A broad-based expert group will help to highlight key perspectives and make recommendations. The group had its first meeting in March 2023.
Why we are doing this project:
We are experiencing an “iPhone moment” for artificial intelligence
Since the Norwegian Board of Technology published the report Artificial Intelligence – Opportunities, Challenges, and a Plan for Norway in 2018, the field has evolved very quickly. The emergence of large-scale language models, developed by companies such as Google and Open AI, is a turning point for the most powerful technology of our time.
The iPhone succeeded because it was highly advanced technology that could be used for almost anything, was easy to use, and made it hard to imagine life without it. Large language models do the same for artificial intelligence.
The language models are:
- General and powerful
With a powerful and general purpose, the models create unique and advanced content themselves, based on short instructions. Trained on content from the internet, they have written books and music and won art competitions.
- Solving new tasks
The models have an unforeseen ability to solve new tasks, from generating new antibodies to predicting how proteins fold. Stanford University believes the technology could alter 4 out of 5 jobs.
- Easy to use
The models are easy to use and have quickly become very popular. ChatGPT reached 100 million users in two months, making it the fastest growing digital service of all time.
- Available to everyone
The language models are becoming available as APIs, and the programmes based on them are becoming part of most digital systems. The technology is already present in Google search and in Word, Excel, and Teams.
Developments are happening fast – and regulation is not keeping pace
Development is not without risk. Language models are unpredictable and unreliable, and this is inherited by the programmes that build on them. The models are too complex for developers to anticipate everything they might do, and while what they say may appear correct, they have no concept of truth or falsehood. They hallucinate if they must and can be discriminatory and manipulative. It is also difficult to control and know how they operate because of the secrecy of the companies.
There are now several attempts to set a framework for artificial intelligence. The US has launched its AI Bill of Rights – a list of voluntary recommendations for developers of the technology. The EU is negotiating its Artificial Intelligence Act, the world’s first comprehensive regulation of artificial intelligence. China is introducing a new law to ensure that generative AI safeguards socialist values and does not undermine state power.
Hundreds of experts and technologists, including Elon Musk, Apple founder Steve Wozniak and historian and author Yuval Noah Harari, argue in an open letter that the pace of development is too fast. They fear that machines will flood our information channels with propaganda and untruth, automate all the jobs, and eventually take control of our civilisation. Therefore, they believe that all development of large-scale language models more powerful than those currently available should be put on hold to give policymakers time to act.
Aim of the project
The aim of this project is to map this paradigm shift for artificial intelligence, assess which questions will be particularly important for Norwegian government to consider, and to make recommendations.
Some of the topics the project will address:
- The need for large language models developed in Norway based on Norwegian data
- How the technology will affect schools and education
- Which jobs will be affected, and how
- The use of large language models in public administration
- Implications for creative professions, copyright, and patents
- Measures for dealing with the increasing amount of machine-generated content online