Canada is in trouble when it comes to defence artificial intelligence (AI) and is positioned to become a cautionary tale of the early AI years. Although Canada is well-placed globally for AI research, development, funding, and implementation, the country’s defense force is badly positioned to embrace digital transformation. This is a consequence of the organization’s structure, history, and culture, rather than of technical shortcomings. Canada’s positive advances in AI are minor, fleeting, and scattershot, not representing systematic effort, and enjoying little priority. These positive advances are hardly worth talking about in comparison to the problems, which are potentially insurmountable. This chapter deviates slightly from the other chapters in this volume by presenting the historical background of Canada’s legacy problems for defense AI, before proceeding with assessments of the current state of defense AI in Canada.

The Department of National Defence (DND) is the arm of Canada’s federal public service related to defense. The Canadian Armed Forces (CAF) are the country’s uniformed military services. The “Defence Team” refers to both DND and the CAF, as well as other defense stakeholders.

1 Background: Canada’s Legacy Problems for Defense AI

Three historical developments heavily influence how DND/CAF behaves as an organization. First, a cultural overemphasis on mission success and operations primacy has contributed to a dysfunctional organizational mentality. Second, the transformation from three services to one service (unification) in the 1960s had the long-term effect of creating an enduring organizational fragmentation. Third, drastic budget and personnel cuts in the 1990s destroyed much of the organization’s collective memory and left it without the ability to manage its own data and information. None of these problems is directly linked to AI, but they collectively have created a culture that possesses major limiting factors for the implementation of AI systems in Canada’s defense establishment.

1.1 Cultural Problems

One of the basic assumptions of the Canadian Armed Forces is the idea of “operations primacy” or “mission first”: that mission success receives priority over everything else (Hansen 2022). This assumption dates back to the Second World War, and recently operations primacy was reaffirmed as a professional expectation in the new CAF professional ethos (Department of National Defence 2022a: 33; Leslie 2011). This expectation promotes a “get-it-done” mentality privileging mission accomplishment over wellbeing, and even accepting a degree of wilful disobedience so long as results are achieved (Rozema-Seaton 2019: 15). There is very strong cultural pressure in Canada for every mission to be no-fail. Justice Louise Arbour’s (2022) external review of the CAF noted that, “the long-established way of doing business in the CAF is anchored in operational imperatives that are often nothing more than assumptions” (Arbour 2022: 9).

The basic assumption of operations primacy has often been illustrated using the biological metaphor of “teeth and tail,” imagining the CAF as an animal whose fighting components are the “teeth,” while the supporting functions of the organization are the “tail.” The metaphor lionizes deployable, operational fighting elements, while denigrating and even vilifying the supposedly non-essential “tail.” The metaphor justifies cuts to “tail” areas (headquarters, administration, record keeping, data analysis, etc.) and assumes that it is always preferable to cut the “tail” of the CAF instead of its “teeth.” “We are going to have to reduce the tail of today while investing in the teeth of tomorrow,” wrote Canadian Army Commander Lieutenant-General Andrew Leslie in a 2011 capstone report on one of the CAF’s failed transformation efforts (Leslie 2011).

This basic assumption persists for perfectly logical reasons. Unlike the United States, where the armed services wield significant political clout, the CAF has little independence and no meaningful ability to shape political decision-making or strategy (English 2004: 88–89). The ability to complete those missions assigned to it in superb fashion is a protective mechanism for the CAF: government and public indifference to the military is so great in Canada that mission failure could feasibly provoke fresh rounds of budget cuts. The “teeth” must always be kept sharp.

However, the metaphor of the teeth and the tail is a faulty one. No higher animal consists only of teeth and tail; biological organisms are complex systems of systems, the most important of which is the guiding intelligence in the central nervous system. As historian Allan English has asked, “Is it better for the animal to lose a tooth or two, or a significant part of its brain?” (English 2016: 202–203). However understandable it may be, the primacy of operational requirements and the persistent privileging of “teeth” over “tail” has far-reaching consequences.

