What do you think about AI, Finland?

Summary: key findings from the analysis of the Voxit discussion

A total of 6,429 people took part in the Finnish public discussion on the use of artificial intelligence in public services. Participants voted on 194 statements, of which 169 were added by citizens themselves. The discussion took place on the digital Voxit participation platform in September 2025.

The Voxit platform identified two distinct opinion groups within the discussion:

• AI-positive (approx. 13,5% of all participants): view AI as useful and ready for broader adoption
• AI-cautious (approx. 59,1% of all participants): wish to proceed slowly, emphasising risks and privacy

Both groups highlighted the importance of responsible development, EU-level regulation, and the use or support of domestic/European AI. Citizens also expressed a strong need for more information about AI: they do not know enough about where and how AI is currently used in public services.

Key themes included:

• Finns approach AI with a generally critical but constructive mindset; polarisation was surprisingly low. There was broad consensus on safety, data protection, transparency, and the role of humans in decision-making.
• Data protection and privacy were regarded as extremely important – access to data by AI systems should be tightly restricted, and citizens should have the right to know what data about them is collected.
• Health and social care topics revealed differences between the opinion groups, particularly concerning AI doctors, AI-assisted therapy, and the combining of sensitive data.
• Citizens want AI systems to be regularly assessed. They expect openness about how AI is used, and they want final decision-making power in critical services to remain with humans.

Overall, Finns hope that AI functions as a tool that supports people, rather than as an independent decision-maker.

1. Introduction

In September 2025, Sitra, in cooperation with the technology consultancy Reaktor, organised a large national discussion on the use of artificial intelligence in public services. The discussion was titled “What do you think about AI, Finland?”. The aim was to gather the views of ordinary Finns on the acceptable limits of AI use in public sector services. The theme was considered timely, as AI is developing rapidly and becoming increasingly embedded in public administration processes and services.

At the same time, several studies and reports (University of Tampere, 2026: User trust in AI and major tech companies in twelve countries) have shown that a significant share of Finns do not trust AI or those who develop it. One of the key aims of the citizen discussion was therefore to increase understanding of how trust in AI‑enabled public services could be strengthened.

The national AI discussion was part of a broader experiment. A central goal of the experiment was to create an AI rulebook based on the opinions of Finnish citizens. This rulebook could then help public authorities ensure that AI is developed, used and governed in ways that Finns consider acceptable. The outcome of the experiment – the rulebook (LINK TO MEMO HERE) – contains citizen‑driven rules that can be used both to train AI models and to support the development and governance of AI systems in public services.

The What do you think about AI, Finland? discussion took place on the Voxit participation platform from 5 to 28 September 2025. People were invited to take part through letters, digital advertising, events, emails and social media posts. The aim was to involve a broad range of Finns: at least several thousand participants. The goal was not to assemble a group that perfectly represented Finland’s population.

Voxit is a low‑threshold digital platform that enables large groups of people to express their views constructively on a selected topic. On the platform, participants vote on statements related to the theme by agreeing, disagreeing, or skipping the statement. Unlike traditional surveys, participants can also propose their own statements to make sure their perspectives are included.

Alongside the creation of the rulebook, the experiment produced a large and interesting dataset that sheds light on Finnish views regarding the use of AI in public administration. The content of this article is based on the statements generated during the Voxit discussion.

It is important to note that the concept of AI was not defined in detail for participants, and general public perceptions of AI inevitably influenced the discussion. As a result, some participants likely focused on, for example, the use of generative language models in public services, while others considered the broader service processes or IT systems of the public sector. It is also worth noting that the statements proposed by citizens were not excluded for being considered ‘self‑evident – meaning already covered, for instance, in Finnish legislation or public sector guidelines.

In the analysis phase, around fifteen statements added to the Voxit discussion late in the process were excluded, as they did not receive enough votes in time. All statements included in the analysis were ones in which at least one fifth of the members of two different opinion groups participated.

The discussion on the use of AI in public sector services was introduced on the Voxit platform with the following message: “Everyone has the right to take part in the discussion on the use of AI. Let’s create together the world’s first citizen‑generated rules for AI. The discussion is organised by Sitra and Reaktor. The results will be shared with decision-makers and civil servants.”

