a common-method-bias analysis before any further consideration, and to ask them to substantially temper the causal and societal-generalization language. If the authors cannot provide these materials, the paper would not be publishable in its current form."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a standard UTAUT/Delone-McLean hybrid with an ethics-attitude construct added, tested on 237 self-reports. The headline claim—that attitude toward ethical behavior predicts AI use (H5, beta = 0.207)—is plausible but not verified. The measurement instrument and data are not in the manuscript, there is no common-method-bias test, and a few internal inconsistencies (nine vs eight hypotheses; VIF text vs Table II) need cleaning up. I would send it to review, but I would not accept it as is.\n\nWhat's actually new: the paper recombines established models into a specific path from ethical attitude (AFCE) to use, and it collects a fresh dataset. It does the usual PLS-SEM checks—Cronbach's alpha, composite reliability, AVE, Fornell-Larcker—and reports nonsignificant hypotheses as nonsignificant rather than burying them. That is honest work as far as it goes.\n\nThe soft spots are real but not fatal to the idea. The biggest one is that all constructs come from a single online questionnaire on 1–7 Likert scales, so common method variance could inflate the H5 path. Without the item wording and data, you cannot test whether AFCE is measuring something distinct from general positivity or a conditional use intention. The paper should have run Harman's single-factor test or used a marker variable. Second, the causal language in the conclusion ('proven') outruns the cross-sectional design; this can only support associative claims. Third, the instrument, cross-loadings, and AVE tables are referenced as Annexes but not present, so the validity evidence cannot be audited. The nine-versus-eight hypothesis count and the Inner VIF contradiction (text says 0.768–0.921; Table II shows 1.284+) are small but embarrassing.\n\nThe citation pattern is mostly standard, with several self-citations; that is not itself a flaw, but it does not add evidence for the new path.\n\nFor a reader, this is a workable starting point for a survey-based AI adoption study. I would not cite it in my own work this year, because the central result is not yet falsifiable from the manuscript. But the research question is relevant and the model is testable, so it deserves a serious referee. My recommendation: send to peer review with a strong request for the measurement instrument, the data (or a public repository), a common-method-bias diagnostic, and careful revision of the causal wording. If the authors can show the H5 path survives those checks, it becomes a modest but useful empirical contribution.","headline":"A plausible but unverified ethics-to-use path in a small PLS-SEM study; deserves review but needs data, instrument, and a common-method-bias check.","tokens_in":13745,"tokens_out":2877,"would_cite":false,"duration_ms":26117,"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":"The paper claims that an attitude toward ethical behavior is a significant predictor of AI use and that seven of the nine hypotheses in its ethics-anchored adoption model were validated.","keywords":["artificial intelligence","ethics","technology adoption","behavioral intention","attitude toward ethical behavior","structural equation modeling","net benefits","UTAUT"],"falsifier":"Collect objective AI-usage logs from the same 237 respondents and re-estimate the model with logged use replacing self-reported use; if the AFCE-to-Use path (H5) is no longer significant after controlling for behavioral intention and facilitating conditions, the paper's central claim fails.","tokens_in":12690,"feed_emoji":"⚖️","tokens_out":10681,"duration_ms":91910,"temperature":0.7,"pith_summary":"The paper sets out to show that AI can be adopted in society without setting ethics aside, and that a person's attitude toward ethical behavior is one measurable driver of whether they actually use AI. Using partial least squares structural equation modeling on 237 questionnaire responses, mostly from IT professionals, the authors report that seven of their nine hypotheses held: performance expectations, facilitating conditions, and behavioral intention behave as technology-acceptance theories predict, and the Attitude Towards Ethical Behavior construct has a significant positive path to self-reported AI use (H5, $\\beta = 0.207$, $p < 0.05$). The two rejected paths both involve social influence. If the result is right, ethics functions as a support for adoption rather than a brake on it, which gives organizations and regulators a concrete lever: cultivating ethical attitudes may increase AI uptake.","feed_headline":"Ethical attitudes predict AI use, survey model finds","feed_subtitle":"A 237-person structural model finds ethical attitude is a real path to AI use, not just a constraint.","key_machinery":"The carrying mechanism is a partial least squares structural equation model built by joining technology-acceptance constructs (performance expectation, facilitating conditions, social influence, behavioral intention, use) with the net-benefits dimension of the information-systems success model and with an adapted construct, Attitude Towards Ethical Behavior (AFCE), drawn from Internet-ethics research. Each research hypothesis is a path in this model, and the reported betas, p-values, F-squared effect sizes, and R-squared coefficients decide which paths survive. The argument rests on the AFCE-to-Use path (H5), because that is the route by which the conclusion claims ethics changes adoption behavior.","core_discovery":"The central claim is that an ethics-grounded adoption model is empirically viable: the Attitude Towards Ethical Behavior (AFCE) construct, defined as an individual's view of the fundamental ethical principles of Internet use applied to AI, directly and positively influences the Use of AI, with a reported standardized path