Operations primacy means that those elements of the CAF that are immediately deployable on operations, and specifically those that fulfil direct combat functions, are the most esteemed by the organization. The system rewards people and perpetuates structures that are operationally focused. The CAF suffers an organizational addiction to mobility, and expects key people (especially leaders) to move jobs every 2 to 3 years (Wakeham 2022). Promotions in the CAF hinge upon achieving short-term goals and then moving on, and privileges breadth and variety in postings—particularly on operations—over the development of expertise or depth. The CAF’s culture has inadvertently rewarded dysfunctional behavior. The cultivation of technical expertise in the CAF is not part of a career path that leads to promotion, and those who stay “locked” in jobs or locations for extended periods forfeit all organizational advancement (Department of National Defence 2024: 22; English 2011: 11). The CAF demands a willingness from its members to move jobs constantly to meet short-term operational imperatives.

1.2 The Long Hangover of Unification

Unification was a cost-saving program of the 1960s that transformed Canada’s traditional three military services (the Royal Canadian Air Force, the Royal Canadian Navy, and the Canadian Army) into one unified service containing different subordinate Commands. This restructuring was unprecedented among the military forces of the Western democracies, and was closely examined by policymakers in allied countries; notably, none of them followed the Canadian example (Irwin 2002: 41). Ironically, after dispensing with the three “strong services,” unification eventually created dozens of semi-autonomous, highly siloed organizational entities—the “Level 1 s” (L1s). The head of each L1 reports directly to the two “Level Zeros” above them: either the uniformed Chief of the Defence Staff (CDS), the Deputy Minister (DM) of National Defence, or both. The L1 system can charitably be described as federated, and more accurately as balkanized. Each L1 typically pulls decision-making authority together at the top of its own silo, to ensure control and to minimize the influence of both subordinate elements and enterprise-wide initiatives (Author Correspondence with LGen R. Crabbe 2022).

The way that data and information are handled within DND today—that is to say, poorly—is a consequence of this history. The L1s all operate semi-autonomously and have separate IT systems, independent procurement processes, and differing data requirements (Department of National Defence 2019: 29). As of 2019, the civilian Assistant Deputy Minister (Information Management) (ADM(IM)) “provides IM direction, procedures, and enterprise tools, [but] each L1 is responsible for implementing IM plans and activities within their respective operational areas” (Chief Information Officer 2019: 8).

1.3 “Records-Keeping Bedlam”

The second highly relevant development on Canada’s current path was the significant budget cuts and reductions in personnel following the end of the Cold War. The 1994 cut reduced the size of Canada’s Regular Forces by 32%, cashing out the peace dividend by reducing active military personnel from 89,000 to 60,000 as part of the Forces Reduction Program (FRP). However, there was a concurrent sharp rise in operational tempo, and many new high-intensity deployments of troops to the former Yugoslavia. Because that operational tempo had to be sustained and the government needed infantry battalions in Croatia and Bosnia-Herzegovina, the cuts were not distributed evenly.

In order to keep the fighting power of the organization sharp, the cuts fell on the back end, reducing administration and headquarters functions by half (Chief Review Services 2001a: 1). As Canadian historians wrote two decades ago: “In the relentless paring of military personnel in the [CAF] and civilian staff in [DND], inevitably many of the first positions to go have been the information handlers such as clerks, secretaries, archivists, and librarians,” whom the FRP deemed over-staffed and expendable in comparison to combat arms operators. These were the people who constituted the record-keeping and information management (IM) backbone of Canada’s defense establishment. With them gone, the well-disciplined analog record-keeping systems inherited from the Cold War military disintegrated, just as digital technology became widespread. The CAF let these personnel go before the IM tools were in place to support a smaller staff (Chief Review Services 2001b). E-mail and instant messaging took root as the preferred communication media, allowing everyday business to bypass a crumbling centralized record-keeping system (Lizotte 2019: 7). By 2001, observers described DND/CAF as already being in a state of “records-keeping bedlam” (English et al. 2001: 479). It has never recovered.

Canada’s DND/CAF is probably two decades behind where it needs to be on information management. The CAF’s 2022 Digital Campaign Plan accurately described the organization as being at the lowest stage of digital maturity:

Legacy analog systems and processes, stove-piped capability development, and generally low levels of digital literacy. Members of the CAF struggle to access data, analyse the data, and to generate decision-ready information supported by descriptive analytics. Data manipulation is predominantly done manually. Users adjust their behaviour and actions to existing systems and processes. (Canadian Armed Forces 2022: 6)

As of 2024, the CAF still has no centralized record-keeping system, and its inability to manage its own information has been the source of repeated scandals and professional malpractice for three decades (Desbarats 1997: 59–70; Sharpe 2000; Sabry 2015: 33; Berthiaume 2021; Arbour 2022: 54–55). Due to operational imperatives, key information managers within the L1 entities are typically double-hatted from their regular jobs, and are not dedicated specialists. Some powerful digital tools are available enterprise-wide within the CAF, but user training and instruction are not. Most areas of the CAF’s information ecosystem are effectively ungoverned, with networked shared drives, websites, SharePoint instances, mail servers, and document repositories holding huge volumes of ad hoc, disorganized content. “The result,” one CAF information management officer has written, “is an information environment which is untrustworthy, inefficient, that frustrates users, and limits the value the DND/CAF can glean from its own information” (Lizotte 2019: 2).

