• May 30, 2024
• admin

TechSweden releases a report on the present and future of AI in Sweden. You can read the full report here.

Background

In March 2024, TechSverige arranged three roundtable discussions on Sweden’s role in the ongoing AI development. These discussions brought together leading experts and key stakeholders to discuss three perspectives: AI development, AI use and the AI ecosystem in Sweden. Each conversation had a focus on examining Sweden’s strategic position to build competitiveness in AI and the needs this entails. The conversations were conducted according to the Chatham House Rule, which means that the information can be shared, but no statements are attributed to individual participants.

This report summarises the three round tables. If one aspect was only dealt with in one conversation or if one group deviated from the others, this is reported explicitly. In addition, the report provides an overall summary of the issues discussed and a broad consensus on Sweden’s opportunities and challenges in AI development. Neither the report nor the talks give a complete picture of everything that is happening in the area, but they identify important issues to work on further.

The talks were led by Joakim Wernberg, research leader at the Swedish Entrepreneurship Forum and lecturer in Technology and Society at Lund University. The list of participants can be found at the end of the report.

Summary

A common insight from the conversations is the need for strong leadership to drive and coordinate AI work both within organizations and at the national level. It is stressed that AI should not be reduced to a matter for the IT department, but integrated into broader political and economic strategies.

Sweden’s Strategic Advantages

Sweden has two significant strategic advantages in the AI economy: access to high-quality data and the ability to develop local and specialised AI applications. The overall development of AI has been driven by three factors: computational capacity, data and algorithms. With current technical limitations, it is important to focus on qualitative data and efficient models that require less data and computational capacity.

The Importance of Data

Data is considered more important than the AI models themselves. Sweden’s public register and the principle of public access to official records constitute a strategic resource, especially in health, education and social care. Local, high-quality datasets can also be used to create synthetic datasets for AI training. One obstacle, however, is the interpretation of data protection rules such as the GDPR, and Sweden’s slow progress with open data indicates additional challenges.

Competition through Specialized Applications

Swedish companies can compete by developing local and specialized AI applications based on larger models. These applications can be scaled up and reach a wider market, similar to how apps leverage smartphone technology.

Digital Transition in the Public Sector

The public sector has been slower in its digital transformation compared to the rest of the economy and other countries. Legal and technological fragmentation between public activities is a significant barrier. Regulatory harmonisation and better interoperability are necessary to promote the use of AI.

Skills needs

The report identifies three types of skills needs: users, rule interpreters and builders. User and legal competences to manage AI regulatory frameworks are particularly important. Practical technical expertise is also crucial for implementing and fine-tuning AI systems.

Introduction

AI development is in a crucial phase where technological innovation, use and regulation of technology are simultaneously changing at a rapid pace. This dynamic creates opportunities for business leaders, government heads and policy makers to identify strategic resources and strengthen the competitiveness of the emerging AI economy. At the same time, the outcome depends on how these three areas—technological development, use of the technology and its regulation—interact. Developments in one of these areas are likely to affect the other two, underlining the importance of an integrated approach.

Summary
All roundtable discussions emphasized the need for leadership to drive and coordinate AI work, both within individual organizations and from the government. AI and digitalisation must not be reduced to the responsibility of the IT department. Instead, better policies are needed that make use of AI in all areas.

Sweden has two potential strategic advantages in the AI economy: the availability of high-quality data and the ability to develop local or specialised AI applications. Developments in AI have depended on three factors: computational capacity, data and algorithms. Now, progress seems to be peaking, increasing incentives to build AI models that require less data and computational capacity.

Data is more important than AI models. Public registers and the principle of public access to official records are strategic resources for Sweden, especially in health, education and social care. However, barriers such as the GDPR and the lack of open data hinder its potential. Swedish companies can compete with local or specialised AI applications that build on large models. These applications can be scalable and reach a larger audience, similar to apps that use the capabilities of smartphones.

The digital transformation of the public sector has been slower than the rest of the economy and compared to other countries. Legal and technological fragmentation between public activities hampers interoperability and data sharing. Promoting experimentation with new technologies in the public sector is important.