2. What themes were discussed?

Artificial intelligence is often seen as a topic that requires extensive expertise and specialised knowledge. Public discussions on AI are typically led by experts. When the discussion on the use of AI in public services was launched on the Voxit platform, the organisers also wondered whether Finns would find the theme too difficult or too distant.

This concern proved unnecessary. Citizens proposed more than 560 statements for the Voxit discussion, of which 171 were accepted. In addition, the organisers provided 23 background and seed statements. The number of rejected statements in relation to the published ones was typical for Voxit discussions. In discussions held on the platform, statements are typically moderated out if they are similar in content to statements that have already been published. Many statements were also left out because they were too general or their connection to the use of AI in public administration was considered too weak.

A total of 6,429 people took part in the citizen discussion on AI, and of these, 4,667 participants (72.6% of all participants) voted often enough for the Voxit platform to place them into the opinion groups that emerged in the discussion. Altogether, participants cast 352,929 votes on the statements.

The citizen‑proposed statements published on the platform covered a wide range of perspectives. Citizens wanted to discuss, among other things, information security, data protection, and the governance and transparency of AI systems used in public services. However, what engaged Finns most were topics at the societal level. When the published citizen statements are grouped into sub‑themes, it becomes clear that the two largest themes relate to such broader issues.

The highest number of statements added by citizens – 55 in total – concerned the possibilities, risks and ethical limits of AI use in social and health care services or elsewhere in public administration. The second‑largest group, 41 statements, dealt with the broader societal or democratic impacts of AI. Citizens would have liked to discuss these societal issues even more, but many statements on this theme were moderated out to ensure that the discussion remained focused primarily on AI use in public services.

The following sub‑themes were represented in the citizens’ statements published on the platform:

Societal level

• AI in social and health care services and other areas of public administration: 55 statements added by citizens
• Impacts on society and democracy: 41 statements added by citizens
• Impacts on skills and education: 8 statements added by citizens
• Individual rights and equality: 6 statements added by citizens

Governance and security of AI systems

• Safety and information security: 30 statements added by citizens
• Governance of AI systems, including technical management of AI systems and service design: 19 statements added by citizens

Procurement of AI systems

• Procurement rules: 15 statements added by citizens
• Compliance with laws and regulations: 4 statements added by citizens

Transparency and protection of privacy

• Data protection and privacy: 13 statements added by citizens
• Transparency of AI systems: 9 statements added by citizens

Sustainability

• Responsible use and development of AI systems: 9 statements added by citizens

Some of the statements listed above fall under two different themes, which is why their combined number is higher than the 194 statements that were voted on.

3. Which AI‑related themes did Finns agree on, and where did opinions diverge?

The real‑time results report generated by the Voxit platform is based on statistical and machine‑learning methods. Using these methods, the system produces, among other things, a consensus scale of the statements published in the discussion. Each point on the scale represents one statement. At the left end are the statements with which the majority of participants agreed. At the right end are those on which opinions were most divided.

Figure 1: Consensus scale from the “What do you think about AI, Finland?” discussion

The scale shows that participants in the AI discussion agreed on most statements reasonably well, and only a small number of statements were strongly divisive.

Below are example statements with vote counts, shown in the form in which they originally appeared in the discussion.

Statements with the highest levels of agreement among citizens

• “Under‑18s should be taught how to search for information also without using AI.” (Agreed 97%, n=2109)
• “I should have the right to know what data about me has been collected by AI, where it comes from, and where it is stored.” (Agreed 96%, n=2142)
• “The usefulness of AI should be assessed regularly to ensure that its errors do not create additional work for public‑sector professionals.” (Agreed 94%, n=2099)
• “AI must not be given unlimited access to people’s data. Access should be limited strictly to the areas necessary for its task.” (Agreed 93%, n=2206)
• “There must always be a clear and visible label on data or images generated by AI.” (Agreed 93%, n=1982)

There was similarly strong agreement on the following statements, but in these cases, citizens agreed in rejecting the statement. A clear majority disagreed with the following:

