The AI Advisory Expert Group in Norway Delivers First Note on Structuring the AI Billion
The Norwegian artificial intelligence advisory expert group submitted its first input to the Research Council of Norway
The following is a quick translation from the first note delivered to the Research Council of Norway the 15th of November 2023, on the AI billion (NOK). I have slightly altered the formatting of the text from the original that can be found here.
It is not my intention to change how the text is interpreted, but rather to highlight some of the text for the purpose of this article on Medium. As such some text is larger than it may appear in the note or bold text appear in different sections.
You can find more information about the process structuring the AI billion investment in research here:
Underneath you will find the 1st note translated from Norwegian to English.
1st note from the advisory expert group for the AI initiative — current situation and expectations
AI (Artificial Intelligence) and digitalisation affect our society in an extensive way. digitalisation is crucial to meeting key societal challenges around the economy, demographic changes, skills needs, the climate crisis, global health, inequality, increased polarisation, digital alienation and trust.
At the same time, the use of the new technologies will have major and partly unforeseeable social consequences.
The AI billion alone will provide limited opportunities to solve all the challenges. A separate investment in AI will, however, be able to set a direction for the research effort, which in turn can be followed up with other research and innovation projects, and new political measures.
Collectively, this can make Norway an active participant that helps to set the premise technological development.
As described in the mandate, the initiative includes three integrated tracks
- Consequences of artificial intelligence and digital technology
- Digital technologies as a research area in itself
- How digital technologies can be used for innovation, including in research
‘Digital technologies’ is a very broad area. We have chosen to interpret the mandate not to include all digital technologies, but to limit this to AI and digital technologies that are related to or enable AI. This is referred to hereafter collectively as AI.
With this as a starting point, we have looked at research within the three the tracks.
We assume that more sector-oriented applications and innovation projects are primarily financed within other programs or from other sources.
Large infrastructures for generative language models, data and processing infrastructure for machine learning and other AI methods are also outside our mandate.
At the same time, such projects and infrastructures will be important for some of the effects of the research to be possible to materialise.
The initiative’s three tracks are closely linked. The connection between the tracks is emphasised by the fact that some topics will be important in research within all the tracks in the initiative.
We will highlight three examples of such cross-cutting themes:
- Trust: In society, in the technology, and in the technology and services created using it ensure.
- Accountability: Responsible, including safe and non-discriminatory technology, and responsible and sustainable technology use
- Creativity: Innovative and ground-breaking research. Innovative development and use of technology. Creative AI-based problem-solving methods for innovation and new creation.
In order to take advantage of the links between the three tracks, interdisciplinary research is needed, which combines technology, social sciences, law and humanities.
However, ground-breaking research is not always interdisciplinary. We also need focused research from many different disciplines, and tools that ensure that knowledge from these is available across subjects and tracks. This will be relevant in all tracks.
For the investment to have an effect and have a clear footprint, it must have a clear direction. At the same time, the investment must be sufficiently open to opportunities and issues we are not yet aware of. Below, we will describe the status and opportunities within the three tracks.
1 Consequences of artificial intelligence
1.1 Areas in need of research
Artificial intelligence is being used in almost all areas of society and the list of research needs is inexhaustible. This applies to both research on AI that has been put into use, follow-up research when AI is put into use, and research into mechanisms that are relevant when new technology is to be put into use. We have therefore identified some overarching themes with examples of sub-themes. We emphasise that this is not a complete overview, and that new areas will appear in the 5-year period.
Trust in society
The Nordic societies are characterised by a high degree of trust. This applies to both interpersonal trust and confidence to social institutions and administration, but also trust in technology. It is important to gain more knowledge about how artificial intelligence and the use of artificial intelligence affect trust (and the basis for trust).
Examples of subtopics are security, privacy and surveillance, information and disinformation, freedom of expression, democratic processes, influencing elections and voter turnout, AI-supported decision-making systems, ethics, and regulations and frameworks for technology use.
