Artificial Intelligence as a Research Instrument and an Editorial Burden: Scientific Judgment, Forestry Scientometrics, and Academic Incentives

Abstract

The increasing production of fake manuscripts with generative artificial intelligence (AI) is a genuine concern that leads editors to call for special countermeasures. In this opinion piece, I argue against targeting AI for these abuses, and I support this position with two levels of argument, micro and macro. First, AI is like other scientific tools and should be used just as they are. Second, the real reason for the escalating pressure to publish is a system that rewards the number of publications over their quality.

At the first, micro level, I examine examples of individual research tasks, competence, verification, replicability, and a defensible contribution distinguishing accountable assistance from delegated authorship. I use for it an example of an author-directed forestry scientometrics workflow on the Stanford/Ioannidis career database (the Stanford list). The provided examples show what guided exploration produces when used as an analytical tool, similar in nature to, for example, stepwise regressions, nonlinear mixed models, and even general applied mathematics and mathematical software, and what depends on documented author decisions. Along with the examples, I give some opinionated arguments that the evolutionary aspect of AI's game-changing empowerment is greater but no different in principle from other past advances, such as the emergence of mainframe computers, desktop computers and portable calculators, the Internet, software availability and other breakthroughs in scientific developments.

At the second, macro level, I consider the general predicament of publication systems and academic institutional requirements that shape and condition individual practices and values. I provide examples of an institution's public pay records, which track the volume of output by authorship position but show no appreciation of the impact of single-author papers, i.e., of citations per single-author paper. Awarding countable output more reliably than ambitious inquiry unavoidably promotes cheaper and faster manuscript production, resulting in the production of more lower-quality manuscripts pretending to be worthwhile research just to get published. The implied conclusion is that in this context AI is just the topping on the cake, while the reward system is the cake, and the observed trend towards fake publications is merely accelerated and facilitated by AI, not created or motivated by it.