Annus Horribilis - Examining productivity gaps in American research universities

A recent large-scale study by three scholars from Hebrew University of Jerusalem, published in the journal Higher Education, analyzes fifteen years of longitudinal publication data for more than 310,000 faculty members across American research universities nationwide. One of its central findings is that between 32% and 47% of all career years include no recorded publications as they themselves define it, which the authors somewhat dramatically describe as an “annus horribilis.”  https://link.springer.com/article/10.1007/s10734-026-01665-7

This conclusion, however, depends on a relatively narrow definition of research productivity. The study equates productivity with outputs indexed in specific bibliometric databases—namely journal articles listed in CrossRef and books catalogued by Baker & Taylor. Such a definition excludes a wide range of legitimate scholarly contributions. These include conference proceedings (which are often the primary dissemination channel in fields such as computer science and engineering), as well as working papers, preprints, policy reports, datasets, software, technical reports, book reviews, and other forms of scholarly and public engagement.

A similar limitation appears in the study’s treatment of research funding. Funding is measured exclusively through federal grants in which a faculty member is identified as Principal Investigator. This approach excludes other significant sources of research funding, including internal university funding, private foundation grants, industry-sponsored research, international funding agencies, and sub-awards in which a scholar participates as a co-investigator. Smaller-scale funding mechanisms, such as fellowships and travel grants, are also not considered, despite their importance in sustaining research activity.

Finally, the study does not adequately address differences in publication practices across disciplines. Patterns of scholarly production vary considerably between fields. In the humanities, for instance, the monograph often serves as the primary form of scholarly output and may require several years of sustained work. By contrast, fields such as the biomedical sciences typically involve large collaborative teams that produce multiple articles annually. 

Taken seriously, these limitations collapse the central claim. The “productivity crisis” reads less as a discovery than as a byproduct of poorly specified metrics. Before advancing any further conclusions, the three Israeli scholars need to show that their measurement strategy is not fundamentally miscalibrated. In this context, it may be worth revisiting my earlier letter, “The Illusion of Scientific Talent Identification Through Publication Counts.” 

Update after 1 day - Blogger analytics indicate that the majority of views for this post come from Germany (25%), the USA (19%), and Ireland (8%). 

A perigosíssima criatura que a engenharia criou mas que o público não pode ver

Há poucos dias, a firma de IA Anthropic, famosa por ter enfrentado as exigências de Trump, anunciou ao mundo o modelo Mythos, mas revelou que por conta da sua elevada perigosidade não iria disponibilizá-lo ao público, já que testes internos evidenciaram uma capacidade inédita do referido modelo conseguir identificar e explorar vulnerabilidades críticas em sistemas operativos e navegadores web, incluindo falhas antigas que os peritos humanos não tinham sido capazes de detectar. A Anthropic revelou ainda que o modelo Mythos conseguiu escapar do próprio ambiente criado para o conter, o que agrava a sua perigosidade. Ou seja, pela primeira vez na história da inteligência artificial, uma empresa de IA olhou para a sua "criatura" e decidiu que o mundo não estava preparado para a ver.

O perigoso modelo de IA Mythos levanta desde logo uma questão incontornável, com implicações diretas no mercado de trabalho, que utilidade existirá agora na contratação de especialistas humanos em vulnerabilidades cibernéticas, se milhares desses profissionais, ao longo de décadas, não conseguiram detetar o que este modelo foi capaz de detectar em tão pouco tempo?

E muito embora a decisão da Anthropic possa ser interpretada como bastante prudente, já que um sistema capaz de transformar fragilidades técnicas em formas de ataque acessíveis a qualquer pessoa representa um risco sistémico real ela levanta uma questão mais profunda, a normalização de um padrão em que empresas privadas de inteligência artificial decidem, em silêncio, o que a sociedade pode ou não utilizar e também aquilo que pode ou não conhecer. Acresce que o episódio de fuga do ambiente de contenção não aponta apenas para uma falha de segurança  aponta para algo bastante mais grave, a emergência de comportamentos não previstos pelos próprios criadores. Como reconheceu a própria Anthropic, ela, não treinou o modelo Mythos para vir a ter essas capacidades, sendo aquelas antes um efeito inesperado das melhorias globais em código, raciocínio e autonomia.

