Author : Jurij Selan
The phenomenon of artificial intelligence (AI) is inherently paradoxical. On one hand, it is generative. This generative quality has benefited contemporary research by enabling researchers to generate ideas and enhance research opportunities while saving time and costs.
On the other hand, the generative nature of AI appears inevitably to lead to antagonism, resulting in entropy through model collapse and becoming degenerative. In this article, we explore the extent to which the implicitly degenerative nature of AI could be regarded as the main long-term threat to research integrity (RI), as many other problems associated with the impact of AI on RI may be seen as its effects.
In the first part, we provide an overview of the impact of AI on RI, including AI ethics, the use of AI in education (AIED), AI as a “grey area” or questionable research practice (QRP), the implementation of principles for AI use in codes of conduct, and the attitudes of academic publishers and universities towards AI. In the second part, we examine how the collapse of generative AI into degenerative AI poses a critical threat to RI in the future.
We emphasise that the only way to prevent the harmful effects of the degenerative nature of AI on RI is to retain the original human-generated datasets as the basis for AI systems and continually add new human-generated datasets.
One of the key principles regarding the impact of AI on RI is therefore the responsibility to ensure that AI remains grounded in human-created reality. This, however, leads us to the sociotechnical perspective on degenerative AI, which we address in the third part, where we evaluate the broader social and moral impact of degenerative AI.
We stress a fundamental shift in human trust requirements towards society and make a plea for more inclusive anticipatory risk management of AI with respect to RI.
