Does AI pose a threat to research integrity? To answer this question, I would like to take a brief tour through the concepts of normal science (Kuhn 1962) and big science (De Solla Price 1962). Kuhn argued that normal science is carried out inside a scientific paradigm, in which researchers look for evidence that confirms the accepted theory and discard non-confirmatory findings as anomalies. A number of institutions of modern science are normal science institutions. For instance, the process of peer review aimed at ensuring the quality of research is a process that normalizes science: an article is seen as having high quality if its methods reproduce exiting scientific praxis, if its theory is robust according to existing scientific theories, if it reviews and creates a dialogue with the existing literature and frames its contribution as an incremental addition to the existing paradigm. I would argue that the same tends to hold for papers that ‘challenge’ existing ideas: they do so by referring to the literature and by working from within the paradigm, in many cases. As a result, a substantive part of a scientific article is normal science, it reproduces and rehearses accepted knowledge and from this framing, it adds its contribution. AI can help reproduce and rehearse accepted knowledge, and in this sense its use fits well with normal science paradigms and may help researchers better comply with the ‘scientific article’ genre, and thus advance their career. A second example of normal science institution is PhD training, which requires learning and applying the research methods of the field, doing literature reviews and situating one’s research with respect to accepted knowledge, making a research plan with clear objectives that state which puzzle one wants to solve and how. Similar arguments hold for writing research proposals. All in all, a lot of scientific research involves paying tribute to theories, literature and methods in a normal science fashion – and AI can assist in doing just that. Please note that I’m not saying normal science produces bad research, my point is that its practices are easy to emulate by AI and it creates the conditions for AI to be seen as a useful “assistant” in doing research. Is this a threat to research integrity? Was research integrity already threatened before AI, and we’re just seeing an intensification of the threat?
The second concept I would like to refer to is that of big science. De Solla Price distinguished between little science, done by solitary researchers or small teams, with little infrastructure and funding, driven by curiosity, and big science, characterised by large-scale projects that require significant funding, extensive collaboration, and advanced technology – think of the Manhattan project, where large teams of scientists were brought together and guarded by the military to create the atomic bomb. Big science leads to the professionalisation of scientists: being a researcher is a job, and one learns to be efficient at it, productive. Once again, big science provides a great breeding ground for the use of AI. Professional science and competitive funding create phenomena like ‘publish or perish’, reinforced by promotion mechanisms at universities, which lead to a situation where researchers may like, in an ideal world, to devote time to slow, rigorous, careful research, but feel pressured to deliver and may turn to AI. In this case, I think that it is clear that AI does not create the threat to research integrity, it merely amplifies the crisis of scientific practice, sometimes identified with the reproducibility crisis (Ioannidis 2005), retraction scandals, publication mills, etc.
And then, the message I receive as a researcher is that I should make ethical and responsible use of AI. As long as I check what AI produces, there is no problem in using it, apparently. Integrity is the responsibility of the individual – never mind that the system of incentives out there makes a joke out of integrity. Never mind that if I use AI to write more papers, I contribute to the paper mills phenomenon, to a productivist version of science as an article-maximising business, to paraphrase a phenomenon Latour and Woolgar already identified in 1979.
By the way: If I don’t use AI, I am told, I will fall behind those who do – I really don’t buy this argument. Remember the MIT Technology Review’s cover from 2012 “You Promised Me Mars Colonies. Instead, I Got Facebook”? I wonder if, after all the promises that AI will solve major social challenges, what we’re going to be left with is using AI to write social media posts. Now back to the main argument.
Let me close with a bit of a doomsday reflection. I don’t think that AI creates it, but it intensifies the hollowing out of science and of knowledge. Some say that AI will make the peer review system collapse – but so-called predatory journals and paper mills haven’t made the system collapse and have been around for over 15 years now. Maybe it hasn’t collapsed yet. Or maybe, and here comes the doomsday part, they have rendered scientific publications irrelevant, an activity that sustains the professional mechanisms of academia but that is largely irrelevant to the outside world. Maybe the irrelevance of at least part of scientific outputs can be observed in the post-truth turn and the emergence of governments (notice the plural, this is not just a US thing) that outright deny the value of scientific knowledge.
Integrity is important: it is the virtue of being whole, of being one and not many-faced. Many researchers are driven by integrity and can surely find ways of using AI with integrity – but that’s not an easy-fix. AI emerges in a specific historical moment, reinforcing specific trends in science at the expense of others, and colluding with specific tendencies in politics at the expense of others. To me, integrity is not about the use of a tool, but about asking questions such as: which practices, values, institutions and politics does AI, in its current configuration, reinforce?
References:
Ioannidis J.P. (August 2005). “Why most published research findings are false”. PLOS Medicine. 2 (8) e124.
Kuhn, T.S. (1962). The Structure of Scientific Revolutions (1st ed.). University of Chicago Press.
Latour, B. and Woolgar, S. (1979). Laboratory Life: The Social Construction of Scientific Facts. Sage.
de Solla Price, D.J. (1963). Little Science, Big Science. New York: Columbia University Press.
0 Comments