The Round Table identified three skills needs: users, rule interpreters and builders. The lack of technical user competences and legal competence is greater than the lack of technical excellence. In addition to cutting-edge expertise, craft skills are needed to implement and fine-tune AI systems, for example through technical industry upper secondary education.

User competence is crucial and extends across the economy. The need for data literacy is central. Legal skills are needed to enable the use of AI. The problems with open data and GDPR in the public sector demonstrate the need for harmonised rule interpretation and resources to lower legal thresholds. Many companies, especially SMEs, lack the legal skills to use AI technologies. Increasing regulatory frameworks can be a significant obstacle to the development of AI.

Sweden has the potential to develop a strong domestic AI ecosystem, but there is untapped potential in collaborations beyond the country’s borders at Nordic or Nordic-Baltic level.

Technology

Over the past two decades, developments in machine learning, especially deep learning, have made great strides and become a driving force in AI. Recently, major AI models, especially language models such as ChatGPT, have attracted a lot of attention globally, affecting politicians, business leaders and the general public alike.

However, it is important to understand that AI is a broad concept. Many AI technologies are more specialised and less visible than generative AI, but their economic impact is significant. The AI concept has been around since the 1950s, and many older technologies work similar to today’s AI by using data to adjust their behavior, such as control technology and optimization developed in the early 1900s.

Despite this, there are important differences with today’s AI technologies. A key difference is that modern AI models, such as ChatGPT, generate responses in a way that even their creators cannot explain in detail. This “black box” issue places high demands on users’ understanding of both AI
capabilities and limitations. AI’s
ability to detect statistical patterns that humans cannot predict has great potential for growth and prosperity, exemplified by DeepMind’s model for predicting protein folding.

Understanding the
impact of AI on the economy and society requires specifying which technologies to study in order to avoid inaccurate comparisons. Studies on work automation due to AI are often affected by which AI technology was up to date at the time of study. We need empirical studies and theoretical discussions to understand how AI’s
unique ability to perform analytical work will affect society in the long run.

AI is basically software (which can be integrated with hardware and robotics) and part of the broader digitalisation of society. This brings great potential for innovation, growth and prosperity. However, organisations and societies that are not adapting to this digital transformation risk missing out on the full potential of AI technologies, both now and in the future.

Use

On the surface, it may seem that ChatGPT suddenly appeared, but the development of major language models has been going on for a long time. What was new with ChatGPT was the user-friendly interface that made the technology accessible to everyone. Because it is a software-based and data-driven service that spreads over the internet to already existing computers and mobiles, it could be quickly integrated into the economy and society.

When new technologies such as generative AI tools are distributed via existing devices, the technology becomes accessible to more people and its use spreads from the grassroots upwards, not just through decisions from higher levels. The increased availability creates opportunities for more organizations to experiment with the technology, but also poses challenges.

History shows that technology investments alone do not automatically lead to increased productivity. Working methods, organisation and sometimes business models need to be adapted to take full advantage of technology. Statistics measuring digitalisation often focus on technology investments, which does not provide a complete picture. Investments in technology are known costs, while organisational adaptations and innovations are often underestimated.

It is a challenge when employees gain access to new AI tools via their mobile phones while their workplaces need to invest in data-driven innovation and skills development. It is about integrating AI tools into working methods and processes within organizations and according to current regulations.

AI can complement human work, but what this division of labour should look like is not self-evident. Studies show that doctors who use AI to assess X-ray images can be affected by their expectations of AI
With generative AI and language models, there is a risk that people overestimate the technology because it can feel like the machine is thinking like a human.

In general, larger companies are faster at adopting new technology than small businesses, but technology-driven startups are often faster at taking advantage of new technology than large established companies. This pattern also applies to digitalisation. Large companies with over 250 employees have progressed further in their digital transformation than SMEs and were early in investing in AI. Statistics Sweden’s
surveys from 2019 to 2021 show that the use of AI increased most among medium-sized companies with 50-249 employees.

