While AI is accelerating data analysis and article production, machine-generated texts threaten the reliability of scientific publications and are forcing researchers to fundamentally rethink their established working methods. At a discussion held Friday at Opinion Festival titled "A Million Sensors, One Planet: How Artificial Intelligence of Things (AIoT) Is Changing Science," experts discussed the challenges and opportunities for science posed by AI as it becomes intertwined with billions of sensors and Internet of Things networks.
The participants noted that just a decade or so ago, collecting and interpreting data for a single research project could take years. Today, algorithms working alongside researchers can generate drafts of scientific articles in a matter of days or even less. As a result, the editorial offices of academic journals are being inundated with AI-generated material.
Helen Eenmaa, associate professor of governance and legal policy at the University of Tartu, sees the integrity of the scientific record as a serious concern. "Journals are receiving a great many scientific papers that have largely been generated using language models. Journal editors say the volume of manuscripts has increased enormously and that 90 percent of them are AI-assisted garbage. In addition to the extra work involved in filtering the material, many journals are publishing this kind of garbage," Eenmaa said.
The junk content making its way into publications is testing not only editors' patience but also the evaluation model entrenched at universities. At present, many academic institutions measure researchers' success largely by the number of articles they publish, but in the age of machines, raw publication counts are losing their original meaning. In this new environment, evaluation criteria need to be reviewed as a matter of urgency.
As a scholar in the humanities, Eenmaa even welcomed the fact that the problem is becoming more grotesque, as it highlights a longstanding sore point. "The number of publications is growing, but it no longer indicates anyone's scientific caliber or allows researchers to be compared. In Estonia, we have strongly embraced the trend toward quantification, and our system for evaluating research has been based on it. AI makes the problem even clearer: We urgently need to think about how to assess the quality of a researcher," Eenmaa said. Some universities abroad, for example, assess the content of a researcher's five best articles instead.
The current evaluation model's demand for a constant stream of new results also forces early-career researchers to work at a frantic pace. It is therefore little surprise that many are turning to automated assistants to advance their careers. TalTech doctoral student and junior researcher Uku Sildoja explained: "You can see among fellow doctoral students how enormously the volume of research has increased: In their first year, they produce six scientific articles and attend three conferences, whereas in the past it took one to two years to produce a single article. That's incredibly fast, but it's dangerous to operate too much on autopilot, and at some point something is going to give."
The academic race can tempt researchers to cut corners. In high-pressure situations, ethical rules can often be relegated to a secondary formality. "AI speeds up processes, and to remain competitive we don't have time to think — we have to churn out articles like a sawmill churning out boards. In doing so, we may push ethical issues into the background, but that's as good as unleashing an AI agent on the world when there's no way to put it back in the bottle," warned Alar Kuusik, director of TalTech's Thomas Johann Seebeck Department of Electronics.
The participants noted that reducing these risks will require a new type of leadership in the scientific community. Rather than being narrowly focused specialists, researchers are increasingly expected by society to have a thorough command not only of their own fields but also of legislation and ethical boundaries. These new responsibilities inevitably leave many researchers uncertain about their role.
In effect, researchers must take responsibility for the entire research process. "The question is what the researcher's role is: Does the researcher decide what to measure, or do they have to start managing the entire decision-making process? Researchers are increasingly becoming managers of the research process who have to make decisions about data protection, ethics and legal issues they did not previously have to deal with. This creates uncertainty for many researchers because they need to quickly become competent across a broader range of issues," Eenmaa said.
On the other hand, intelligent agents can independently process vast quantities of data, meaning routine data processing is increasingly being left to machines. The time this frees up gives researchers greater scope to look for new and more complex connections.
However, reducing the human role carries serious risks because machines have no sense of moral responsibility. Sildoja highlighted the hidden dangers of automation: "It's a double-edged sword: AI can speed things up and allow one person to do the work of an entire research group, but on the other hand, it also makes it possible to produce 10 times as much nonsense. If you don't critically keep your finger on the pulse and stay on top of what the agents are doing, you can end up with some striking examples — for instance, AI can start adjusting results on its own in analysis scripts."
If machines are left unsupervised, science can therefore quickly become unreliable. Researchers should use the working hours freed up by AI to improve the substantive quality of their research. Sildoja added: "There may be endless amounts of data, but ultimately someone has to decide what is important, why it is important and how to use it. AI allows researchers to accomplish significantly more with their time, but it is important not to become lazy or let AI do the grunt work simply so that you can work less. Instead, the same amount of effort should be used to pursue more ambitious projects." Every expert should therefore recognize the social dimension of their field and take responsibility for the consequences, while research teams working on future projects should learn to set boundaries independently. "I dream of us being able to discuss ethics without having to turn separately to a philosopher or lawyer — responsibility should be an inherent part of every researcher's work. The accountability mechanism should be better. Ethicists and lawyers do not always understand the scientific substance of another field, and advising researchers is a secondary activity for them," Helen Eenmaa said.
The discussion was organized by the Estonian Research Council (ETAg).
-- Editor: Jaan-Juhan Oidermaa, Marcus Turovski