of $\\beta = 0.207$ and $p < 0.05$. The authors interpret this as evidence that ethics 'impacts' people's decisions and that it is possible to adopt AI in society using AFCE as the model's mainstay. The same estimation supports the standard acceptance chain — performance expectation and facilitating conditions feed behavioral intention, behavioral intention feeds use, and use feeds net benefits — while social influence shows no significant association with either intention or use in this sample.","pith_inferences":["The causal language of the conclusion goes beyond what the cross-sectional, single-source survey design can establish; a longitudinal study that measures AFCE before any AI use and then logs actual usage would test the claimed direction.","Because the sample is mostly IT professionals, the null social-influence finding may not generalize; in less technical populations, social influence could resurface as a meaningful path.","AFCE is adapted from Internet-ethics items, so an AI-specific ethics scale covering bias, autonomy, and transparency might change the size of the H5 effect; that is a direct testable extension.","A common-method-bias check, such as a single-factor or marker-variable test, would clarify how much of the AFCE–use correlation is shared questionnaire variance rather than a real behavioral link."],"forward_implications":["Organizations can treat ethical attitude as an adoption lever: strengthening data-privacy norms and ethical-behavior expectations may increase AI use, not merely satisfy compliance.","Performance expectation is the largest driver of intention in the model, so users who believe AI improves their work are more likely to intend to use it.","Social influence plays no visible role in this sample, meaning the validated model does not depend on peer pressure or supervisor pressure to predict use.","Use strongly predicts net benefits, so adoption itself is positioned as the route to personal and organizational gains.","An ethics-anchored model can be empirically estimated and mostly supported, which the authors take as evidence that ethical AI adoption is feasible today."],"supporting_citations":[{"why":"Supplies the core UTAUT constructs (performance expectation, facilitating conditions, social influence, behavioral intention, use) that the proposed model fuses.","marker":"[7]"},{"why":"Provides the AI-specific acceptance model and the expectation that performance expectation and social influence drive behavioral intention, which the paper extends and partly contradicts.","marker":"[18]"},{"why":"Gives prior evidence that facilitating conditions and social influence affect AI use in organizations, the comparison for hypotheses H2 and H3.","marker":"[19]"},{"why":"Source of the Attitude Towards Ethical Behavior construct, adapted from Internet-ethics behavior to the AI context.","marker":"[17]"},{"why":"Supports the claim that ethical behavior intentions and attitudes influence actual behavior, reinforcing hypothesis H5.","marker":"[24]"},{"why":"Supplies the Net Benefits dimension and the use-to-success logic that underlies hypothesis H6.","marker":"[9]"},{"why":"Supports the direct relationships between facilitating conditions, behavioral intention, and use in consumer technology acceptance.","marker":"[16]"},{"why":"Provides the perceived-usefulness and ease-of-use foundation from which the performance-expectation path is derived.","marker":"[5]"}],"fun_headline_variants":["Ethical attitudes drive AI adoption, not just restraint","Ethics is a key path to AI use, study finds","AI adoption depends on ethical stance","Ethical behavior attitude predicts AI uptake","Model shows ethics shapes AI use decisions"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that one-time, self-reported survey answers from 237 mostly IT respondents can establish a causal direction in which ethical attitude drives AI use, rather than merely correlating with it.","fun_headline_variants_meta":{"raw":{"variants":["Ethical attitudes drive AI adoption, not just restraint","Ethics is a key path to AI use, study finds","AI adoption depends on ethical stance","Ethical behavior attitude predicts AI uptake","Model shows ethics shapes AI use decisions"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000189,"raw_usage":{"total_tokens":1338,"prompt_tokens":949,"completion_tokens":389,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":565,"completion_tokens_details":{"reasoning_tokens":321}},"tokens_in":565,"tokens_out":389,"duration_ms":4075,"temperature":1.0,"reasoning_tokens":321,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T05:28:57.764178+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Collect objective AI-usage logs from the same 237 respondents and re-estimate the model with logged use replacing self-reported use; if the AFCE-to-Use path (H5) is no longer significant after controlling for behavioral intention and facilitating conditions, the paper's central claim fails.","supporting_citations":[{"cited_title":"A new acceptance model for artificial intelligence with extensions to UTAUT2: An empirical study in three segments of application,","cited_arxiv_id":null,"evidence_quote":"Provides the AI-specific acceptance model and the expectation that performance expectation and social influence drive behavioral intention, which the paper extends and partly contradicts."},{"cited_title":"What drives students’ Internet ethical behaviour: an integrated model of the theory of planned behaviour, personality, and Internet ethics education,","cited_arxiv_id":null,"evidence_quote":"Source of the Attitude Towards Ethical Behavior construct, adapted from Internet-ethics behavior to the AI context."},{"cited_title":"What influences IT ethical behavior intentions —planned behavior, reasoned action, perceived importance, or individual characteristics?,","cited_arxiv_id":null,"evidence_quote":"Supports the claim that ethical behavior intentions and attitudes influence actual behavior, reinforcing hypothesis H5."}],"review_version":1}