AI tools delivered as cloud services or via digital platforms are more accessible to SMEs as they do not require large investments in own hardware. At the same time, technology is evolving rapidly and competition for the skills needed to leverage AI tools and manage data is fierce. Smaller companies tend to wait until the technology is standardized and the risk of failure is lower, in part because of their smaller financial margins.

In the public sector, the challenges are different, as operations are governed more by rules and laws. When digitisation statistics began to include public digital services, Sweden’s position in international rankings fell. This is partly due to the fact that large organisational investments are required to provide efficient and legally secure public services.

Two lessons from digitalisation can be applied to AI development: Firstly, AI should be integrated into existing policy areas such as the labour market, social services and education, instead of being treated as a separate policy area. Secondly, new regulatory frameworks need to be interpreted and applied consistently across different authorities. GDPR has resulted in varying interpretations, which is costly and difficult for smaller municipalities to handle.

It’s important to remember that AI doesn’t automatically improve everything. In 2010, historian Nina Wormbs described the “digital imperative” – the belief that everything will be better if it is digital. This insight should be taken into account when navigating future AI technologies. AI is a tool for achieving other goals within organisations and society, not an end in itself.

Policy and AI regulation

There is a lot of development in politics, but regulating AI is challenging because AI is a broad concept that includes both new and older technologies. The EU has adopted an AI regulation to be implemented in all member states. In parallel, work is underway within the OECD and the UN to establish AI-related regulations that will also affect Sweden. These regulatory frameworks are horizontal, designed to cover all uses of AI, although the EU
AI Regulation divides certain areas of application into different levels of risk that are regulated separately. This means that in addition to the implementation, the regulations must be integrated with existing industry-specific rules.

whereas the AI Regulation and similar regulatory frameworks are not technology-neutral, creating questions of interpretation and the risk of biased incentives in the economy; The adoption of the Regulation does not mean that the process of establishing an effective regulatory framework is over, but rather that it has only just begun.

Nationally, Sweden has appointed an AI Commission and has a Productivity Commission that will investigate how Sweden and Swedish companies can benefit from digitalisation and AI development to strengthen competitiveness. In addition, Sweden’s national digitalisation strategy will be updated in 2024.

The policy approach is about both regulating and promoting AI development. Regulatory frameworks can prevent unwanted phenomena and create reliable rules that promote transition, innovation and entrepreneurship. The discussion on responsible and ethical use of AI has dominated public debate, but it is also important to highlight the potential values of technology. Many countries are developing their own AI strategies so as not to be left behind. The question is what we have learned from digitalisation so far and how we can avoid repeating old mistakes while leveraging new AI solutions to accelerate the digital transition.

The Way Forward

The rapid development of AI technologies, their use and policies are mutually affecting all these areas. This creates great opportunities to influence development, but it is also a challenge to meet the right goals in a changing environment. That’s why it’s important to be humble about forecasts, future skills needs and uses for AI. Flexibility may be more important than detailed planning, while the market needs stable rules despite an uncertain future.

Through roundtable discussions with experts from different sectors, TechSweden has promoted knowledge exchange and dialogue to discuss Sweden’s strategic position in AI. These talks do not aim to provide definitive answers, but to formulate better questions for future work. I thank TechSweden and all participants for their efforts. Hopefully, these discussions mark the beginning of a broader exchange around AI development.

Sincerely,

Joakim Wernberg
Research leader at the Swedish Entrepreneurship Forum, Senior Lecturer in Technology and Society at Lund University and Head of Research Group for Socioeconomic Technology Studies (SoeTech)

1. Leadership with AI as a Means

Leadership as a catalyst

The need for strong leadership to drive and coordinate AI work, both within organisations and at political level, was highlighted in all roundtables. Experience from previous digitisation has shown that too much focus has often been put on technology, which has delayed the transition. AI technology is accessible to everyone with the internet, creating both opportunities and needs in many different areas. Leadership is not about having all the answers, but about acting as a catalyst for AI work across society. In addition to strong leadership, the participants highlighted the importance of good examples, ambassadors and significant reforms to promote the use of AI. The question is whether there is a “home PC reform” for AI.