Justice, diversity, language and culture
AI reproduces biases in the training data sets. Generative AI can provide this problem on an even larger scale, not only in the form of discrimination on the basis of e.g. gender, skin colour or place of residence, but also because Norwegian data is only to a limited extent included in the training datasets for large models, whether it concerns language and images, or medical data.
We need research that uncovers such discrimination, and how AI affects Norwegian and Sami language and culture. At the same time, we need humanistic and social science research that can ensure that ethical assumptions are built into the technology and not just become an afterthought.
Labour and business
Working life is changing with the introduction of new technology, and artificial intelligence can contribute to major changes in the labor market and the content of various occupations. Examples of sub-topics are technology’s impact on job creation and loss, changes in professions, work processes, management, business models, competence needs in business, health and the public sector, and the need for further and further education, as well as the sustainable use of AI, and what requirements should be set to transparency.
Knowledge and creativity
Technology enables new ways of learning and expressing oneself. Artificial intelligence can be used to obtain and process information, for large-scale analyses, and for creating and creative processes. There is today considerable discussion about the use of digital technology in schools. Artificial intelligence will bring increased challenges, but also new opportunities. Examples of sub-themes are the individual’s need to master technology, new opportunities in art and culture, creativity, and artistic development.
1.2 Subjects where Norway and Norwegian research environments have an advantage
Since the 1990s, Norway has had significant research investments in ICT and digitalisation.
Professional environments have been built up with strong expertise in the importance of technology for society and culture, for the public, democracy, working life and education. And we have considerable research into the regulation of technology, as well as how technology use, user participation and culture contribute to shaping the technology.
There are strong traditions for interdisciplinary collaboration between technological subjects and humanities and social science subjects in Norway.
Many of these professional communities are at the forefront internationally in their areas. These professional communities are now orienting themselves towards artificial intelligence, and Norway has the opportunity to consolidate its position internationally in research into the consequences of artificial intelligence.
2 Artificial intelligence as a research area in itself
2.1 Areas where research is needed
AI is technology in rapid development. Much of the technology used today is immature and not very transparent. It may lack robustness, has insufficiently quantified uncertainty, often fails to generalise across domains, relies on massive amounts of curated training data, and is energy-intensive in both training and use.
At the same time, we see that technology’s potential to contribute to necessary and important innovation and development is great (cf. part 3). It is therefore crucial that the technology is further developed. The research effort should contribute to a radical transformation of AI to make it (1) more accurate, (2) take uncertainty into account, (3) responsible and explainable, and (4) more sustainable.
More accurate
If we are to have reason to trust artificial intelligence, it must in several contexts be sufficiently accurate and provide answers we have reason to trust. This applies whether it is to be used for decision support, analysis or against other technologies. Central research areas for more reliable and accurate artificial intelligence are: Data curation, including good data bases for machine learning, methods for transforming data into information, for example by combining ‘semantic’ methods or other knowledge (physics and mathematics) with machine learning methods and statistical models, as well as the possibility of measuring model quality through the transparent use of meta-data.
Precision in AI and the “reliability” of the model will vary by application. For AI-generated art, for example, precision can mean creativity and the ability to follow a certain artistic style, but also the ability of the models to support creativity and experimentation. In predictions, such as precision medicine, weather or market analysis, precision refers to the accuracy of predictions compared to actual results.
Takes uncertainty into account
Data are often incomplete and noisy; models are often not accurate enough, and algorithms only approximate exact solutions. Therefore, estimates, generalizations, predictions and decisions produced by AI are inherently uncertain. To provide the confidence needed to inform decisions, uncertainty about outcomes must be quantified. This will in turn increase the reliability of AI based recommendations. It is also important that quantification of uncertainty reveals disagreement between data and knowledge, as well as between different data sources. Relevant research areas are: uncertainty quantification, statistics, robustness guarantees, models with hard and soft constraints, error margins for autonomous systems, user interpretation, digital humanities (e.g. critical dataset studies), and philosophy of technology.
