When I give public talks and training on AI in government, I am often asked why Canada has no “Department of Artificial Intelligence” to govern AI. Some jurisdictions, like the United Arab Emirates, have a minister for AI, and a few others, like the UK, have a small government office focused on AI; but to my knowledge there are no jurisdictions with a department of AI.
The call for a department of AI is a well-meaning response to the growing importance of AI and highlights the gap that exists between technology adoption in government and the much higher proficiency of the private sector with this technology. A dramatic state of affairs should be addressed through decisive action and, so the logic goes, that decisive action should be nothing short of updating the machinery of government to create a new department or agency directly focused on the governance of AI. While the sentiment is understandable, the idea of creating a department of AI is riddled with misunderstandings and needs to be handled with care.
First, few applications called “AI” have much in common with one another from a technical standpoint, at least not in the way that the term “AI” is generally used today. For instance, the type of AI used in security cameras to identify suspects is a very different technology than the AI that is used to support advanced translation services. There is a large and common overestimation of AI’s shared characteristics across applications. In reality, when we talk about AI, we are talking about many different things.
What’s more, a great many of the new applications we call AI are fairly linear, albeit impressive, advances from a pre-existing technology or process, most of which already had a governance framework. In the world of policy, the permitted applications of AI with regard to traffic cameras have much more to do with the existing rules governing traffic cameras than with, say, an AI application that can play chess.
In that sense, AI is less analogous to a field like astronomy and more analogous to something like electricity: both AI and electricity are found in a wide range of fields and would be poorly served by an oversight body created on the basis of all potential applications. Imagine creating a “department of electricity” to be responsible for regulating every instance and application in which electricity is found, or a “department of computers” to oversee all conceivable applications of computing power. Not only would such an organization be unwieldy, but it would be unlikely to have clear goals and purposes.
Making new departments
From an administrative and organizational effectiveness standpoint, it’s not clear that AI capacities should be housed together in a single department. The computer science and data science at the heart of AI are functionally very different from the operation of things like postal services or seaport governance, where activities must be clustered together to achieve economies of scale in the use of a particular piece of machinery or infrastructure.
In contrast to the technologies of the industrial age, which are highly location dependent and benefit immensely from clustering, software and data science applications do not depend on location, or at least not nearly to the same degree. Even if significant benefits of co-location did exist for AI uses in the federal government, being housed in the same departmental structure is no guarantee that staff would even be housed in the same building; most departments are split across multiple sites, and even across multiple provinces. It’s hard to see how a new “department of AI” would ever resemble an army of AI professionals saluting from the same cubicle farm.
Concentrating all the government of Canada’s AI capacities in a single institution would also come with negative side effects. With the transfer of those capacities to a centralized institutional vessel, other departments would be left with very little or no AI capacity of their own. AI would be monopolized by the single department of AI, which would then support the AI projects of other departments as needed.
There is a precedent for such an arrangement: the strategy behind the creation in 2011 of Shared Services Canada, which sought to put all government IT services under one roof. For a variety of reasons this change in machinery has been held responsible for harming the quality of government technological services, and the Trudeau government assessed SSC as being in need of “renewal” only nine years after its founding. It’s hard to imagine that a duplication of this approach for AI would fare much better.
Our newsletter about the public service.
Nominated for a Digital Publishing Award.
Alternative institutional structures
Instead of a megadepartment of AI that houses all potential applications, it is possible to imagine taking a broader approach, permitting applications and governance to be decentralized but with a complementary concentration of expertise and resources that can be lent out to other departments as necessary. But this modified vision of new AI machinery is unlikely to be uniquely different from existing institutions like Statistics Canada, which already has renowned strengths in the data, statistics and modelling techniques that are central to AI. Statistics Canada needs to be brought up to speed on AI, but that can be done without creating a new department.
Perhaps a department of AI could better be viewed as a much more focused federal R&D instrument, responsible for the sorts of specialized, leading-edge AI research that might benefit from the concentration of facilities for supercomputers and the like. An institution with this sole mandate does not exist at the federal level and could even make some sense at a technical level; however, the idea is a complete non-starter due to jurisdictional issues. While the National Research Council does operate in this space for the federal government, most public R&D functions and funding are passed off to Canada’s 260 or so post-secondary institutions, which are nominally independent and overseen by provincial governments.
Bringing all of these AI activities under one (figurative) roof would require a huge bureaucratic street fight, to strip powers and functions away from their hundreds of existing owners. With such high cost in political and administrative capital in order to unlock a dubious and unknown benefit, it’s implausible that such a change would ever be attempted. Certainly, there may be no solitary department addressing AI, but it’s not obvious that having one would be an improvement.
If it ain’t broke…
Under the existing arrangement, overarching AI policy for the government of Canada is the purview of central agencies and Innovation, Science and Economic Development, with political leadership provided by the new cross-departmental minister of state for digital government. Meanwhile, responsibility for delivery and applications is housed in the line departments, which use AI applications and are closest to their effects. This set-up has been successful so far in practice; Canada ranks among the leading countries in the world for AI preparedness. Even the minister of digital government is not assigned sole responsibility for AI.
The idea of a departmental “stovepipe” for the new and exciting field of AI is tantalizing, but the state can meaningfully commit to doing more about AI in other ways. Creating a new department would be an incredibly expensive and disruptive undertaking, adding new layers of administration and technical complications. Machinery of government changes are in fact an immense distraction from the everyday business of government; very little AI governance will get done while staff are being moved and new mandates debated.
Due to its onerous nature, changing the machinery of government is seldom regarded as a way of acting either quickly or decisively. Petronius Arbiter, a Roman administrator writing on governance, is generally credited with the observation “We tend to meet any new situation by reorganizing, and what a wonderful method it can be for creating the illusion of progress while actually producing confusion, inefficiency, and demoralization.” While AI is growing in importance and needs better governance to match, a machinery change should take place only if there are clear problems with the existing institutional structure and clarity that the change will represent an obvious improvement. It’s not clear that either of these would be true in the case of AI.
Do you have something to say about the article you just read? Be part of the Policy Options discussion, and send in your own submission. Here is a link on how to do it. | Souhaitez-vous réagir à cet article ? Joignez-vous aux débats d’Options politiques et soumettez-nous votre texte en suivant ces directives.