These problems must be addressed with culture change, as there is at present no culture of “working horizontally” among L1s at DND/CAF. However, the current structure, processes, and incentives are working against the necessary changes. In its 2022 digital strategy, the Canadian Army correctly described the current era in DND/CAF as an ongoing “digital winter” (Canadian Army 2022: 12). And external reviewer Louise Arbour strongly condemned the CAF’s information systems, writing that, “a more thoughtful approach would ensure that the sum of each organization’s data represents the whole picture … [but] with the current silo model focused on achieving individual organizational mandates, this is simply not possible” (Arbour 2022: 52).

1.4 Consequences of these Legacy Problems

These historical developments have contributed to a Canadian military culture and organization with notably toxic characteristics, and this presents substantial barriers for the adoption of defense AI. To return to the “tooth-and-tail” metaphor: when confronted with the choice of what to cut, the DND/CAF animal has historically preferred lobotomy to dentistry, shedding brain matter rather than risk losing “teeth.” This attitude, married to the cloistered, siloed structure of the organization and the dystopian state of its IM, is anathema to building a meaningful AI capability, especially at scale. By DND/CAF’s own reckoning, the basic prerequisites for understanding, developing, and fielding AI systems include research and development (R&D), agile project management, software as an essential capability equal to hardware, massive investment in digital infrastructure and information management, and application development across the enterprise. While DND has some promising R&D capabilities, none of the other elements are close to being met.

These historical, cultural, and organizational problems shape everything about the potential for Canada to develop a meaningful AI capability for its armed forces.

2 Thinking About Defense AI

The real tragedy of the legacy problems discussed above is that informed elements within DND/CAF are fully cognizant of the transformations that are necessary to implement defense AI. A significant amount of intellectual heavy lifting has been done on incorporating AI into the future of the CAF. In all cases where they have adopted explicit stances on AI, DND/CAF entities have maintained a commitment to keeping humans “in or on ‘the loop’” when it comes to decision-making, and there is no thought at present to permitting fully autonomous offensive weapon systems that “complete the loop” and engage targets without human oversight. Attempting to ensure that combat remains an activity featuring meaningful human involvement is a cornerstone of Canadian AI thinking (Department of National Defence 2024: 7). The CAF’s most-used leadership framework, the Pigeau-McCann Model, explicitly deals with how “cybernetic control” systems involving some degree of autonomy are feedback mechanisms, fundamentally lacking the properties of command (Pigeau and McCann 2002: 54).

Thinking about defense AI begins at the top, with Canada’s 2017 defense policy, entitled Strong, Secure, Engaged (SSE), which enshrines many of its formal military aspirations for the future. According to SSE:

Canada is committed to employing new technological capabilities in a manner that rigorously respects all applicable domestic and international law, and ensures full oversight and accountability. As a country that has led several efforts to advance human rights and establish new international norms, Canada is also well-placed to advocate among international partners for the highest standards. (Department of National Defence 2017: 55)

After SSE, the highest-level document concerning AI is the DND/CAF Artificial Intelligence Strategy, written in 2022 but only approved in early 2024, which is a comprehensive and soul-searching examination of the path towards defense AI for Canada’s military. It envisions five lines of effort that will allow DND/CAF to become an “AI-enabled organization” by 2030, “with ethical, inclusive, and trusted AI for interoperability and advantage in the battle space and improved stewardship in the corporate space.” These lines of effort are: (1) Key Capabilities; (2) Culture; (3) Ethics, Safety, and Trust; (4) Talent and Training; and (5) Partnerships (Department of National Defence 2024: 1). The AI Strategy is an outstanding effort to confront the many problems facing DND/CAF in this space. However, it has some problems of its own. The AI Strategy, like other organization-wide strategies before it, operates on the assumption of the siloed L1 structure of DND/CAF and hesitates to even broach the topic of overarching governance, preferring to devolve responsibility and authority to the L1s. No one entity is responsible; which in Canada often means that nobody is responsible. The authors completed the AI Strategy in final draft form in mid-2022, but it languished on the desks of distracted “Level Zeros” awaiting signatures for 18 months. This is a poor start for a strategy meant to modernize Canada by 2030; it is suggestive of what kind of priority AI can expect within a department filled with conflicting, competing demands between L1s.