• “I believe that political decision‑makers have sufficient knowledge about AI and its effects.” (Disagreed 92%, n=2273)
• “I am prepared to rely on AI‑made decisions for my life and health.” (Disagreed 89%, n=1743)
• “I do not need to know whether a decision concerning me (from Kela/Social Insurance Institution of Finland, the Tax Administration, the Police, etc.) has been made by a human or a machine.” (Disagreed 88%, n=2327)

Statements on which opinions were strongly divided

• “I would like to have a personal AI doctor on my phone, with which I could chat and show images.” (Agreed 25%, disagreed 61%, n=2270)
• “I believe that AI can coordinate a client’s care and social‑welfare services better as a whole than a health and social care professional.” (Agreed 20%, disagreed 62%, n=2306)
• “I want AI to assess my risk of illness and guide me to make lifestyle changes.” (Agreed 34%, disagreed 48%, n=2332)

4. What kinds of opinion groups emerged from the discussion?

The Voxit platform forms opinion groups based on voting data. These groups differ from one another according to voting behaviour and the key statements that characterise each group. In the analysis phase, examining these opinion groups helps to create a clearer overall picture and form a more coherent narrative about how the groups relate to one another.

Two distinct opinion groups were identified among participants in the AI discussion. Based on their voting patterns, they were named “AI‑positive” and “AI‑cautious”. The AI‑cautious group was clearly the larger of the two, with 3,799 participants (81.4% of grouped participants, 59.1% of all participants), while the AI‑positive group consisted of 868 participants (18.6% of grouped participants, 13.5% of all participants).

Of all participants, 27.4% remained outside the opinion groups. A participant must respond to at least seven statements for grouping to be possible.

Additional analysis revealed further nuances among participants

The diverse data generated through Voxit discussions makes it possible to conduct various supplementary analyses and to refine the categorisation of opinion groups. In this AI discussion, in addition to the platform’s own analysis, a data analyst examined the discussion data and identified the following opinion groups:

• Optimists (38% of respondents) believe that the benefits of AI significantly outweigh its risks.
• Critical adopters (24% of respondents) accept AI and use it themselves but demand strong governance due to AI’s shortcomings and risks.
• Sceptics (38% of respondents) doubt the promised benefits of AI.

However, the proportions of respondents in each group should be interpreted with caution. The analytical methods used group the data statistically, and there are always borderline cases that might be classified differently using another method. It is likely that many who were classified as “optimists” or “sceptics” could just as well fit into the group of “critical adopters”.

It is nevertheless interesting that, compared with the platform’s classification, the large AI‑cautious group split into two parts in the data analyst’s observations. One part, the sceptics, approach AI with fundamental doubt. The other part, the critical adopters, are interested in AI and even use it themselves but simultaneously demand strong and well‑considered processes for the governance, development and use of AI.

The number of statements voted on per participant varied considerably. More than one thousand participants cast only a single vote, while only about half of participants cast more than 20 votes. This variation suggests that the available data may be somewhat skewed.

Despite these interesting additional analyses, the thematic examination in this article is based on the two opinion groups created by the Voxit platform. These groups did not differ significantly in socio‑demographic characteristics: both groups included, for example, a relatively high share of people with higher education, and the majority of participants lived in cities. The single factor that clearly distinguished the groups was their stated experience of AI. In response to the statement “I use AI rarely or not at all”, only 27% of AI‑positive participants agreed, while the corresponding share in the AI‑cautious group was 66%.

Cautiously interpreted, the number of AI‑positive individuals could be increased by encouraging and supporting people in trying out AI personally. Due to limitations in the background data available for participants, the analysis cannot delve more deeply into the reasons behind their limited use of AI. These reasons may, for example, be principled, related to data protection, or reflect a lack of skills.

Opinion group A: AI‑positive (868 respondents / 13.5% of all participants)

Members of this group have a practical and unafraid attitude towards AI. They see AI primarily as something that supports everyday life and improves the efficiency of the services they use.

AI‑positive participants believe that the benefits outweigh the risks, provided that AI development is guided responsibly. They are clearly more willing to allow their personal data to be used by public authorities if this leads to better services. They also do not support placing significant restrictions on the use of AI.