AI as a Catalyst

There is broad agreement that Sweden must act to strengthen its position in the rapid development of AI, but there is not the same agreement on why this is important and how it should be implemented. Political leadership is needed in the field of AI, but the focus should be on how technology can be used to create value in other policy areas rather than on technology itself. Leadership must understand AI’s
potential and limitations, and prioritize resources to test, learn, and implement AI across different activities and policy areas. AI policy must not be isolated as previous IT and digitalisation policies have often been.

AI is not an end in itself

AI leadership needs to be anchored at several levels, up to the level of government. Important questions to answer are what society we want to build and how AI can help realize that vision. For example, a national data infrastructure and strategy for data-driven innovation can help solve the healthcare crisis, develop education or promote Swedish exports. Sweden does not need a separate AI policy, but better policies using AI in all areas. Every technology initiative should be anchored in concrete expectations of how it can contribute to a better world. One way forward could be to focus policy efforts on specific areas or functions such as information and cybersecurity, and to integrate AI as a natural part of all policy areas.

A home PC reform for AI?

The home PC reform became a symbol of the early computerization in Sweden, which led to many households getting computers that children and young people used to play games, learn to program and practically use the technology. That generation grew up with technical skills that later contributed to innovation and entrepreneurship. However, this result was not the original intention of the reform.

During the roundtable discussions, what could correspond to the home PC reform for AI was discussed. If the goal is to increase access to AI technology, large language models and services such as Chat GPT can be seen as a solution. These models weren’t developed overnight, but Chat GPT introduced a user-friendly interface that quickly spread the technology. AI is more than generative AI tools, and as people experiment with these, they gradually learn more about the technology.

For skills-enhancing initiatives, there are initiatives such as AI Competence for Sweden and courses such as Elements of AI that make the subject easily accessible. Unlike the home PC reform, skills needs in AI are not as homogeneous across the population. A challenge with AI is that applications are often like “black boxes”, meaning that their outputs cannot be predicted, but must be understood in order to use them effectively.

Another perspective is that there may not be a direct equivalent to the home PC reform for AI. Instead of focusing on AI as an end in itself, AI should be used as a means to achieve other goals. This means that AI should not be isolated as a separate policy area but integrated into all policy areas to foster innovation and growth.

2. Sweden’s competitiveness and strategic advantages in the AI economy

Finding their place in emerging value chains

New value chains around AI, in particular generative AI driven by large language models (LLM), are taking shape. What these chains will look like is still unclear. OpenAI is creating its own app store for its language models while Meta has opened its language model and companies like Google and Microsoft are integrating language models into their product portfolios. Generative AI is only one part of the broader AI technologies used in the economy.

Developments in AI have been driven by three factors: computational capacity, data and algorithms. Over the past 20 years, the development has mainly been due to increased access to training data and computational capacity, but now we are approaching a point where computational capacity and its energy consumption become new bottlenecks. This creates incentives for developers to create AI models that use less but more high-quality data and/or require less computational capacity.

For Swedish innovators, entrepreneurs and companies, there are two important aspects to taking a strategic position in these new value chains. Firstly, AI value chains are data-driven, and secondly, much of the value creation lies in locally adapted or specialised applications. Therefore, Sweden’s competitive advantage can be found both in managing data as a resource and in developing local and specialized AI applications.

This chapter addresses issues around data as a resource, the development of local and specialised AI applications, and the ecosystem around the actors developing AI technologies and new applications in Sweden.

Data is more important than AI models

Data plays a bigger role than AI models in many contexts. AI relies on technical and institutional structures to manage, share and use data effectively. Even when AI models are used for specific services, they often need data from users. whereas AI is more commonly used as a multi-party data-driven service, reducing the autonomy of individual organisations but creating more intertwined economic relationships, similar digital platforms and cloud services;

It is important to have good conditions for collecting data and maintaining data flows both within and outside Sweden in order to maximize the benefits of AI applications. Internet data used to train large AI models is often of low quality and is expected to deteriorate over time due to machine translations and other errors. This can lead to an increased demand for small but high-quality datasets for specific applications.