The environmental L1s (land, air, maritime, special forces) have also put thought individually into their future integration of AI within their siloes. Since this is the level where implementation is most likely to occur, it is worth examining overviews of the thinking about AI on the part of the Canadian Army, Royal Canadian Air Force, Royal Canadian Navy, and Canadian Special Forces Command:

  • The Canadian Army

The Canadian Army is the most forward-thinking of the environmental L1s concerning AI. In its 2020 Modernization Strategy, the Army stated that it is their responsibility “to examine the potential of AI and machine data [sic] to transform some aspects of land operations, including exploiting data and information to produce intelligence and predictive modelling to support decision-making” (Canadian Army 2020: 51). The Army’s more recent addendum, the 2022 Modernization Vital Ground: Digital Strategy, reflects thinking on technological drivers, committing to the concept of “human-machine teaming” (integrating soldiers and autonomous systems) and the vital need to utilize “Big Data” in to process more information while decreasing the cognitive load on humans (Canadian Army 2022: 9). The Army’s Land Warfare Centre has written thoughtfully on how the adoption of AI by the Army “must proceed with caution and be informed by a realistic set of limits … nonetheless, if pursued and applied carefully, much of what AI offers generally aligns well with [Army] requirements” (Priems and Gizewski 2021: 43).

  • The Royal Canadian Navy (RCN)

The RCN issued its Digital Navy strategy in 2019, addressing a broad range of technologies, including AI, machine learning, automation, and data analytics. Digital Navy establishes the need to cultivate a “data-centric mindset” in the RCN, as “quality data will be a fundamental enabler” of success going forward. The RCN stresses three categories of defense AI applications: (1) Autonomous Things (advanced robots, autonomous vehicles, intelligent agents); (2) Augmented Analytics (using AI to enhance analysis of structured and unstructured data); and (3) AI-Driven Development (using AI in design process for naval equipment and systems). The Digital Navy strategy concludes with the promise to establish a Digital Navy Office “to facilitate the implementation and evolution of this initiative,” with a mandate including program alignment, communications, performance measurement, look-ahead functions, process enhancement, training, and contract vehicles (Royal Canadian Navy 2020: 6–17). The plan apparently remains intact, though no updates have been provided by the RCN on its progress.

  • The Canadian Special Operations Forces Command (CANSOFCOM)

The CANSOFCOM enjoys special privileges in terms of access, autonomy, and authorities that allow it to sidestep bureaucracy better than other L1s. In its 2020 strategic plan, Beyond the Horizon, CANSOFCOM provided broad details on Gradient Ascent, a digitalization and data analytics initiative designed to ensure the same competency in the digital space that its operators have achieved in the kinetic space (Canadian Special Operations Forces Command 2020: 31). Gradient Ascent saw CANSOFCOM investing in a complete modernization of its data architecture to permit data analytics at scale, supporting streams ranging from operations to intelligence to enterprise management, providing dynamic analytics products leveraging data from CANSOFCOM IT systems and sensors. Gradient Ascent reportedly “changed the game” in terms of how CANSOFCOM develops software and solves data problems through insourcing (Gonthier 2022b: 3–4). Little information is publicly available.

  • Royal Canadian Air Force

The last environmental command, the RCAF, has published the least on AI. The organization’s Future Air Operating Concept is almost a decade old. Recent Aerospace Warfare Centre publications offer few comments on technology and nothing specifically about AI. (Goette 2020). The RCAF Journal has published a few articles relating to AI between 2019 and 2022, but has just entered a three-year hiatus due to funding shortfalls. Most of the RCAF’s work on AI—if it is happening—is being done outside the public eye.

