Rapid AI Prototyping for Policy Innovation: Transforming Government Decision-Making

As a No10 Innovation Fellow, I have the opportunity to build key AI projects within the government and contribute to the Prime Minister's priorities. Over the past seven months, I have created several AI-based solutions to address various government needs. In this blog, I will focus on three key projects that I have developed:

1. A mobile application for citizen reporting
2. A platform for policymakers to visualise and predict urban scenarios
3. A secure AI assistant for civil servants

In addition to these tools, working alongside teams from i.AI and the Cabinet Office, I have built platforms to detect outliers in urban areas, understand crowd patterns, and address riots and criminal activities across UK cities, and developed a deep learning-based anomaly detection model to identify severe interactions in polypharmacy and prescriptions, helping to prevent systematic errors and improve citizens' quality of life. I am currently building an AI-powered product for housing allocation and policy recommendations.

These projects have the potential to greatly improve the government's ability to tackle challenges promptly and make evidence-based policy decisions. This work demonstrates the practical impact of AI research on policy implementation, showing how AI can be effectively applied to real-world governance issues.

About me

My research and engineering background, which enhances my quick prototyping abilities in product development, is rooted in Artificial Intelligence, Spatial Data Science, and Urban Analytics. I am a lecturer in Spatial Data Science at the Institute for Spatial Data Science at the University of Leeds. Previously, I was a project investigator and postdoctoral researcher at The Alan Turing Institute, and I completed my postgraduate studies and PhD in Engineering at UCL. My research and product development focuses on using AI, especially computer vision and deep learning, to study the built environment, human movement, and behaviour in urban areas. This work builds on my earlier research on human mobility at the Alan Turing Institute and my development of URBAN-i and the AI-enabled camera, URBAN-i Box. My diverse background in architecture, urban design, and engineering offers a unique perspective when applying AI to urban and policy challenges. 

A few details of major projects I have worked on are given below:

Product 1: Street.AI: An on-device AI mobile app for reporting issues in the built environment (working prototype)

Street.AI, a potential mobile app for citizen reporting, could establish a new channel for public engagement, allowing citizens to actively participate in community improvement. This has the potential to enhance the responsiveness of local governments to public needs and concerns. The AI-driven tools I have developed can pave the way for more personalised and efficient public services, improving citizen satisfaction and trust in government institutions. Running computer vision algorithms directly on users' phones keeps data on their devices, ensuring privacy. This approach allows only necessary information to be sent from the user's phone, eliminating the need for additional cloud processing expenses. By processing data locally, the system enhances user privacy, reduces data transfer, and avoids extra costs associated with cloud-based computations. Street.AI is currently purely a functioning prototype, the team are considering how this might be rolled out for use, but it demonstrates a useful capability.

Product 2: MapSpeaker: Visualising, querying and building future planning scenarios of the built environment

MapSpeaker is a platform powered by spatially aligned Large Language Models (LLMs) that generates maps and creates future scenarios based on user inputs. Users can interact with the tool to display various types of maps, such as population distribution in the UK, areas with high crime rates, regions affected by flooding, or potential locations for new housing developments. This versatile tool allows policymakers and planners to quickly access and visualise spatial data, supporting informed decision-making in urban planning and development.

Product 3: Mimo: Secure, offline AI assistant MacOS

Government operations often have significant concerns about data privacy. My development of a secure offline AI assistant, similar to an offline version of ChatGPT, addresses these concerns. The offline operation of Mimo ensures that sensitive information stays protected and out of reach for unauthorised users. This approach also makes the system immune to cloud service disruptions and eliminates the need for additional cloud computing expenses during deployment. Operating locally on the device provides enhanced security, reliability, and cost-effectiveness compared to cloud-based alternatives.

Future Outlook

The work completed so far has established a solid base for ongoing impact, showing the practicality and benefits of quick AI prototype development in government environments. Building on these achievements, my future plans aim to broaden and strengthen the influence of AI in public service and policy creation. The reputation, resources, and knowledge gained from my initial projects directly support these plans.

I intend to expand successful prototypes by moving these effective product prototypes from test projects to full operational use across several government departments. This expansion will allow for more widespread adoption and impact of these AI-driven solutions in public service delivery and decision-making processes.