Local and specialised datasets can be used to create synthetic datasets for AI training. It is no longer self-evident that data volume or computational capacity will drive AI development forward. Instead, computationally efficient AI tools based on smaller but more qualitative data can become more important.

Policy regulation that cannot clearly define AI technologies creates uncertainty in the market and can lead to skewed incentives. By focusing more on data as a resource rather than specific AI technologies, policy initiatives can become more technology neutral and foster a diversity of technological solutions.

Access to high-quality data is a significant strategic advantage for Sweden. Sweden’s extensive public register and the principle of public access to information provide opportunities to provide high-quality datasets in areas such as health, education and social care. However, Sweden has lagged behind in the digitalisation of the public sector and the availability of open data, which needs to be addressed in order to realise the potential competitive advantage. Other countries have now caught up or overtaken Sweden in these respects.

Importance of Local and Specialized AI Applications

All groups stressed the importance of local or specialised AI applications, especially for generative AI. These applications mean that more general AI tools are adapted to local conditions or further developed for very specific uses. The economic value of AI is likely to require local adaptations and specific datasets to meet quality requirements. These applications often occur in short value chains with few actors, where much of the value lies in the adaptation work itself.

The group that focused on development issues noted that many companies that want to use AI internally will need to develop adaptations of general applications to suit local needs. Therefore, it may be strategic to focus on developing local and specialised AI applications to increase both the supply and uptake of AI-enabled services and products.

Specialized applications may have narrower uses but are often scalable and can reach a wider audience. This is similar to the development of apps and software programs that leverage the general capabilities of smartphones to deliver specific services. The availability of high-quality data in specific areas is an important resource for such development work.

Swedish companies can compete with large technology companies by creating local or specialized applications based on large AI models. This is similar to how Google, Amazon, Meta and Apple lowered the thresholds for new start-ups to develop and scale up digital services. Studies show that innovation in the economy has become more software-dependent, partly due to the availability of technical building blocks. But regulation of data and digital markets can be a bottleneck for future entrepreneurs and innovators.

Custom AI Solutions and Ecosystems for AI

Both groups focusing on AI usage and ecosystems stressed the need to develop opportunities for testing and experimentation in addition to the current regulatory sandboxes. Specifically, it highlighted the need to experiment within existing value and supply chains as well as through collaboration between small and large companies, or between public purchasers and private suppliers. This need arises in part because many practical applications must initially be adapted to local conditions rather than being tested in a standardised form that can later be widely applied. In addition, needs differ, for example, between technology-driven start-ups, existing small businesses and large companies.

The group focused on the use of AI underlined the importance of enabling SMEs to experiment with and implement AI tools in their operations at an early stage. The history of digitalisation shows that the pace of transition has been uneven, with smaller companies often showing low digital maturity in terms of technology investments and changing ways of working. This is due, among other things, to their small margins, limited resources and high thresholds for taking advantage of new technologies early on. At the same time, the group of SMEs is very heterogeneous, including both innovation-driven start-ups and established small businesses.

It is likely that we will see the same pattern of uneven transition for AI tools. If the objective is to promote the use of AI among these companies, their needs will vary greatly, and there is no universal “AI lift” that works for everyone. Therefore, general measures (horizontal, rather than vertical technology-specific or industry-specific measures) are needed that lower the thresholds and make it easier for smaller companies to experiment with a wide variety of new technologies for different purposes. If successful, such measures can foster both an increased supply of advanced AI applications and a growing demand for these applications.

3. Digitalisation and AI in the public sector

Starting uphill or a chance to catch up?

How the public sector adapts to and benefits from AI and digitalisation is crucial for Sweden’s position in AI development, as well as for the country’s competitiveness, growth and prosperity. This involves both streamlining public services and acting as a demander and client of new technology and innovation.