3 Developing Defense AI

Canada represents a potentially rich site for AI research and development, and for decades the country has been a locus of AI work. Some of the seminal work in the field came from academics at Canadian universities. The Government of Canada’s Pan-Canadian Artificial Intelligence Strategy has invested hundreds of millions in AI research institutes, education, and student talent development. However, Canada’s relative position in the world of AI research continues to drop, and Canadians only hold about 0.5% (and dropping) of the world’s nearly 1 million AI-related patents (OECD AI 2021). Between 2018 and 2022, one-third of Canadian AI firms either permanently closed or were acquired by foreign firms (Araya 2022: 5).

The main L1 within the Canadian defense community that is working on AI technologies today is Defence Research and Development Canada (DRDC). DRDC’s role within the Defence Team is to provide leadership and advice on issues of science and technology, to engage and collaborate with a network of domestic and international partners, and to exercise functional authority to “ensure coherent of defence and security science, technology and innovation investments” (Defence Research and Development Canada 2022). The organization serves as the bridge between Canada’s AI potential and its defense applications. DRDC defense scientists carry out research internally on behalf of or in partnership with other L1s, and commission contract research from approved third-party vendors, sometimes working in partnership with them. DRDC has invested heavily in research related to AI and machine learning for decades, and a considerable corpus of relevant work associated with and funded by DRDC has accrued: 45 defense research projects and reports relevant to AI are on file as being completed between 2009 and 2022.

Presently, the main route for pursing defense AI within DND/CAF is through DRDC’s Innovation for Defence Excellence and Security (IDEaS) program, announced in 2017. The intention behind IDEaS is to “bring together academics, industry, and other partners to form collaborative innovation networks.” IDEaS is a competitive funding model intended, in part, to bypass the cumbersome and archaic CAF procurement system (discussed later in this chapter) by streamlining academic and industry technical cooperation with the military. According to DRDC, seven in every ten proposals from academia and industry for IDEaS grants each year involve AI components (Directorate S&T Strategic Partnerships 2022).

Many of the IDEaS related to defense AI are through the program’s Competitive Projects funding mechanism. “The Competitive Projects element funds projects fast,” according to the program website. “It advances promising technology quickly through a phased approach.” It begins with up to 6 months of funding, after which there is an option for a further 12 months of funding at a much higher rate. After this, DND also has the option of pursuing the project further using non-IDEaS funding through Science & Technology Solution Advancement. (National Defence Undated) IDEaS’s “Spot the Hack” challenge, issued on behalf of the RCAF, studies cyber vulnerabilities in the Military Standard 1553 bus used by RCAF aircraft avionics networks; many of the bids which received first- and second-round funding, including those from CAE, Palitronica, and Queen’s University, included AI learning agents for intrusion detection purposes (National Defence 2022).

Outside of DRDC, a potentially significant development avenue for AI is a relatively new DND program called “Mobilizing Insights in Defence and Security” (MINDS). MINDS is governed by DND’s office of the Assistant Deputy Minister (Policy) responsible for the development and management of defense policymaking. The MINDS program is based on the idea that “policy and decision-making are strengthened when assumptions are challenged, and diverse viewpoints are considered.” MINDS provides collaboration opportunities between DND/CAF and the academic defense and security community, allowing for bespoke briefing engagements, targeted engagement grants, support for emerging scholars, a “rapid response mechanism” for addressing evolving priorities, and the creation of Collaborative Networks. (Mobilizing Insights in Defence and Security (MINDS) 2022). One Collaborative Network, the Security-Policy Nexus of Emerging Technology (SPNET) based out of Concordia University in Montreal, specifically focused on AI as an emerging policy issue.

The lynchpin of Canada’s defense AI ecosystem, however, are the private interests working as contractors for DND/CAF. Given the internal problems facing Canada’s military, the importance of working with trusted AI vendors is magnified. The Treasury Board of Canada Secretariat maintains a list of interested AI suppliers. Federal departments can use these pre-qualified suppliers to launch streamlined procurement processes for technologies and services, for up to CAD9M before taxes. As of December 2023, there are 121 companies on the Treasury Board List, ranging from very small startups to large enterprises such as Palantir Technologies, Amazon Web Services, IBM Canada, and Microsoft Canada (Treasury Board of Canada 2022).

4 Organizing Defense AI

The most pressing question for organizing defense AI in Canada involves governance: who owns the AI problem, and to whom are they beholden and responsible? Canada’s DND/CAF is reluctant to commit itself to governance standards for the use of AI and is equally reluctant to propose firm internal governance models. Given the department’s “federated” L1 system, the lack of broad governance mechanisms to organize the use of AI will likely default to each L1 within DND/CAF doing as it pleases in developing AI, with little higher accountability.

