By Pål S. Malm, Senior Project Manager, SINTEF Digital

(This article was first published in NorwAI Annual Report 2023, 2024-04-01)

In 2022, attempts were made to defraud DNB customers of NOK 1.2 billion. With the help of AI and other tools, the bank managed to save one billion of this. No wonder NorwAI’s partners are impatiently waiting to implement the AI models they see prototyped in the project.

Artificial Intelligence is an enabling technology with short distance from cutting-edge research to innovations. The launch of ChatGPT in November 2022 has caused awareness of generative AI and large language models to skyrocket. The accessibility of generative AI has lowered the barrier to conduct internal experiments, and many of our partners’ innovative initiatives are pertinent to our NorwAI.

Innovation happens through discussing new ideas, sharing new knowledge and collaborating on developing technologies that can be used for industry partners. The term “innovation” has different meanings in academia and industry. According to the EU Innovation Radar, industry innovation is usually focused on “business-ready” and “market-ready” technology, while academia defines innovation as “exploring innovations” and “tech-ready” technology. Research and technology organizations (RTO) like SINTEF, is often bridging academia and industry. Use cases are useful to describe the “job that needs to be done”. In NorwAI Annual Report 2021 this innovation approach was described as a way to start with the user, brand need or problem, as opposed to having data or technology as a starting point.

Innovation elements

The Research Council of Norway’s definition of innovation describes most of the elements we associate with innovation in NorwAI:

“New or significantly improved products, processes, production or distribution processes, organizational or management forms, or concepts introduced to enhance value creation or benefit to society.”

NorLLM is an innovation that has had a positive impact beyond its technical capabilities. Norwegian language model is on the political agenda, and there is public engagement for a national LLM for cultural preservation and representation for public services and education. Furthermore, we have gained an increased understanding and sense of the urgency of making national content and training data available.

Using AI, DNB has saved itself several hundred million kroner in recent years, and customers one billion, in 2022 alone. The technology they
use is cheaper and more accessible than ever. – Maria Ervik Løvold (COO DNB) to Dagens Næringsliv

DnB says, for example, that they have benefited greatly from NorwAI and NorLLM. Via NorwAI, they have had access to researchers with hands-on experience for DnB’s data science community to discuss to see how large language models in general can be tailored for internal use.

Already experimenting

DnB and Sparebank1 SMN are experimenting with AI and language models to detect fraud and financial crime. Attempts were made to defraud DnB customers of NOK 1.2 billion in 2022. With the help of AI and other tools, the bank managed to save one billion of this.

Criminals use AI to make more sophisticated fraud attempts and banks need to be more proactive in preventing financial crime. However, legal regulation restricts the use of transaction data internally and the sharing of data with other banks, although this could improve banks’ ability to detect fraudulent activity. Banking, finance and telecom have similar issues; They want to utilise internal data to improve services and customer dialogue and want to share data with other banks to prevent fraud and financial crime but find that sharing data is a demanding regulatory area.