Experience so far shows that the transition has been significantly slower in the public sector compared to the rest of the economy. Sweden has fallen in international comparisons and is now below the average in the OECD
Digital Government Index. This backlog is partly due to the stricter regulation in the public sector, which makes experimentation with new technologies and changes in working methods more difficult. Although Sweden had a leading position in the early stages of digitalisation, this lead has fallen sharply.

The question is whether Swedish authorities, regions and municipalities are starting to face headwinds due to the slow pace of digitalisation, or whether this is an opportunity to catch up. The roundtable discussions have primarily highlighted external perspectives on digitalisation in the public sector, focusing on the problems that arise in relation to public services and activities.

Technical and legal fragmentation

There are important lessons learned from how the public sector, especially regions and municipalities, manages and implements GDPR. Interpretations vary widely, leading to inconsistent data management and requirements. Within the same municipality, there may be different interpretations of GDPR. This fragmentation reflects the well-known technical problem of incompatible IT systems within the same public entity.

It is impractical and inefficient that each public organisation independently interprets regulations for digital services and data management. This need is further reinforced by new regulatory frameworks on the horizon, making centralised, harmonised interpretation of rules and advice in the public sector necessary. Such a function would streamline public activities, promote the exchange of experience and facilitate interaction with private actors.

Increased interoperability and harmonisation of technical data management in public activities is also crucial to foster data sharing and data-driven innovation. Although it is a known problem, the Round Table stresses that it is difficult to solve. IT infrastructure varies between and within public services. One proposal is to introduce common data standards on top of existing infrastructure, inspired by Estonia’s X-roads digital platform architecture and Government Cloud. This would prevent public entities from creating their own, non-compatible AI systems.

Another aspect is the difficulty of maintaining client competence in public services. Customers have expert knowledge of their own operations, but it is difficult to quality assure external suppliers technically and legally. This problem is exacerbated by new regulations in the field of AI and data. One proposal is to introduce a common function for certifying suppliers, which would strengthen client competence, increase trust and make it easier for small and medium-sized enterprises to supply public services, without increasing market thresholds.

Easier to experiment

In order to promote experimentation with new technologies in public services, it needs to be simpler and more justified. In addition to managing regulations, technical systems and customer expertise, internal incentives are required to test new technology, learn and gradually adapt the business. Skills and mandates for change management at several levels are needed. The Swedish Association of Local Authorities and Regions (SKR) can act as a central point for promoting learning and the dissemination of good practice.

Often there is a fear of mistakes in the public sector, which hampers innovation. It must be accepted to make small mistakes early to avoid bigger problems later. One example is the conflict between the City of Stockholm and the Open School Platform. Although opinions vary in the specific case, it is clear that the interests of the parties should have been compatible from the outset. This case highlights the importance of collaboration and learning by mistake.

4. Skills needs and supply

Builders, users and rule interpreters

whereas AI technologies can facilitate or automate elements of analytical work, changing the way work is organised both in the workplace and in society; The discussion of AI
impact has evolved from focusing on what jobs are lost or created, to highlighting how all types of work will change. This requires extensive skills development in order to use new technologies and adapt to new ways of working.

The Round Table identified three areas of competence: users, rule interpreters and builders. Participants stressed that technical user competences and legal competences to manage data and AI regulatory frameworks are more critical than technological excellence. This is reflected in the expression “AI does not take your job, but a person who can use AI may do so.” Organisational investments and innovations are crucial for realising productivity gains from new technologies, rather than technology investments per se.

Technical skills are needed, but the debate about the need for more university-educated engineers needs to be nuanced. What is in demand is practical technical expertise to implement and adapt AI solutions, which does not always require university education.

The discussion on technical excellence, user competence and legal competence is further developed in the following sections.

Technological excellence

The public debate has strongly emphasized the need for technical expertise and civil engineers with knowledge of AI. In all three groups, however, it emerged that the need for university-trained technical excellence is important but not sufficient, and that the debate needs to be nuanced.

