The potential for AI in government is massive. Through digitization of services, intelligent process automation, and leveraging the many various data sources that fall within the purview of the public sector, AI presents an unparalleled opportunity to dramatically improve our public institutions. Our team recently got back from Toronto, where we attended both tracks of the REWORK AI in Government and the Deep Learning summits. In addition to the numerous interesting and productive connections we made, the two days were full of engaging talks from leaders in AI research such as Geoffrey Hinton, the Chief Scientific Advisor of our partner the Vector Institute, Adam Oberman, and Andrew Moore, industry players from across the spectrum, such as Huawei, WeWork, and Snap, and public policy experts from the Government of Ontario, World Bank, and International Labour Organization. Here are my key takeaways from last week: The Difference Between Research and Production The biggest practical takeaway I want to share with public sector officials looking to implement AI is to differentiate between AI in research and AI in production. In research, the goal is generally to optimize for accuracy, whether that comes in the form of an F1 score or loss minimization. However, AI in the production environment requires a more diverse set of considerations, such as speed, restrictions on memory and computing messy data, with the ability to adapt to new data. Applied research scientists must focus on much more than only model research, and work hand in hand with engineers and developers to build a well integrated deployment. This includes measures to enable data and model auditability, which in the words of Diego Oppenheimer, CEO of Algorithmia, means knowing “who called which model, when, and with what data”. Critically, these production requirements will often necessitate trade-offs that make or break the viability of the solution for the real world. Applied research scientists working on AI solutions in the public sector should be comfortable working with many different models, managing and comparing their performance, appropriately scheduling retraining to avoid bias, and access permissions specific to different scenarios. Governments should aim to work incrementally, first deploy within a constrained sandbox environment, and perform rigorous algorithmic impact assessments to ensure they are maximizing for metrics that reflect the public good, as opposed to figures in a research paper. Panel on “Understanding the World Through Video” with Roland Memisevic, Graham Taylor, Tegan Maharaj, and moderator Evan Nisselson. The Legal and Ethical Implications of AI We have written before and will continue to write about the importance of keeping the legal, ethical, and social implications of AI top-of-mind when deploying solutions. Nowhere is this more important than in the public sector. During a panel titled “Should Government Regulate AI”, the panelists all agreed that “yes, government should regulate AI”. But when it came to the money question: “how?”, nobody could present a comprehensive framework. As we’ve said previously, regulating rapidly changing technology is like shooting at a moving target: if you act too early you will miss, and if you act too late your strategy will be outdated. The one point where they agreed was that in order to be proactive, working groups focused on developing regulations need to source perspectives from a variety of diverse stakeholders in the ecosystem. Particularly, Abhishek Gupta, Founder of the Montreal Ethics Institute and an ML Engineer at Microsoft, pointed out that more practitioners are needed in these high-level conversations to keep discussion grounded in the reality of the technology and what its capabilities are today. Additionally, these discussions need to be happening globally to ensure that practices across North America, Europe, and Asia are aligned and mutually inclusive. “Should the Government Regulate AI” panel with Dan Ciuriak, Abhishek Gupta, Gosia Loj, Chris Calabrese and moderator Steven Kuyan. The Role of Government Good governance carries the obligation of building societies that create opportunity for all of its citizens. In the context of AI, this means that governments must ensure social and economic gains from the technology are evenly distributed across all layers of society. For players in the public sector, this means focusing AI solutions on areas that maximize collective well-being, whether it’s optimizing transportation infrastructure and traffic flows, extracting meaningful insights into public policy, or making education more accessible. Lawrence Eta, the Deputy Chief Information Officer for the City of Toronto, joined us for a fireside chat to share his experience leading the digital transformation of North America’s 4th largest city. From his perspective, the benefits of AI need to be clear to everyday people to avoid disillusionment and a loss of confidence in the technology since it will touch on every aspect of our lives. The learnings that he shared from a pilot project to create a proof of concept to address medical issues within the homeless community resonated deeply with me, and was exactly the kind of AI for Good implementations we need to see more of. It was encouraging to hear that the City of Toronto is taking a long-term, product focused approach that will allow its digital infrastructure to remain modular, flexible, and scalable in the years to come. Fireside chat with Lawrence Eta and Brent Barron. The Future of AI in the Public Sector The overarching feeling at the conference could be best described as one of reserved optimism. Since the attendees were mostly experts at the forefronts of their fields, whether machine learning, deep learning, enterprise or the public sector, we were all enthusiastic about the potential of AI to revolutionize the processes and organizations we interact with every single day. We’ve been happy to participate in an active conversation with our own governments at various levels, who have taken a proactive approach to interacting with industry and exploring the unique opportunities presented by AI as well as its potential legal, ethical, and social implications. At the same time, as a part of the AI industry, we recognize the challenges we will need to overcome and the work that is left to be done. Which, for a fast-moving company like ours, with scientists who relish good problems, only makes the prospect of an AI-driven economy and society all the more exciting. Interested in starting your AI journey? Contact us today.