AI Transformation Strategist
Here are some of the projects I’m most proud of:
A unified AI assistant for the Swedish public sector
Overview
Svea is a groundbreaking nationwide initiative to develop a unified digital assistant for the Swedish public sector. The idea is to leverage large language models to streamline text-related tasks across all levels of government, enhancing efficiency, effectiveness and quality.
Challenge
The Swedish public sector faces significant staffing shortages due to an aging population and decreasing workforce. Simultaneously, public sector employees handle extensive and varied text-related tasks daily, which are often time-consuming and labor-intensive. While public sector organizations are among the largest potential beneficiaries of a digital assistant capable of assisting in a broad variety of text-related tasks, they face several barriers to unlocking this potential:
These challenges make it difficult for individual organizations, especially smaller ones, to develop and implement effective AI solutions independently.
Approach
1. National Collaboration: Coordinated effort across multiple government agencies, municipalities, and private sector partners to overcome resource limitations.
2. Resource Pooling: Sharing critical resources such as AI expertise, compute power, and legal knowledge.
3. Collective Data Generation: Engaging public sector workers and domain experts to generate data for a shared pool of instruction data representative of the Swedish public sector's needs.
4. AI Development: Fine-tuning a Large Language Model (LLM) for sector-specific support, leveraging the collectively generated data.
5. Holistic Strategy: Addressing technical innovation, change management, legal frameworks, and data generation comprehensively.
6. Inclusive Design: Ensuring the solution is accessible and beneficial to all public sector organizations, regardless of size or resources.
Solution
Svea, a prototype digital assistant based on the open-source Mixtral language model, developed to address the specific needs of the Swedish public sector:
The solution is designed to be scalable and adaptable, allowing even smaller organizations to benefit from advanced AI capabilities without needing extensive in-house expertise or resources.
Results & Impact
Since its conception in the summer of 2023, Svea has made significant strides. The project successfully secured $900,000 in funding for its initial 'zero to one' phase, demonstrating strong support for the initiative.
Currently, Svea serves approximately 1,000 users across participating organizations, marking a substantial user base for a nascent public sector AI tool. The project has fostered increased AI literacy among participants and inspired new AI-driven workflows in public organizations.
A noteworthy achievement is the creation of an open database of general routines, guidelines, and handbooks for municipalities, enhancing knowledge sharing across the sector.
Most importantly, Svea operates in a continuous improvement cycle, with users generating instruction data that is used to fine-tune the underlying language model, ensuring the assistant evolves to meet the specific needs of the Swedish public sector. This collaborative approach has not only improved the tool's performance but also demonstrated the potential for widespread impact through shared AI development in the public sector.
Back to homepage
Svea is a groundbreaking nationwide initiative to develop a unified digital assistant for the Swedish public sector. The idea is to leverage large language models to streamline text-related tasks across all levels of government, enhancing efficiency, effectiveness and quality.
Challenge
The Swedish public sector faces significant staffing shortages due to an aging population and decreasing workforce. Simultaneously, public sector employees handle extensive and varied text-related tasks daily, which are often time-consuming and labor-intensive. While public sector organizations are among the largest potential beneficiaries of a digital assistant capable of assisting in a broad variety of text-related tasks, they face several barriers to unlocking this potential:
-
Widespread lack of AI expertise
-
Insufficient compute resources
-
Limited legal competence in AI implementation
-
Nationwide shortage of training data tailored to the needs of the Swedish public sector
These challenges make it difficult for individual organizations, especially smaller ones, to develop and implement effective AI solutions independently.
Approach
1. National Collaboration: Coordinated effort across multiple government agencies, municipalities, and private sector partners to overcome resource limitations.
2. Resource Pooling: Sharing critical resources such as AI expertise, compute power, and legal knowledge.
3. Collective Data Generation: Engaging public sector workers and domain experts to generate data for a shared pool of instruction data representative of the Swedish public sector's needs.
4. AI Development: Fine-tuning a Large Language Model (LLM) for sector-specific support, leveraging the collectively generated data.
5. Holistic Strategy: Addressing technical innovation, change management, legal frameworks, and data generation comprehensively.
6. Inclusive Design: Ensuring the solution is accessible and beneficial to all public sector organizations, regardless of size or resources.
Solution
Svea, a prototype digital assistant based on the open-source Mixtral language model, developed to address the specific needs of the Swedish public sector:
-
Answers sector-specific questions using a curated database of public sector documents (RAG - Retrieval Augmented Generation)
-
Summarizes documents and extracts specific information
-
Rewrites texts for clarity and accessibility, adhering to public sector communication standards
-
Assists with various text-based tasks common in public administration
-
Provides a user-friendly interface accessible to all levels of public sector employees
-
Incorporates legal and ethical considerations specific to Swedish public administration
The solution is designed to be scalable and adaptable, allowing even smaller organizations to benefit from advanced AI capabilities without needing extensive in-house expertise or resources.
Results & Impact
Since its conception in the summer of 2023, Svea has made significant strides. The project successfully secured $900,000 in funding for its initial 'zero to one' phase, demonstrating strong support for the initiative.
Currently, Svea serves approximately 1,000 users across participating organizations, marking a substantial user base for a nascent public sector AI tool. The project has fostered increased AI literacy among participants and inspired new AI-driven workflows in public organizations.