A common misconception is that technical excellence is only about more years at university. In reality, practical technical skills are in demand, which means the ability to implement and fine-tune systems based on existing AI technologies. This competence, called “builder”, lies somewhere between a “prompt engineer” and a master’s degree in computer science. For example, technical upper secondary education in industry has proven to be good at imparting this practical technical competence.

This type of practical competence is also needed to be able to utilise the research expertise that already exists in the field of AI in Sweden. A recurring example in the discussions is the Wallenberg Autonomous Systems and Software Program (WASP), which will produce one researcher per week during its remaining project period. In order for these researchers and advanced developers to contribute effectively, whether by implementing existing solutions in companies or starting new companies, more people with building skills are needed.

There is still a need for technological excellence at university level. However, the discussions showed that the need for technical excellence is significantly less than the need for user and legal skills right now. Moreover, the need for technological excellence cannot be met by training more engineers in universities alone. Practical and artisanal building skills linked to digitalisation and AI are needed.

Another important point raised in the group that focused on AI development is that the need for technological excellence decreases over time as the technology matures and becomes more accessible as standardized products and services. Chat GPT’s
user-friendly interface has made advanced AI technology available to more people. This also means that AI usage is increasingly entering organisations through their employees.

User and Legal Competence

Even more important than technical competence is the knowledge that enables experts in other fields to use AI technology in their work. The need for such user skills is great and extends across the economy, making it far more important than the need for technological excellence. Experience so far shows that user competences have often been underestimated. One group stressed the importance of training courses for teachers and doctors, for example, including AI technology, so that students learn to use the tools practically and become comfortable with the technology before they start working. This contrasts with how digitalisation has been handled in primary schools, where teachers’ user skills have often developed reactively rather than proactively. Another group stressed that researchers in fields other than computer science also need to learn how to use AI tools to streamline their research. All groups stressed the importance of college and university students being able to easily adapt their education so that, for example, a journalist student can take a course in AI and an engineering student a course in journalism. Both competencies are important.

A central part of user competence is to understand and manage data as a resource (data literacy), which is often more important than the AI models themselves. The need for user competence is great in the public sector, partly because of the regulated ways of working, and partly because the public sector is an important client of AI.

There is also a strong need for legal expertise to enable the use of AI. The public sector has shown that there are major challenges with the interpretation of GDPR and other regulations. Many companies, especially smaller ones, lack the legal skills to ensure how they can use AI. All groups highlighted the need for a central authority that can reduce legal uncertainty by providing broad information, acting as a knowledge resource, developing standardised documentation and providing advice in individual cases.

Discussions on AI in primary and secondary schools identified two main reasons for including AI in education: to teach students to use technology and to understand its impact on society. Making AI a topic of its own is not necessary, but the technology should be integrated into existing topics. Within STEM subjects, AI tools can facilitate learning, while the societal impact of technology can be addressed in social studies and similar subjects.

Finally,

We are facing a transformation comparable to the internet revolution of the 1990s or the rise of electricity. In this context, it is important to be humble in the face of our limited knowledge of the future, including in the short and medium term. That is why we need dialogue and responsiveness, which these round tables represent. Adaptability is more important than long-term planning.

The views expressed here reflect an ongoing process. The report is not a final answer but a snapshot that serves as a temporary compass. We need to continue the dialogue and create opportunities for more people to experiment with and benefit from AI developments.

Participants

The roundtable on the development, use and ecosystem of AI was held in Stockholm on 6-8 March 2024. Participants included leading experts from different sectors:

