The Nordic State of AI

Companies that need large language models can use foundation models that have been trained using publicly-funded resources, and fine-tune their own AI on top of it. The open-source foundation models add to the transparency of the tools and democratize access to these new technologies.

The low priority of “Explainable AI”, especially compared to the previous year, is interesting. With the significant increase in the use of Generative AI/LLM technologies, this indicates a strong prioritization of productivity increases. The lack of adoption of Bayesian learning can be explained by two factors: concerns over the practicality of Bayesian counterparts of cutting edge neural network methods, and a lack of understanding of the advantages of the Bayesian methods, such as better data exploitation especially in lower size datasets and a natural expression of uncertainties in predictions. The rest of figure 10 exhibits predictable changes in the use of technologies, going hand in hand with the overall industry and academic trends.

The trend now is that foundation models are being fine-tuned for domain-specific areas. With high quality data, you could have smaller models that have the same performance as larger models.

I think it's important to have knowledge internally, and not to do everything from scratch. You need to be able to choose external vendors, to know what tools to buy.

AI is practically an extension to conventional software. The development and utilization of trustworthy AI requires foundational understanding of software development but also of data processing and machine learning. The AI landscape will continue to evolve a lot in the foreseeable future, which emphasizes the need for continuous talent development. Recruitment and outsourcing are probably both needed, given the high demand for senior talent required to guide less experienced professionals. A necessity, along with recruiting and outsourcing, is significant investments into talent development of both seniors and juniors.

There are companies in the financial industry who are still debating if AI is something they should do or not. I think that is a strategic mistake. There are ethical concerns, but not trying is a choice that puts you in the backseat when it comes to adoption.

As the AI landscape matures, increasing attention is being paid to data-centric AI. Focusing on obtaining and engineering high-quality data that is consistently labeled would unlock the value of AI across a variety of domains. As such, developing shared, systematic practices around data is key to ensuring a high chance of success of new projects, while ensuring a high level of operational readiness and minimized turnaround time.

The skills required in an AI project depend on at which stage the project is. At the beginning it’s important to have experts that can scope the project well, and at the later stages more technical expertise is required. For this reason it’s important to have diversity in the project team and to utilize everyone’s strengths to get the best outcomes.

I believe that successful adoption of AI starts with the leadership, at a level where there is a solid understanding of the big picture; the business objectives and the resources needed. That way there is a strong motivation and commitment to take AI to production.