Aspen Digital

What is Public A.I.?

Public AI describes an approach to artificial intelligence development and use that promotes the common good. Public AI has three key features: public access, public accountability, and permanent public goods.

• Public Access – Certain capabilities are so important for participation in public life that access to them should be universal. Public AI provides affordable access to these tools so that everyone can realize their potential.
• Public Accountability – Public AI earns trust by ensuring ultimate control of development rests with the public, giving everyone a chance to participate in shaping the future.
• Permanent Public Goods – Public AI is funded and operated in a way to maintain the public goods it produces permanently, enabling innovators to safely build on a firm foundation.

Public AI can play a key role in helping Canada and the world confront its biggest challenges by expanding access to the tools of responsible innovation. For more information about public AI, see Public AI: A New Approach to Public-Interest AI Investment, a whitepaper from the Public AI Network.

Background on Canada’s National A.I. Institutes

In 2017, the government of Canada put in place the Pan-Canadian Artificial Intelligence Strategy, a wide-ranging policy backed by a $568 million financial commitment to accelerate the research and commercialization of AI across Canada. Among the largest recipients of the funding are Canada’s three National AI Institutes: Amii in Edmonton, Mila in Montreal, and the Vector Institute in Toronto. These Institutes carry out cutting-edge research on a variety of AI subfields, from AI and health to robotics.

Why are Canada’s National A.I. Institutes fertile ground for the development of public AI? 

Much of the research carried out by Canada’s National AI Institutes is explicitly geared toward advancing the public good, and many projects contain the seeds of the three cornerstones of public AI, even if they don’t fully embody all three yet. For instance, through its AI4Humanity initiative, Mila uses public funds for its AI Against Modern Slavery (AIMS) project, which uses advanced machine learning techniques to combat forced labor. Under UK and Australian law, large companies are required to publish annual reports outlining their efforts to eliminate slavery from their supply chains. During Phase 1 of its operations, AIMS developed AI-based tools to swiftly analyze these reports and build a “ground truth” database to make information extracted from these statements transparent. Now in Phase 2, AIMS is working on equipping governments, NGOs, investors, and citizens “with the knowledge they need to make informed decisions about the companies they engage with.”

This use case exemplifies the spirit of public AI, though its implementation of the three pillars is incomplete. AIMS is a publicly-funded, not-for-profit project aimed at the public good. Phase 2 explicitly aims to make the use of its tools as widely accessible as possible. Its results, data, and models are open-source, and it encourages the public to use them to replicate its results and improve on its techniques. These features satisfy the public access criterion. However, they do not yet satisfy the permanent public good criterion, since the project does not provide a sustainable mechanism to ensure permanent accessibility. Additionally, in the absence of a clear feedback loop that gives the public a direct voice in the system’s governance after its implementation, it falls short of the third pillar of public AI: public accountability. Giving the public a say in the system’s development is a crucial first step, but true accountability requires public governance throughout the life cycle of the product.

Other Mila projects under its AI4Humanity initiative are in the same boat: they implement some (but not all) the pillars of public AI, and they do so imperfectly. For example, its Artificial Intelligence Alignment for Inclusion (AIAI) project was created to engage citizens in the co-development of urban projects to ensure that they have a say in the design of their own communities. It aims to create a prompt-based AI image generator for third party facilitators during the community engagement process of an urban development project to “improve the likelihood that landscape architects incorporate community needs into their designs.” This tool is being developed in consultation with vulnerable communities in Montreal, and is also using an open-source development model. The public value is clear, and there are clear mechanisms for community feedback. Once again however, the public only has a say in the design phase rather than throughout the life of the system.

These are just two of many such examples. All three of Canada’s National AI Institutes are working on many projects explicitly aimed at furthering the public good, many of which can be described as public AI-adjacent. Through these projects, the Institutes have recognized the value of creating AI-based public goods, and meaningfully engaging the public in their development. These instincts can and should be taken further. By explicitly undertaking projects that fulfill the three pillars of public AI, Canada’s National AI Institutes can more fully realize the goals they have already spent significant amounts of time and resources on.

How can Canada’s National A.I. Institutes lead the way in the development of public AI?

The public AI-flavored work that has already been done suggests that the National AI Institutes are fertile ground for further progress. First, the Institutes already have the necessary infrastructure, namely world-leading principal investigators, motivated graduate students, and compute. The projects described above suggest that many of these researchers are motivated to seek the public good through their work. They also have an intuitive appreciation for public AI’s three pillars, though they have not always fully realized their potential. Second, shifting this existing infrastructure within the Institutes toward public AI would have a low administrative burden, since the basic structure of the public AI research projects would be the same as other research projects, just with a shift in focus.

In the future, this groundwork can be transformed into world-leading public AI research and development infrastructure. Future governments can set aside funding for the National AI Institutes that is explicitly reserved for projects that fulfill all three pillars of public AI. They can require researchers to make the fulfillment of these criteria explicit in their grant applications, incentivizing recipients who are already motivated by the public good to take that commitment to its logical conclusion and directly build public AI. This would in turn help build a public AI talent pipeline as demand for research assistants, laboratory assistants, and graduate students in the field increases. It would also increase the prestige of the field, which would help attract top talent. Finally, as the field grows in both numbers and prestige, it will naturally extend into the wider world of AI research. For instance, many of the researchers at Amii also work at BC universities. Directly investing in public AI at Canada’s National AI Institutes would thus likely have a multiplier effect, drawing researchers not affiliated with the Institutes into the field through their colleagues. These investments could therefore be the seeds of a pan-Canadian public AI research network.

Anthony De Luca-Baratta is a Public Service Fellow and MA International Relations Student at the Johns Hopkins School of Advanced International Studies (SAIS). He currently works as an intern at the Macdonald-Laurier Institute, Canada’s premier national think tank, where his research focuses on Canadian national defense and U.S.-Canada relations. He is also the Managing Editor of the SAIS Review of International Affairs, a position he has held since May 2024. During the Fall 2024 semester, Anthony worked as a Google Public Policy Fellow at Aspen Digital, where his work focused on public AI in the Pan-Canadian Artificial Intelligence Strategy.