The Responsible Organisation

The City of Helsinki, the capital of Finland, is the largest employer in Finland, with over 38,000 employees dedicated to providing a wide array of public services, including education, healthcare, and culture. Recognising the potential of emerging technologies, the city's digitalisation unit, has been actively testing different technological solutions and exploring the integration of artificial intelligence (AI) into its daily operations. 

The PSTW interviewed the city’s Head of Data and Analytics, Tomas Lehtinen, with the aim of understanding how the government team is experimenting Generative AI solutions in this context.

The problem

The City of Helsinki provides over 800 different services and produces massive amounts of data, presenting a significant level of complexity in terms of skillsets required among public servants, information management and data processing. The government team has been continually seeking ways to optimise processes, improve internal knowledge and capabilities, and provide better services for citizens.  As Tomas Lehtinen explains in the interview, “in Finland we have digitalised (…) services very, very long time ago, many of them have been at least 30 years and they have produced a lot of data. That city, I would say have  at least 600 different data systems, apps, and similar”.

However, recently, the government team also observed that employees started to already spontaneously experiment with generative AI tools like ChatGPT, using these tools in their daily work. As Thomas Lethien explained, a challenge emerged consequently regarding how to better ensure a more structured and secure way to employ generative AI tools that could be integrated into the city’s existing workflows.

The solution and its implementation

To address this challenge and harness the potential of generative AI in a secure, controlled and trustworthy manner, the City of Helsinki decided to explore integrating AI tools into its existing work environment. Microsoft 365 Copilot emerged as a promising solution, offering AI-powered assistance within the familiar Microsoft 365 suite. The tool's functionalities include generating meeting notes, summarising lengthy documents, and drafting emails, among other tasks. In his words, "we wanted to have some kind of tool that would be as easy as possible for our people to use, and they are using Microsoft 365 tools every day." The city's familiarity with Microsoft 365 products made Copilot a natural choice for initial testing.

For this reason, the City of Helsinki decided to launch a pilot program in 2014 to explore Copilot's potential to enhance knowledge work across various departments. Lehtinen noted that Copilot was seen as "the easiest way to get an understanding of how generative AI can help us". The main objective behind the pilot, as Lehtinen stated, was not to try how Copilot performs but to learn about how a large group of city staff understands generative AI, and how they can use it in their work."

The pilot program

The pilot program adopted a phased approach, gradually introducing Copilot to four different groups of employees across various city departments. The selection process for the pilot was strategically oriented towards the selection of entire city teams with an explicit attention to areas of government that did not handle sensitive data. 

Firstly, instead of selecting individuals, the city selected entire teams to participate in the pilot to ensure a comprehensive sample to test the solution. This approach also allowed for a wider range of perspectives and use cases to be considered. In total, 1000 employees participated in the pilot, representing a diverse cross-section of the city's workforce.

Secondly, a distinctive feature of the pilot was the strict data protection standards to minimise data-related risks. In practice, one of the key decisions was to exclude from the pilot those areas who regularly handled sensitive data. As Tomas Lehtinen highlighted, "We decided that if our employees are working with [citizens] data, they cannot join the pilot." Additionally, HR teams that handled employees or potential new hires data were also excluded for this phase. This restriction aimed to minimise the risk of inadvertently exposing personal or confidential information to the AI tool, building on the provisions agreed with Microsoft prohibiting the reuse of the city’s data to train their language models. Furthermore, the roll-out of this pilot aligned with the City of Helsinki’s eight ethical principles to guide responsible AI adoption, i.e., human-centeredness, transparency, explainability, fairness and non-discrimination, accountability and trust, privacy, safety, and human oversight over AI operations.  

To prepare the participants for the pilot and ensure high adoption rates, training sessions were delivered to equip staff with the necessary skills to utilise Copilot effectively. The trainings covered not only the technical aspects of using the tool, but also broader concepts related to generative AI and its potential applications in their work.

Throughout the pilot program, the city monitored the usage of Copilot. This involved tracking the number of entries and queries made by employees, providing data on the adoption and utilisation of the tool. This data helped the city understand how employees were interacting with Copilot and identify areas where further training or support might be needed.

In parallel, the city gathered feedback through surveys to understand, from the employees’ perspective, how the solution impacted their wellbeing and daily work. These surveys provided a platform for employees to share their experiences and perceptions of Copilot, highlighting both the benefits and challenges they encountered. The feedback gathered through these surveys played a central role in informing future decisions regarding the tool’s wider implementation. Feedback data gathered highlights that, while 48% of participants felt that Copilot didn't meet their high expectations, a significant portion reported positive outcomes. Lehtinen shared that "25% of the people in Group #3 [of the pilot test] felt that they are feeling better in their work, [meaning that] it's much easier to come to work when they're using Copilot." The pilot also highlighted specific use cases where Copilot proved particularly beneficial, such as aiding employees with limited Finnish proficiency in understanding and translating documents.

Expected benefits

Among the benefits identified and expected, the following arise as most relevant:

• A positive impact of the pilot was the empowerment of city staff, illustrated by the feedback survey results. The adoption of the tool provided staff with support and more confidence to develop daily tasks, improving their wellbeing and satisfaction with their work results. 
• Expected benefits also comprise an improvement in both the quality and the efficiency of the work across the city teams, potentially leading to increased productivity and time savings in the long term.
• A key benefit of the solution was the seamless integration of Copilot into the government’s existing Microsoft 365 environment. This meant that employees could access and utilise the AI tool within their familiar workspace, minimising disruption: since the city's staff were already accustomed to using Microsoft 365 tools daily, incorporating Copilot into this environment felt like a natural extension of their existing processes. This facilitated a smoother adoption of the new tool and encouraged wider participation in the pilot.
• The pilot project proved feasible to implement a generative AI tool across the city workforce while maintaining the data security. The collaboration with Microsoft was based on a large and comprehensive data privacy agreement that guarantees that the information integrated in Copilot by the employees would not be used to train their language models. Furthermore, to safeguard sensitive data, areas that managed sensitive data were explicitly excluded from the pilot project. Finally, to mitigate further potential risks, the City of Helsinki decided to disable the Bing AI feature within Copilot for all pilot participants. 

Main challenges 

The pilot program also encountered some challenges. The main obstacles identified are the following:

• In the first version of the Copilot tool utilised, the pilot group faced language support barriers. The lack of comprehensive Finnish language support posed a significant obstacle for employees, as they develop the majority of their work in this language. This caused some degree of disappointment among the first pilot group, in particularly among those who primarily worked in Finnish. After Microsoft included the Finnish language support in a more recent version of Copilot, pilot groups showed a higher degree of satisfaction with the tool.
• Emerging challenges as regards managing employee expectations and addressing concerns about the impact of AI on their roles. Clear communication, explicability of the tool and training will be essential to ensure adoption and transition in a responsible manner towards employees. 
• Additionally, other topics related to the ethical implementation of Generative AI tools are relevant, such as considering the potential impact of Copilot on different employee groups, and ensuring that the tool is used in a way that aligns with the city's values.
• As the city moves towards a scale-up phase of the solution and/or further testing and experimenting with other Generative AI tools, maintaining strict data protection standards and prioritising data security and sovereignty will be crucial. The city has already built a large amount of learnings on data privacy agreements, which will support the government team further advance.

Link: https://www.hel.fi/en/news/the-city-of-helsinki-is-investigating-the-benefits-of-artificial-intelligence-in-various-work-tasks

Tomas Lehtinen

Email: tomas.lehtinen@hel.fi