Community Spotlight
Leading by Example to Improve Civic Life
The City of London’s AI Lab is paving the way for local governments in Canada (and beyond) to adopt machine learning
In the Community Spotlight series, TDS Editors chat with members of the data science community about the exciting initiatives that help push the field forward. Today, we’re thrilled to share Elliot Gunn’s conversation with Mat Daley, Matt Ross, and Blake VanBerlo at the City of London’s Municipal Artificial Intelligence Applications Lab.
Mat Daley (LinkedIn) is the Director, Information Technology Services with the City of London. Mat has 20 years of management experience in the public and private sector and has had the pleasure to work in five distinct industries. As an entrepreneur, Mat was involved in the leadership of three start-up companies. Mat completed his undergraduate studies at the University of Waterloo, his graduate studies in Public Administration at Western University and is a designated Project Management Professional.
Matt Ross (LinkedIn) has spent the past decade working in the intersection of technology and social good. Having founded a non-profit, a technology startup, working as an independent data science contractor, and co-founding the AI Lab at the City of London. Matt pursued degrees in Philosophy from Huron University and Chemistry from Western University.
Blake VanBerlo (LinkedIn) is the founder of VanBerlo Consulting, a firm that specializes in machine learning applications research and development. Having previously completed a bachelor’s in software engineering at Western University, Blake is currently a graduate student at the Cheriton School of Computer Science at the University of Waterloo.
It’s uncommon to find an in-house AI Lab at the municipal level of government. I would love to learn more about the team. Could you share what the Lab does at the City of London? What does a typical day look like?
The Municipal Artificial Intelligence Applications Lab is a part of the Information Technology Services division at the City of London. Due to some personnel changes, the lab is in a transition phase. The lab looks for opportunities to leverage AI and analytics to deploy low-cost, low-risk solutions to improve life for the city’s citizens and/or employees. We are currently a very small group, but we attempt to be nimble due to the varying nature of different projects.
A typical day for the lab involves frequent communication with stakeholders and developers. The stakeholders are typically a part of or related to city services. Since our projects are driven by business requirements, we are constantly in touch with stakeholders about ongoing projects, receiving feedback and providing education. Developers, who may be internal employees or external consultants, spend their days working on projects in accordance with the evolving requirements set out by stakeholders. We are also consistently hunting for new projects and approaching folks in the city who may be open to collaboration.
The article you published with TDS was the first blog post that discussed a project that the Lab undertook to apply machine learning to a civic issue (also covered in the CBC). Could you share more about how this project came about?
The chronic homelessness prediction system was indeed the first initiative that the lab undertook to apply machine learning to a civic issue. The idea for the project came from a business systems analyst who works in the city, and it was born at a hackathon aimed at discovering opportunities to apply machine learning. This individual works closely with homelessness prevention stakeholders, and he wondered if there could be a way to use machine learning to make predictions based on information contained in the Homeless Individuals and Families Information System (HIFIS) database, which contains information on individuals accessing homelessness services. The idea then evolved into predicting chronic homelessness from these records. After demonstrating that this approach had promise with a small prototype, we engaged stakeholders in Homeless Prevention. Discussions with them highlighted the importance of explainable machine learning. From there, the project was in a constant iterative cycle with developers and stakeholders.
We wanted to share this work because doing so entails potential benefits to other municipalities. As a municipal government, we believe that we have a duty not only to London, but to the well-being of all communities. By shedding light on our work, we were increasing the chances of other municipalities coming across our work. Our solution is applicable to other locales in Canada due to the prevalence of the HIFIS application. To make our work as accessible as possible, we included detailed instructions in the GitHub repository on how to get started with the code.
Could you tell us about a few projects that have seen promising results?
We are particularly proud of the chronic homelessness prediction project. The project has been deployed and is now available to case workers who interact with clients seeking homelessness services.
Another project we are proud of is our work in water demand forecasting. Given records of past water consumption, we developed a model that forecasts citywide water demand over the coming years. This work may be instrumental in infrastructure planning and budgetary forecasting. A key aspect of this project was that the model we employed is inherently interpretable; as a result, water management can better understand and trust in its predictions.
How does the government identify and prioritize promising projects given its resource constraints? How do you align ML projects with the government’s broader objectives?