A noteworthy achievement is the creation of an open database of general routines, guidelines, and handbooks for municipalities, enhancing knowledge sharing across the sector.
Most importantly, Svea operates in a continuous improvement cycle, with users generating instruction data that is used to fine-tune the underlying language model, ensuring the assistant evolves to meet the specific needs of the Swedish public sector. This collaborative approach has not only improved the tool's performance but also demonstrated the potential for widespread impact through shared AI development in the public sector.
Back to homepage
Overview
Starting out at AI Sweden, I was product manager for the AI Maturity Assessment, a service designed to help organizations evaluate their current AI capabilities, identify gaps and areas for improvement, and develop a roadmap to build a more effective AI program.
Starting out at AI Sweden, I was product manager for the AI Maturity Assessment, a service designed to help organizations evaluate their current AI capabilities, identify gaps and areas for improvement, and develop a roadmap to build a more effective AI program.
Challenge
According to research, nine out of ten companies agree that AI represents a business opportunity for them. However, only a few companies are able to realize value using the technology.
Companies often misunderstand AI's unique requirements, treating it like standard technology. This leads to challenges in development, operation, and maintenance. Implementing AI demands organizational and cultural shifts, as well as new knowledge. Successful AI adoption typically requires significant organizational transformation and novel working methods to fully leverage its potential.
Approach
The AI Maturity Assessment is a three-stage process that consists of assessment, analysis, and workshop.
1. Assessment
Using the AI Maturity Assessment Tool to evaluate current capabilities, identifying strengths and weaknesses.
2. Analysis
Interpret the results, identify trends and benchmark against industry standards.
3. Workshop
Stakeholders discuss findings and co-create an AI capability improvement roadmap.
Results & Impact
Conducting the AI Maturity Assessment requires a relatively small investment of time and effort from the organization compared to the potential benefits. The assessment process provides valuable insights into an organization's readiness for AI, specifically within ten key dimensions:
-
Ambition
-
AI Use Cases
-
Organization
-
Expertise
- Culture
- Data
- Technology
- AI Ecosystem
- Execution
- Steering
By identifying gaps in these areas, organizations can prioritize investments in technology, talent development, and governance structures that support AI adoption and integration.
Moreover, the assessment process encourages organizations to examine their existing culture and identify any obstacles to AI adoption, such as resistance to change, fear of automation, or a lack of data-driven decision-making. By understanding and addressing these cultural challenges, organizations can create a more supportive and collaborative environment for AI innovation and transformation.
Back to homepage
While working for the Swedish Prison and Probation Service, I developed a business case for using AI to monitor inmate phone calls. This innovative approach addressed a critical challenge in the correctional system; how to secure inmates' rights to communication with the outside world while at the same time avoiding security risks and continued criminal activities.
While using AI for monitoring inmates' phone calls may initially sound intrusive, it actually has the potential to revolutionize correctional practices in a more humanitarian direction. By enabling more efficient and comprehensive monitoring, AI can paradoxically increase inmates' privacy and family connection time, as it allows for extended phone privileges while reducing the need for human listeners, thus better protecting inmates' personal conversations and supporting their rehabilitation process.
Challenge
The Swedish Prison system faced a dilemma:
- Inmates' right to communicate with family is recognized as a human right and crucial for rehabilitation.
-
Existing systems were becoming obsolete due to the phasing out of copper-wire networks.
-
Limited monitoring led to potential security risks and continued criminal activities.
- The suggested remedy of the challenge, manual monitoring of all calls (ideally 1-1.5 million call hours annually), would require over 700 full-time employees, an unsustainable resource allocation.
Approach
1. Leverage AI technologies including voice recognition, speech-to-text, and Large Language Models.
2. Develop a system capable of monitoring all calls in their entirety, verifying approved contacts, and understanding conversation content.
3. Implement a holistic solution addressing security, ethics, and operational efficiency.
4. Design the system to support rehabilitation efforts while preventing criminal activities.
Solution
An AI-powered monitoring system that:
- Uses voice recognition to verify callers' identities
-
Employs speech-to-text technology to transcribe conversations
-
Utilizes NLP to analyze conversation content, flagging potential security risks
-
Classifies calls as safe, suspicious, or non-classifiable, with human review for flagged calls
- Continuously improves through machine learning, reducing false positives over time
Results and Potential Impact
Economic:
- Potential to save over 700 full-time equivalent positions, allowing reallocation to more impactful roles
- Estimated to pay for itself within the first year, even with development costs of up to 100 million SEK
Ethical and Humanitarian:
- Increases allowed call time for inmates, supporting better family connections and rehabilitation
-
Enhances personal privacy by reducing human listening to calls
-
Enables quicker response to potential suicide risks or other urgent issues
- Supports the "better out" vision of Swedish corrections by balancing security with rehabilitation needs
Security and Operational
- Monitors 100% of calls, significantly reducing undetected criminal activities
-
Provides faster response times to security threats
- Generates valuable insights for improving rehabilitation programs and security measures
Innovation
- Creates a scalable, future-proof solution adaptable to increasing inmate populations
- Opens possibilities for collaboration with intelligence services, police, courts, and suicide prevention researchers
Conclusion
This case study demonstrates how AI can be applied to solve complex challenges in public sector operations, balancing security needs with ethical considerations and humanitarian values. It showcases the potential for AI to not only improve operational efficiency but also to support better outcomes for individuals within the correctional system.
Back to homepage
© 2024