Agree to create guidelines for the use of artificial intelligence

Senior researcher Daniel Weiss leads the group looking at Nordland Research's use of artificial intelligence. Photo: Marta Anna Løvberg

The five regional social science research institutes are joining forces to develop guidelines for the use of artificial intelligence in research.

Text: Idun A. Husabø, Vestlandsforsking

“Development has been incredibly fast, and many researchers were out early,” says Director Anne Karin Hamre at Vestlandsforsking.

ChatGPT shocked us all when its language model was released online in November 2022 and we realized what it was capable of. Over the course of 2023, one AI tool after another became known and joined the competition to make everything from writing to translation and text analysis child's play. Artificial intelligence has created excitement. But it has also created fear, for example that "deepfakes", videos where you can make anyone look like they are saying anything, are just a click away. 

Can perform many research tasks

Researchers were also among the early ones to explore what ChatGPT and other AI-based tools could be used for. It turns out that AI can do many of the tasks that are part of researchers' everyday lives, such as outlining a lecture, writing a text on a given topic, converting audio recordings from interviews to text, or even creating an overview of the research literature in a new field.

There is a wealth of tips and advice on how AI can perform tasks more efficiently than humans, but what are the consequences for researchers and research institutes of using AI? 

Ethical and legal consequences

Not all forms of artificial intelligence, or the use of AI, meet privacy and ethical requirements. With the sudden availability of artificial intelligence, there was a need for more knowledge and guidance – because perhaps not all tools and applications were equally harmless?

– Artificial intelligence can clearly offer great opportunities for us, but it also offers ethical and legal consequences. We see the need to develop driving rules and internal guidelines for artificial intelligence, says Anne Karin Hamre.

Joint work in the Focus Institute 

Trond Runar Hagen from Sintef Digital gave the directors of the Focus Institute an introduction to the advantages and disadvantages of using AI.

Other regional social science institutes have also begun to put in place frameworks for how employees can use artificial intelligence in their jobs. The Institutes of Møre Research, Telemark Research, Nordland Research Institute , NorSus and Vestlandsforsking, which have an ongoing collaboration under the umbrella of Fokus, are now making plans to develop joint guidelines for the use of AI in research.

When they met recently, the Fokus directors invited Trond Runar Hagen from Sintef Digital to give a professional introduction to the pros and cons of artificial intelligence.

Many pitfalls

Many of the challenges with AI are well-known, for example, that it can easily produce false information and images that can deceive us, while also affecting models like ChatGPT. Other dangers are also in store: If a researcher inputs confidential information or self-produced text into a model that uses such input to train the model, this means in practice that this is shared with the AI ​​system. One challenge is to have full control over where the information ends up and who can access it.

It is very serious that this could threaten privacy in research. But it is at least equally serious that this could give rise to problems around research ethics related to plagiarism, copyright and source criticism.

There are other challenges with AI models. They can be biased and favor or discriminate against certain groups or political views. Or they can – as ChatGPT was known for – “hallucinate”, that is, make up things that are not true. Language models like ChatGPT are not designed to tell the truth, but to calculate what is the most likely next word in a sentence. This means that AI must always have a vigilant user.

– AI and algorithms may respond in different ways than the user expects, and the result must always be quality assured, emphasized Hagen. 

Research leader Hilde G. Corneliussen reminds us that the guidelines should give researchers room to explore AI in their work.

Part of this problem is that those who create AI tools often hold their cards close to their chest:

“We lack insight into the models and how they are trained,” said Hagen. For the user, this means that you can never blindly trust what comes out of these tools.

Joint working group

Both Vestlandsforsking and Nordlandsforsking have already had working groups that have worked on digital tools and AI. Hilde G. Corneliussen, who is research leader for technology and society at Vestlandsforsking and has, among other things, researched artificial intelligence in the public sector, has, together with Halvor Dannevig and Hans Jakob Walnum, designed interim guidelines for Vestlandsforsking to support researchers who are already exploring AI.

When the efforts are combined, an internal working group will be formed across the five Focus Institutes to develop a new, common set of guidelines. These will cover, among other things, what researchers and administration can use AI for and where we need to tread carefully in terms of law and ethics. And as Corneliussen reminds us:

– The guidelines should not only address prohibitions and risks. They should also embrace engagement and give employees the space to use AI in a good and productive way.