Responsible and explainable
Artificial intelligence that is implemented in society must be transparent and explainable to a far greater extent than today. We need to know more about how it actually works (data base, models, algorithms and epistemology), what it can be used for, and what it does not work well for. AI must function securely, as well as reflect a Norwegian context with Norwegian values. To achieve this, a high degree of user participation is also needed in the development of the technology. Important research areas are for example: Explainability, security challenges, language technology for Norwegian text and speech, alignment of models, synthetic data generation and privacy, norms and values, discrimination and bias, and analysis of connections between training data and output.
More sustainable
Artificial intelligence requires enormous amounts of energy to be effective. Much of the work within AI and machine learning in particular uses brute-force with a focus on ever larger models with ever more data. Getting roughly the same results with smaller models and data usage is central to the use of AI being more sustainable. Key research areas to succeed in this include tinyML, knowledge-based ML, energy-efficient processing including new calculation methods (for example NeuroAI and neuromorphic data processing). Because storing data also uses energy, new methods are needed to reduce and compress data, without major loss of information.
2.2 Subjects where Norway and Norwegian research environments have an advantage
Norway has strong professional environments within digital technologies and artificial intelligence. Some areas where Norway has invested over time are machine learning, theoretical informatics, statistics, logic, applied mathematics, energy efficient processing including new calculation methods, image processing, theory of learning. Here there are also opportunities to make breakthroughs in interdisciplinary collaboration with humanities, social sciences and legal subjects, and domain-specific problems within biomedicine, physics and electronics. Another area with a long tradition is language technology, which is important for developing and understanding future language models.
The cyber security environment has extensive experience in handling complex data and systems. Both machine learning and more traditional symbolic AI require a structured understanding and follow-up of data, where Norway also has strong environments. Of new areas, we see an emergence of environments that work with explainable AI (XAI), knowledge-based AI (incl. hybrid models), as well as NeuroAI, which springs from the strong neuroscience community in Norway.
3 How can artificial intelligence be used for innovation, including in research
3.1 Areas with a need for research
There is great potential for using artificial intelligence for innovation in the private and public sector, including research. We will illustrate the need for research by showing some examples of opportunities within these sectors. Next, we will give some examples of cross-cutting needs.
For innovation in business — green and digital transformation
Innovation in the business world is crucial, among other things, for Norway to achieve a green and digital transformation.
AI will be necessary for transformation in a wide range of industries, from aquaculture and maritime industries, to process industry, energy production and distribution, construction, battery/hydrogen, transport, telecoms, logistics, media and creative businesses.
Research is needed on how AI can be used in innovation processes to reduce energy use, material use, pollution or to create new value chains (for example based on circular economy), services and products.
Furthermore, research-based insight into how AI can provide innovation can contribute to important restructuring of workflow, organization and process changes.
For innovation in the public sector — health and welfare
The public sector has a very large need for innovation and AI and digital technologies can play an important role in this. The Norwegian Health Personnel Commission (NOU 2023: 4) illustrates this very well.
The need for health and care services is increasing as a result of demographic changes. In many areas, from diagnostics to rehabilitation and follow-up, AI can provide significant gains and better services.
The need is great, and here both follow-up research and further development of AI and other digital technologies are needed which also take account of domain knowledge and the context in which the technology is to be used. Similar examples can be given for other parts of public administration.
For research
AI has great potential for breakthroughs both in medicine, the natural sciences, the humanities and the social sciences. Although it is within natural science research that we see the clearest results so far (for example AlphaFold), AI is moving at full speed into research in all subject areas. Aspects related to data curation and process support are central to increasing the replicability of the research.
A better understanding of the connection between training data and results is needed, both for noisy and incomplete data, and for unstructured data such as text and images. It is important to develop a better understanding of which methods work on which problems and types of data.
There are many examples of machine learning being used in the wrong ways, without a good tradition of disseminating this knowledge. Language models and the fact that AI has become so easy to use increase the risk of such errors.
Transversal needs
In order for the potential of using AI for innovation in business, the public sector and research to be realised, the cross-cutting themes mentioned at the beginning are important: We must have trust in the technology and how it is developed and put into use. The technology must be and be used responsibly and ensure safety in a good way. And we must use technology in such a way that it promotes creative problem solving.