Municipalities deliver a wide range of services; as a result, there are several streams to apply machine learning. The longstanding, general goal of government is to become more effective and efficient. Better resources and more efficient service delivery directly connect to opportunities for successful machine learning projects. We therefore assess potential projects by their ability to deliver on this goal. For example, the visceral and ubiquitous nature of homelessness and housing security motivated us to apply machine learning to improve delivery of homelessness services.
How does the team work to mitigate risks from deploying AI solutions and interventions with the vulnerabilities associated with marginalized communities?
First and foremost, we engage in constant discussions with stakeholders to identify outstanding risks. We put stakeholders first by ensuring development is led by them and consider data scientists to be supportive. For example, in our first project, we worked closely with homeless prevention (managers and case workers) to ensure that data was treated respectfully.
We follow multiple strategies to prioritize information security and privacy. First, we heavily collaborate with the city’s privacy commissioner to discover and apply best-in-class security and privacy considerations. We require explicit consent for data usage and de-identify data before it is accessed by developers or sent to a deployment environment.
To determine other possible risks, we strive to adhere to the government of Canada’s Directive on Automated Decision Making. To this end, we complete the federal government’s Algorithmic Impact Assessment Tool to ascertain the impact level of our projects.
Lastly, we prioritize explainable machine learning to promote transparency. Models that are not inherently interpretable are always coupled with explanations produced by an explainability method. Explanations ensure that stakeholders understand why models make the decisions that they do. They also may reveal biases present in the model that could trigger subsequent development iterations.
Are there projects that might be better addressed by governmental institutions in Canada rather than for-profit efforts in AI and DS?
Some services are much better delivered by government agencies than by private enterprises. We believe that services that benefit society fall in this category. At the municipal level, this includes delivery of clean drinking water, wastewater management, and public transportation. At the provincial level, examples could include healthcare, education, legal services, policing, fire management, and emergency services. To be clear, we do not believe that the private sector cannot play a role — we simply believe that some services lend themselves more naturally to government delivery.
Another consideration is that the datasets used for projects pertaining to municipal services reside within government bounds. Data sharing is a complex undertaking, especially when personal data may be shared with private organizations. Further, the ubiquity of data science tools and open source code entails free use by government agencies, reducing governments’ dependence on the private sector for these projects.
What role(s) do you see the team playing as a leader in doing data science for social good?
We see ourselves as leading by example. We share codes and methods resulting from our projects with the intention of helping other municipalities to adopt the same tools.
What kind of writing in DS/ML do you enjoy, and what would you like to see more of?
We enjoy reading articles and papers regarding end-to-end machine learning projects and we would like to see much more being produced. By “end-to-end”, we mean coverage of all steps of a ML project: from research to deployment. The data science community is awash with tutorials and ideas, but rarely do we see publications on operationalization, sustainability, and resources.
In what areas do you hope to see the DS/ML community continue to grow in the future?
In the coming months/years, we hope to see a greater push towards explainable ML. Advances in deep learning have produced more powerful and expressive models at the cost of increasing opaqueness. In municipal government, transparency is key to delivering equitable services to citizens. We would like to see a greater push toward application of existing explainability methods, development of novel explainability methods, and reconsideration of inherently interpretable modelling approaches.
We hope that the community places increased focus on sustainability and maintenance of machine learning systems. We have several questions pertaining to these topics that we are continuously seeking answers for regarding current and upcoming projects. What are the best practices for keeping systems relevant as new data becomes available? What if the data distribution and/or feature set shifts over time? Who is supporting deployed systems and how are they being maintained?
We hope that the community continues to adopt an open approach and that code continues to be made available open source. Our projects (and surely those of other governments) would not be possible without freely accessible machine learning software packages.
Finally, we would like to see increased pursuit of AI governance. Legislation regarding the deployment of systems delivering public services is paramount to the equitable and safe delivery of such services.
Curious to learn more about data science at the City of London’s Municipal Artificial Intelligence Applications Lab? Here are the two articles mentioned in the interview that share in-depth case studies of projects that utilize machine learning for social good.
- “Forecasting Long-Term Daily Municipal Water Demand” (April 2021, TDS): Blake VanBerlo shares a case study of a project by the Lab that focused on obtaining daily forecasts for long-term citywide water demand.
- “An Open-Source Interpretable Machine Learning Approach to Prediction of Chronic Homelessness” (May 2021, TDS): Blake VanBerlo explains how the team applied machine learning to the prediction of chronic homelessness in the City of London, from research to deployment.