Their positive attitude is also evident in their readiness to use AI personally in more sensitive areas, such as health and social care services. For example, they would be willing to consult an AI doctor to speed up access to care.

Opinion group B: AI‑cautious (3,799 respondents / 59.1% of all participants)

AI‑cautious participants view the use and adoption of AI through the lens of risk and consider AI applications to be still insufficiently mature. This group is also more sceptical about AI’s productivity potential.

Although members of this group are also willing to use AI in many everyday services, they are more critical about sharing their personal data. Overall, they emphasise the need to remain in control themselves and to be able to choose when and where AI may be used.
Concerns about AI’s environmental impacts appear more strongly in this group than in the other. They also strongly believe that ordinary people should have more influence over how public authorities use AI and develop services.

6. Differences and similarities between the opinion groups when examined by theme

The differences between the two opinion groups formed from the Voxit discussion data also became visible when the material was examined in greater depth, theme by theme. The analysis highlighted numerous themes related to the governance, development and use of AI, on which there was broad agreement between members of the different opinion groups. Despite the differences between the groups, it appears that participants’ underlying values are largely built on similar foundations.
When reviewing the percentages for each statement, it is important to note that every statement was voted on by a different number of participants. This number is shown alongside the statement with the symbol N.

6.1. Themes on which both opinion groups held broadly similar views

Safety and information security

One of the themes that generated the strongest consensus among participants in the AI discussion concerned the safety and information security of AI systems used in public administration. In both opinion groups, a clear majority demanded that public authorities should have precise and regular assessment processes in place to ensure the ethical and appropriate use of AI. Citizens expect AI systems to be evaluated both before their deployment and throughout their operational use.
Below are several statements related to this theme, along with the number of votes each received, broken down by group. Additional statements concerning this and other themes can be reviewed in the results report produced on the Voxit platform (LINK TO PDF REPORT).

The geography of AI solutions

Participants in the discussion paid considerable attention to the question of which countries AI solutions should be sourced from, and, on the other hand, where and how data should be processed. These types of statements are challenging to interpret, as for example language models are only one part of a broader system solution. What does it mean, for instance, to “support Finnish or European AI development”? Does it refer to favouring such solutions in procurement, or to providing funding for product development?

As the theme was clearly important for participants, the discussion should be continued and further specified.

Human control in decisions and benefits

There was also broad agreement among citizens that AI should not be given excessive power when decisions are made about individual benefits or access to services. In such critical points of decision‑making, responsibility and final authority should, in the view of Finns, remain with public‑sector employees.

The idea of human control looks somewhat different when the focus shifts from the general level to questions in health and social care, where attitudes towards the use of AI are more negative overall.

AI and sustainability

There was some variation between the groups in statements related to the responsibility and sustainability of AI, but the majority of respondents in both groups thought in broadly similar ways. Both environmental and social challenges were considered important factors in AI solutions, and many felt that these should also influence public‑sector procurement decisions.
It is noteworthy, however, that the groups’ views differed sharply on the expert‑generated seed statement “The use of AI in public services should be restricted because of its environmental impacts”. Overall, fewer than half of all respondents who voted on the statement selected Agree.

6.2. In which themes did the two opinion groups differ?

Social and health services

The AI‑positive group expressed significantly more confidence than the AI‑cautious group in the maturity of AI as a practical tool and were more willing to accept social and health services produced with the help of AI. The groups also differed markedly when assessing the use of AI as support in mental‑health services.

Data protection and privacy

Clear differences could be seen in citizens’ attitudes towards the collection and use of data for AI purposes. These differences concerned both the sharing of other people’s data and the sharing of one’s own. Despite this, a significant majority in both groups still expected people’s privacy to be protected and access to sensitive data to be restricted.

AI‑driven guidance and the role of AI in decision‑making

In these statements, the differences in citizens’ views reflect their level of trust in AI, which in turn has a major impact on how much decision‑making power they are willing to give it. If people do not trust the maturity or capability of AI, they are not willing to accept AI‑driven guidance or decision‑making. Opinions on decision‑making are noticeably stronger than views on AI‑based recommendation or guidance.