Furthermore, as explained in Part 2, the technology must be accurate, sustainable, accountable and explainable, and take uncertainty into account. In addition, transparency around processes, such as the role technology plays in decision-making or production processes, is crucial. AI is increasingly becoming part of a central interaction system where humans interact with machines in constantly new contexts for, among other things, decision support, art, creativity, health and social relations, which creates a need to develop the knowledge within system development and information systems for how to adopt new technology in organized business.
3.2 Subjects where Norway and Norwegian research environments have an advantage
Norway has strong research environments that are well suited to realise the needs for innovation in the public sector and business mentioned above. In addition to what is mentioned in parts 1 and 2 of the note, we would like to highlight two matters in particular:
Firstly, successful innovation work requires user involvement. Norway has professional environments with strong expertise in including people in the development of technology.
Second, successful innovation depends on rethinking work processes and business models. Norway has world-leading environments in system development and the business value of IT.
The areas of application that are particularly important for Norway to develop can further be divided into two main categories. First, we have the areas where Norway either already has a business community that should be able to draw on research, or where there is reason to believe that such a business community can be developed. Health, clean energy, aquaculture and maritime industries are examples of this.
In addition to this, there are areas where we do not necessarily have an advantage, but where it is crucial for Norway that we develop our own knowledge and expertise. Norway needs autonomy in digital security. The public sector and Norwegian language support are areas that are important without us having an advantage internationally. Resource optimisation, transport/logistics and intermodality are other examples of such areas.
4 Objectives of the venture
The venture together with additional activities should result in:
- Research excellence (forskningstyngde): Norway is clearly placed on the map as a research nation with heavy AI competence. Research and new methods contribute to handling challenges with advanced AI, in the light of Nordic values.
- Cross-cutting collaboration: Increase the number of interdisciplinary projects and collaborations that cross sectors and disciplines to address complex AI-related issues.
- Competence development: Results from research and application are taken into teaching, both by technologists, but also in all other fields.
- Knowledge environments: Collaboration is developed between research and practice fields that facilitate good application of research results.
- Security: Strengthen digital security, including resistance tests in AI systems.
- Innovation: Implementation of AI-based solutions in business, health and the public sector.
- Public information: Basis for informative resources for the general public that increase awareness of AI consequences.
- Environment and sustainability: Concrete examples of how AI has contributed to solving environmental problems or improving the management of natural resources.
5 Footprint of the investment in 2030
We describe here a rather ambitious footprint. This will not come from the AI billion alone, but by the AI billion setting a direction and being used to build up research-based knowledge that can attract other funding from public and private actors in the same direction.
The ambition for using the billion for artificial intelligence must be long-term, fruitful, ground-breaking and interdisciplinary:
- Long-term: The effect of the billion goes far beyond the next 5 years and the billion itself, and is in the connection economically sustainable.
- Fruitful: The effect of the billion brings benefits far beyond research. In addition to research results, the research contributes to the constructive and regulated application of AI, especially in health/welfare, the public sector, and to a green transition in the business world.
- Groundbreaking: The AI billion produces important, ambitious and influential research that prioritises the core challenges and places Norway on the map of AI research-heavy countries. Ground-breaking research takes place in international collaboration and Norwegian researchers are helping to solve a grand challenge in science with the use of AI.
- Interdisciplinary: The results of ground-breaking AI research must benefit society and people in general, and contribute to a fairer and more sustainable world. This can only be achieved through interdisciplinary research.
In total, the investment, together with adjacent projects, will place Norway as a central AI nation.
In selected areas, Norway must be recognised as a driving force at Nordic and European level, and through this in the rest of the world.
This should make Norway more attractive to international AI researchers, both as a place of work and as a partner in international projects. Greater recognition should also contribute to us having a political influence on the direction of AI, also internationally.
This is also part of my personal project #1000daysofAI and you are reading article 521. I am writing one new article about or related to artificial intelligence for 1000 days. The first 500 days I wrote an article every day, and now from 500 to 1000 I write at a different pace.