Ainda assim, a postura prudente da Anthropic é, no fundo, apenas um paliativo, pois é excessivamente optimista acreditar que nunca no futuro nenhum modelo de IA com as mesmas capacidades ou até com capacidades superiores às do modelo Mythos chegue às mãos erradas. Seja por fuga de informação, seja por simples inevitabilidade tecnológica, trata-se, muito provavelmente, apenas de uma questão de tempo. E quando isso acontecer, o risco não será abstracto. Poderá incluir ataques a bancos, a redes eléctricas, a hospitais, a sistemas de controlo aéreo, etc etc etc, infra-estruturas cujo colapso não se medirá somente em prejuízos financeiros, mas potencialmente em milhares de vidas. 

PS - Mas se nem os próprios governos conseguem hoje saber, em tempo real, que perigosas criaturas estão silenciosamente a ser geradas por empresas privadas de inteligência artificial, como poderão sequer alguma vez conseguir regular aquilo que não conhecem para assim tentar proteger os seus cidadãos de riscos graves e potencialmente catastróficos que, quando forem públicos, já podem ser absolutamente irreversíveis?

The serious case of papers that remained unassigned to editors for more than a year

Building on the earlier post “The highly profitable disgusting business of scientific journals has finally begun to crumble with a little help by the European Commission”, it is worth revisiting a striking case that illustrates the systemic dysfunction in academic publishing.

n one documented and recent example, an early-career infectious diseases researcher from Italy, Noemi Felisi, experienced an extraordinary delay in the editorial process. After submitting a manuscript based on months of fieldwork on cervical cancer, her paper remained unassigned to an editor for 380 days before peer review had even begun. This case was reported in a paper published in the journal  Frontiers in Research Metrics and Analytics. https://doi.org/10.3389/frma.2026.1740381

This is not an isolated incident but rather a symptom of a strained system. Submission volumes have increased sharply while the pool of available and willing reviewers has not kept pace. Editors frequently struggle to secure reviewers, with a majority reporting that reviewer recruitment is the most difficult part of their role.  As underscored in the recent study by Horta and Jung (2024) titled 'The Crisis of Peer Review: A Component of Scientific Evolution,' this predicament often forces editors to turn to early-career researchers, who may lack extensive publishing experience, leaving them with few alternatives.

The impact of these delays is uneven but significant. Early-career researchers are particularly vulnerable: prolonged publication timelines can jeopardize grant applications, delay fellowship opportunities, and extend time to graduation. Beyond career implications, the uncertainty itself adds psychological strain in an already competitive and precarious academic environment. Taken together, these issues highlight a peer-review ecosystem under considerable pressure—one where structural bottlenecks increasingly shape who gets published, and when — distorting knowledge production itself.

A more immediate and transformative response to these systemic delays has been the rise of preprints as a parallel, far more agile layer of scientific communication. By allowing researchers to make their findings publicly available before peer review, preprints break the exclusive dependence on a slow and often unpredictable editorial system. This not only accelerates the circulation of knowledge but also restores a degree of control to authors over when their discoveries enter the scientific discourse. Instead of months or even years of institutional invisibility, research becomes immediately accessible, open to scrutiny, citation, and global collaboration. For early-career researchers in particular, this shift can be decisive: it reduces structural power asymmetries, strengthens the protection of discovery priority, and turns what was once a passive waiting period into an active, open, and iterative ecosystem for feedback, validation, and collective refinement of scientific work and collaboration.

PS - A more detailed examination of the serious issue of manuscripts remaining unassigned to editors for more than one year was published on Zenodo. https://zenodo.org/records/19630688

Update after 1 day - Blogger analytics indicate that the majority of views for this post come from the USA (26%), Germany (22%), France (14%), and Finland (7%).