7. The big picture of the AI discussion

Some of the key findings from the discussion confirmed what was already known, such as the general scepticism towards AI, while others were more surprising — for example, the low level of polarisation.

Based on the discussion, Finns are not significantly divided over the use of AI. Their underlying values are largely shared, and there is broad agreement, for example, that the governance processes for AI systems used in public administration must be responsible and subject to regular evaluation.

The data reinforces the picture presented in previous studies that a large proportion of Finns take a critical view of AI. However, it adds new dimensions: there appears to be a substantial minority whose views on AI are highly positive. They want to adopt AI widely and quickly, including in public‑sector services. At the same time, the data shows that many Finns who approach AI more critically do not reject its use categorically. They see that carefully and responsibly developed and deployed AI can bring many benefits, but they wish to proceed more slowly than the AI‑positive group so that decisions are made thoughtfully and potential negative consequences are properly assessed and minimised.

A large majority of citizens — in both opinion groups — oppose any form of “being forced”. They do not want AI to be given the power to make final choices on behalf of individuals or to strongly steer people, for example in matters related to health. They are somewhat more open to guidance in the form of recommendations. Overall, the importance of personal control over one’s own affairs is a strong and recurring theme throughout the discussion. The need to improve AI literacy was also emphasised.

Participants paid considerable attention to national and regional AI solutions. Citizens added ten accepted statements on this theme. EU‑level AI legislation was considered important, people wanted data to remain within EU or national borders, and most respondents to these statements hoped that the public sector would procure AI solutions from Finnish providers. It would be valuable to understand better how “European” and “Finnish” relate in this context — especially in comparison with, for example, American or Chinese AI.

Citizens also submitted several high‑level statements in which AI was seen primarily as an opportunity rather than a threat. Each of these statements received more than 2,000 responses, with clearly more than half calling for more discussion on the opportunities of AI instead of “getting stuck in horror scenarios” that do not move things forward.

Overall, Finns take a practical approach to AI, emphasising an assessment of benefits relative to actual needs and potential harms. People are willing to grant AI a role that supports and assists humans.

7.1. Future use of the discussion data

Citizens are both willing and able to discuss AI in a factual, constructive and nuanced manner when given the opportunity. Even though most people do not understand the technical workings of AI or the legislation surrounding it, they are fully capable of assessing how they would like to be treated as users, subjects and customers of public services that make use of AI.

The collected data can help decision‑makers and civil servants identify important themes they should consider when designing new services and communicating about them — or even when drafting new laws. The discussion data helps to reveal which AI‑related themes are associated with widespread concerns, uncertainty or misunderstandings among the public, and which, in contrast, generate strong interest and optimism.

Many issues in the AI discussion remained at a general level and would require deeper examination. For example, it would be useful to understand questions and concerns related to health and social care services in more depth. The discussion could therefore be continued, for instance, through a citizen panel or a new digital public consultation with a more targeted focus.

For service developers, the material has a great deal to offer. By examining especially the statements that divided opinions, they may be able to identify important societal tensions that should be considered when developing and deploying AI in public administration.

The most important overall insight from the discussion relates less to the goal of creating a rulebook that reflects citizens’ wishes for AI in the public sector, and more to the significant need for information. Participants repeatedly expressed that they do not know what AI is currently used for, what is allowed, or what rights they have in relation to it. They also expressed low trust in the level of AI knowledge among political decision‑makers. Civil servants and policymakers would be well advised to invest heavily in active communication. Citizens are looking for specific things, such as clear indications inside systems and decisions when AI is used — but also broader information about AI use in public services.

At the same time, citizens expressed the view that they should have a greater and more active role in shaping how public services are developed and in what direction they evolve. This creates a remarkably rich opportunity for meaningful interaction between public authorities and citizens. Through transparency, active communication and meaningful involvement of citizens at key moments, we can safeguard one of Finland’s unique strengths: societal trust.

Read also Sitra memorandum: Participatory AI development model – A citizen dialogue on the use of AI in public services.