– Daniel Akenine, Microsoft Sweden

– Håkan Andersson, IBM Sweden

– Hannes Berggren, Amazon Web Services

– Björn Blomqvist, Axfood AB

– Catharina Borgenstierna, Archibalds Adventures AB

– Janne Elvelid, Meta

– Anna Essén, Stockholm School of Economics

– Mikael Haglund, IBM

– Johan Harvard, Silo AI/Combient Mix

– Ulf Hertin, Health Invest

– Helena Hånell McKelvey, Amazon

– Hanifeh Khayyeri, RISE AB

– Jens Larsson, Avanto Care

– Mikael Ljungblom, AI Sweden

– Martin Nygren, PA Consulting

– Tomas Ohlson, Einride

– Fredrik Olsson, Gavagai

– Maria Ramstedt, Ericsson

– Olivia Rekman, Nordic Council of Ministers

– Andrea Risberg, Accenture

– Shiva Sander Tavallaey, ABB Sweden

– Erik Sandström, Doctor.se

– Karl Sjöborg, Hopsworks

– Anna Sööder, Schibsted

– Nicklas Tibblin, Atlas Copco

– Jannike Tillå, Swedish Internet Foundation

– Rebecka Ångström, Ericsson Research

– David Österlindh, Nexer Insight AB

– Sara Övreby, Google

Further reading

Here are some selected studies and articles that highlight AI and its applications, with a focus on Swedish companies and organizations:

1. Agarwal et al. (2023): Combines human expertise with AI in radiology. Read more
2. Andersson et al. (2021): Explores software development and innovation in Swedish companies. Read more
3. Batty (2017): Redefining big and small data in urban planning. Read more
4. Bommasani et al. (2021): Opportunities and risks of basic AI models. Read more
5. Brynjolfsson & Hitt (2003): The impact of computing power on productivity. Read more
6. Chen et al. (2022): The impact of GDPR on business performance globally. Read more
7. OECD (2024): OECD Digital Government Index. Read more
8. The Economist (2014): A special report on tech startups. Read more
9. Ledendal et al. (2018): re-use, confidentiality and data protection in the digital society;
10. Legg & Hutter (2007): A collection of definitions of intelligence.
11. Lipsey et al. (2005): Economic transformations and long-term growth.
12. Lundblad et al. (2013): An introduction to working with open data. Read more
13. Manyika & Spence (2023): AI has the potential to reverse the decline in productivity. Read more
14. McCorduck (2004): The history and future of AI. Read more
15. Statistics Sweden (2023): AI use in Swedish companies and the public sector. Read more
16. Thompson et al. (2024): The scope of machine translation on the web. Read more
17. Growth Analysis (2022): Prerequisites, opportunities and barriers for the use of AI in companies. Read more
18. Varadi & Velankar (2023): The impact of the AlphaFold Protein Structure Database on the life sciences. Read more
19. Variant (2018): AI impact on economics and industrial organization. Read more
20. Wernberg (2020): post-pandemic digital transformation of SMEs; Read more
21. Wernberg & Andersson (2022): Digital excellence in business and the public sector. Read more
22. Wernberg (2023a): How data-driven services are changing the economy. Read more
23. Wernberg (2023b): What is AI and why is it important? Read more
24. Wormbs (2010): The digital imperative. Read more

Sources

• Digital Europe (2023), Europe 2030: A digital powerhouse. Read more
• Digitaleurope (2024), The Single Market Love Story. 10 digital actions to save the 30-year marriage. Read more
• ERT (2023), European Round Table for Industry, press release 23 May 2023, Business confidence stabilises, but Europe’s competitiveness is on the decline. Read more
• European Commission (2022), Digital Economy and Society Index (DESI) 2022 Sweden. Read more
• European Commission (2023), Long-term competitiveness of the EU: looking beyond 2030, COM(2023) 168 final. Read more
• European Commission (2024), White Paper. How to master Europe’s digital infrastructure needs?, COM(2024) 81 final. Read more
• Exponential Roadmap (2020), Scaling 36 solutions to Halve emissions by 2030, version 1.5.1, Falk, Johan et al, Future Earth. Sweden. Read more
• McKinsey Global Institute (2022), Securing Europe’s competitiveness: addressing its technology gap. Read more
• Meyers, Zach (2024), Helping Europe’s digital economy take off: An agenda for the next Commission, Centre for European Reform. Read more
• OECD (2024), 2023 OECD Digital Government Index, OECD Public Governance Policy Papers. Read more
• TechSweden (2023), Swedish tech industry 2023. Read more