A shared digital assistant for the public sector
This is a collaboration between Swedish authorities, municipalities, regions, and industry, coordinated by AI Sweden. The objective is to promote national cooperation on AI for text tasks and create conditions for shared solutions in the public sector.
The ultimate goal is to support public sector employees and free up their time so they can focus on more creative and interpersonal tasks. Success requires public sector organizations to share critical resources while sharing the burden of gathering the data needed to create truly competent and flexible solutions.
The first phase of this initiative focuses on data generation, change management, identifying use cases, and developing a prototype for an AI assistant using large-language models.
Building a prototype for a shared digital assistant is one of the identified tasks to support the use of AI in the public sector. In this project we explore and identify how the prototype should be designed in order to be easy to use and accessible to every staff member, regardless of their technical skills.
Examples of tasks the assistant could be able to help with:
- Summarize and analyze large amounts of text
- Improve and edit texts to make them more readable and structured
- Create new texts such as reports, documents, and applications, customized according to specific templates and guidelines
- Classify documents and create decision support in administrative processes
- Develop personalized support to assist citizens best
It is a pioneering initiative that will free up time in the welfare sector and open up for more interpersonal and valuable work.
Julia Olander, innovationsledare på Göteborgs stad
What do we face ahead?
Public sector employees in Sweden deal with extensive and varied text-related tasks daily. These are essential to fulfill the sector's mission and provide services to citizens but are also often labor-intensive and time-consuming. At the same time, a significant shortage of staff is expected in the sector due to the aging population and declining labour force.
+50%
The number of people aged 80 or older will increase by almost 50% by 2031
Source: SKR 2022
+410,000
By 2031, the number of workers needs to increase by 410,000.
Source: SKR 2022
Alongside the challenge of a shrinking labour force and the increasing needs of an ageing population, there is also a lack of AI skills, computational power, legal requirements and, above all, the lack of Swedish data that reflects the specific needs of the sector. These challenges are too big for individual organisations, and therefore a broad collaboration across the sector with joint initiatives for AI is required.
AI technology is part of the solution
AI technology, especially large-scale language models, has the potential to greatly support this challenge. In a 2019 report, DIGG estimated that AI technology could create a value of SEK 140 billion annually in the public sector (1). Moreover, the report was written before big language models had had time to prove their enormous value.
The potential of large-scale language models is particularly relevant for public sector organizations as much of the data handled in the sector is text-based. Research also shows that using large-scale language models can significantly increase the efficiency and quality of text-based work (2).
- Agency for Digital Government (DIGG). (2019). Främja den offentliga förvaltningens förmåga att använda AI
- (2023). Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality. Working Paper 24-013, Harvard Business School.
How will the initiative be implemented?
To date, the initiative has worked on needs assessment in the participating organizations and identified key use cases and areas for the use of a digital assistant. Based on this, the data generation process has started, with staff in the organizations creating data that can be used to improve the language model.
The project has also developed a prototype for a digital assistant called Svea, which can already be used within the participating organizations. Svea is a web-based chat interface designed to be model agnostic and based on open source. This solution runs on infrastructure in Sweden and is provided by the technology company Intel.
In addition, the project requires extensive legal expertise, which is done in collaboration with the participating organizations. The aim is to design the system to be usable and adapted to the legal requirements of public organizations.
Change management is also a major part of the initiative. Employees and different levels of operations are engaged in finding new ways of working and using the digital assistant effectively and responsibly.
It is clear that AI Sweden's starting point is that the solution must be legal in order to be scalable. Therefore, the legal issues are taken extremely seriously, and technical solutions are adapted accordingly.
Linnea Princis, kommunjurist Kungsbacka kommun
We are now launching the next step
The call for expressions of interest for phase 2 of this initiative is now open. In this phase, around 30 authorities, municipalities and regions will work together with actors in the business community in a project that extends over one and a half years, coordinated by AI Sweden. There is great interest and many organisations have already started to register their participation.
Participation in the project provides many benefits:
✓ Employees, in participating organizations, can use the Svea prototype in their work
✓ Organizations own database of documents that can be used in Svea
✓ Access to an open database with national information
✓ Skills development - Increased understanding of AI & responsible use
✓ Business development with new ways of working
✓ Increased AI maturity throughout the organization
A shared digital assistant will be an important tool for current and future challenges in health and social care [...]. Participation in the project is highly recommended for those municipalities that want to drive the development forward.
Jenny Fermby, verksamhetsutvecklare, Tjörns kommun
Please get in touch with Jonathan if your organization is interested in participating in phase 2!
Svea
Svea is the prototype for a shared digital assistant, which is under development in this project. The AI-based prototype will be designed to help public sector workers with text-based tasks. Currently, based on the Mixtral open language model from the French company Mistral, more models will be tested in phase 2.
Employees interact with Svea via a web-based chat. Svea can already answer business-specific questions, summarize documents, extract information, rewrite texts in easy-to-read Swedish, and much more.
Svea aims to make working days more efficient and less stressful and to improve the quality of text-based tasks, especially for those without specialized skills. Employees can provide feedback on the assistant's behavior, helping to continuously fine-tune and improve Svea to meet business needs and handle more advanced tasks.
AI Sweden coordinates data collection, infrastructure, fine-tuning, development of the digital assistant Svea, training, and change management. The organization includes experts in artificial intelligence, data science, and AI transformation.
Timeline
Phase 1
- Until summer 2024
- Funded by Vinnova and partners
- From the public sector: Kungsbacka municipality, Tjörn municipality, Gothenburg city, Region Skåne, Region Halland, and Västra Götaland Region
- Intel is the project's hardware partner and will assist the project with all the hardware required
Phase 2
- Starting autumn 2024
- Planned to last until the end of 2025/beginning of 2026
- Around 30 municipalities, regions, and authorities are invited to participate
Phase 3
- From 2026, the ambition is to hand over the initiative to a long-term operator
For more information, contact
Related articles
AI assistant training for Swedish public sector underway
Significant new initiatives to propel Sweden's AI development
Learn more about our other sector initiatives
Collaboration for AI in Municipalities and Civil Society
Energy
