Last week in this space, I wondered what will happen to academic contestation as AI makes it cheap to scrutinize research. This week, one new paper argues for putting evidence first, while another puts Claude in the peer reviewer’s seat, and the APA’s activist stance comes under scrutiny.
Here are the stories we’re following at HxA.
Empirical Evidence (and Egalitarian Ends)
In a new working paper, Stanford’s Tom Clark and Jonathan Mummolo of Princeton make the case for “Privileging Evidence in Empirical Social Science.” With a politically balanced set of examples from research on topics including policing, elections, and terrorism, the authors make a classic argument for centering quality evidence with key criteria including contestability. The authors argue:
Researchers must be as transparent as possible with regard to data, methods, and assumptions, and be willing to have their efforts scrutinized, come what may. Just as importantly, other scholars must be interested in and incentivized to contest evidence. … We support the ‘open science’ movement, but all the data repositories in the world will not translate into accountability if critics expect character assassination in return; they will simply abstain from criticism in the first place.
What else would be privileged in empirical work, if not evidence? Clark & Mummolo point to recent proposals to use what they call “non-evidentiary criteria” such as the moral quality of the data source, the identity of the researchers themselves, and the assumed political and social impact of the research finding. Using such considerations to limit inquiry, they argue, is an egalitarian-seeming move with inegalitarian consequences: “When evidence is not allowed to speak, its role is too often supplanted by status, credentials, and rhetorical skill. Evidence, in contrast, is a great equalizer.”
Clark & Mummolo don’t measure how often such “non-evidentiary” considerations are put into use, although they suspect that they are not “prevalent or dominant.” I’m not so sure. But that’s an empirical question that deserves good evidence of its own.
Claude’s Peerless Peer Review
Clark & Mummolo approvingly cite Karl Popper’s description of science as the “friendly-hostile co-operation of many scientists.” But it’s 2026, so we must ask: is that only human scientists?
In another paper this week, Harvard’s Gary King and Steven Worthington tested whether AI can check papers like a human peer reviewer. When three former journal editors blind-evaluated a set of manuscript reviews written by human reviewers and Anthropic’s Claude, they preferred the AI review in 26 of 30 judgments. The AI reviews also made about a third as many incorrect claims about the manuscripts (5.5% versus 15.9%) and cited no fabricated sources.
The authors suggest that journals could have authors revise against AI reviews before submitting, publish the prompts their reviews use, or let editors decide from an AI-assisted dossier in place of routine human review. They also warn that scholars who brainstorm with AI can converge on similar ideas, which drift toward what the authors call the “conventional, orthodox, and majoritarian.”
“AI benefits us individually,” they write, “but risks making the collective efforts of science more homogenous and ordinary.”
APA’s Future President Is ‘Accountably Subjective’
The American Psychological Association is fighting to keep its role accrediting psychology training programs. Last month an Education Department advisory board voted 7-4 to recommend suspending recognition of the APA’s Commission on Accreditation, after testimony about antisemitism in the field. In a Wall Street Journal op-ed this week, HxA members Colin Wright and Andrew Hartz argue that the allegations reflect a wider failure in parts of organized psychology to separate activism from science.
This month, the APA elected Emory psychologist Candice Hargons as its 2028 president; she becomes the official president-elect in January. Hargons’ views are on the record in her campaign statement, and they suggest a close coupling of science and activism. “By uniting against intersectional oppression,” she writes, “we will position APA not just as a witness to social change, but as a courageous, unwavering architect of global liberation by collaborating with peer organizations who fight injustice.”
Asked how she defines scientific integrity, Hargons writes that she rejects “the positivist paradigm’s assertion of objectivity” and that “true integrity means being accountably subjective; we don’t hide who we are or lie to the public about what we find in our research.”
Positivist or Always Sex-Positive?
Since her election, Hargons has been criticized online for her own work, in particular a 2022 article with Shemeka Thorpe entitled “#HotGirlScience: A Liberatory Paradigm for Intersectional Sex-Positive Scholarship.” The paper openly calls for researchers to weigh non-evidentiary factors alongside evidence, making it a useful example to consider.
Read past the hashtags and playful hip hop references, and this paper paints a sympathetic picture: research about Black people and sex has been too ignorant of context, too focused on finding the negatives, and overly serious along the way. (Many HxA members understand that certain fields can grow overly pessimistic. What do these dour sexologists think they are researching, American Studies?)
The positivist approach, following Clark & Mummolo, would be to shine the light of careful inquiry on those possibly overlooked angles in sex research. What is the evidence about pleasure, not just pain? What are the truths that become clear with the benefit of richer context? If race shapes people’s sex lives in overlooked ways, what are those ways and what evidence tells us?
Hargons & Thorpe’s proposed research paradigm takes another path, continuing normal science – including quantitative data analysis – but layering in non-evidentiary considerations. For example, instead of universalism, they advocate identity-based particularity: “Only #HotGirlScience scholars born and immersed in the sexualities they study can use quantitative methodologies and do them justice, because their interpretation is the only justifiable frame for authenticity, fun, and sex positivity.”
Instead of methods that are carefully agnostic to outcomes, Hargons & Thorpe’s paradigm
requires the most loving, sex-positive, nuanced interpretation of the research results. This means that even when the results are negative and not sex-positive or liberatory, researchers acknowledge and state the results but interpret them in meaningful and robust ways that reflect the full context of history and politics that inform the outcomes. Researchers, especially those publishing in academic journals, should ask of themselves: ‘How would participants feel if they read what I wrote about them, their sexualities, and their lived experiences?’
This is science with a commitment to share results, but also an intention to celebrate some results and contextualize others. We can trust such a paradigm to notice what it planned to notice, to safeguard the feelings of research subjects, and maybe even to improve the subjective experience of #HotGirl scientists. Can we trust it to describe the world?
Clark and Mummolo's line about critics simply abstaining when they expect character assassination gets at something open-science policies can't fix with repositories alone. Incentives to contest work matter as much as access to the data, and plenty of fields have quietly lost the friendly-hostile cooperation Popper described. The King and Worthington result is striking too. Editors preferring Claude's reviews in 26 of 30 judgments, with about a third of the error rate, will tempt overloaded journals fast. Their warning about convergence toward the conventional and majoritarian may be the bigger story, since an AI reviewer trained on the existing literature could end up reinforcing whatever consensus already dominates a field, including on politically loaded topics. Interested to see whether anyone measures how often those non-evidentiary criteria actually get applied, as you suggest.
I found this part of the article important: “What is the evidence about pleasure, not just pain? What are the truths that become clear with the benefit of richer context?” These questions dig deep when approaching a topic.