4.1 External Governance for AI

Federal entities in Canada are bound by the government’s 2019 Treasury Board Directive on Automated Decision Making, which ensures that AI systems are used responsibly by government institutions (Treasury Board of Canada 2019). This Directive is a supposedly mandatory policy instrument applied throughout federal government departments. The Directive applies to the use of all systems that make, or assist in making, recommendations or decisions. As part of the Directive, every department must complete an Algorithmic Impact Assessment (AIA) of any AI system prior to production, or whenever system functionality changes. The AIA assesses a systems’ impact based upon factors such as the systems’ affect on the rights, health, wellbeing, and interests of individuals and communities, as well as sustainability, reversibility, and duration (Riley et al. 2022). Based upon responses to risk and mitigation questions, the AIA assigns an impact rating and requires publication of the AIA on an open government portal. Having a person make the final decision in a system does not exempt departments from complying with the Directive; any system that provides advice to public servants who make the final decision is within its scope (Beshaies and Hall 2021).

The Directive on Automated Decision Making appears robust on paper but contains escape clauses that dilute its effectiveness. The Directive applies only to the “external” eservices of government—services offered to individuals or organizations by government—and does not apply internally within departments, a “glaring oversight” in the governance regime (Scassa 2022). The team carrying out a periodic review of the Directive recommended expanding the scope to include internal as well as external systems (Bitar et al. 2022).

DND/CAF does not believe the Directive applies to it, and thus far has ignored it. DND/CAF’s AI Strategy notes that any new AI systems used by DND should be developed and implemented “in accordance with applicable laws, policies, and guidelines.” However, it also warns that “because of their application to defence and national security, many DND/CAF use cases [of AI] will fall outside the guidance provided by the Treasury Board Secretariat, and the gap between the development of AI and other emerging technologies and legislative and policy coverage only continues to widen.” The AI Strategy carefully discusses how it will be “aligned with” the Directive in considering risks, without actually stating that it is subject to the Directive (Department of National Defence 2024: 19–21). Recent cases of DND using AI-driven hiring services to help fill executive positions within the department became a small scandal when Canada’s Privacy Commissioner accused DND of skirting the rules surrounding the use of AI. DND did not submit an Algorithmic Impact Assessment to the Treasury Board: a DND spokesperson claimed that because “final decisions” were not being made by AI, the department did not feel obliged to complete the Treasury Board’s algorithmic assessment (Cardoso and Curry 2021). This excuse was neither in the spirit nor the letter of the Directive on Automated Decision Making. In fact, as of 2023 DND/CAF have never made a submission to the Algorithmic Impact Statement, strongly suggesting that DND will resist having AI governance decisions imposed by other parts of the federal government.

4.2 Internal Governance for AI

The AI Strategy does not state how governance of AI will function internally to DND/CAF, though there are hints. Most likely the federated L1 entities will be left to govern their own uses of AI. The AI Strategy proposes the creation of a Defence AI Centre of Excellence for Canada to “accelerate AI experimentation and scaling across the Defence enterprise,” creating a hub of AI expertise. However, centers of excellence are increasingly commonplace and establishing them falls under Mme Arbour’s category of the “flurry of activities” that DND/CAF tends to do that do not necessarily accomplish anything. The AI Strategy does not propose to actually invest this center of excellence with governance responsibilities. Instead, it argues that DND/CAF must “Vest decision authorities for AI at the lowest appropriate level to encourage innovation.” Given how Canada’s military works, the “lowest appropriate level” will mean the L1s.

The office of the Assistant Deputy Minister (Data, Innovation, Analytics) (ADM (DIA)) is in one L1 silo of DND, reporting directly to the Deputy Minister of National Defence. The office of Assistant Deputy Minister (Information Management) (ADM(IM)) and Assistant Deputy Minister (Defence Research and Development Canada) are in different siloes. On 6 December 2022, DND announced the removal of the ADM(DIA) office as an L1, merging it with the Directorate of Knowledge and Information Management (DKIM) under the ADM(IM) to stand up the Digital Transformation Office (DTO). The ADM (IM) L1 is now called the Chief Information Office and is charged with managing both IM and data analytics enablement “in support of initiatives like machine learning and artificial intelligence” (Matthews 2023). CAF insiders suggest that the ADM(DIA) had not accomplished anything since it was stood up in 2017 and will be little missed, so this reorganization may correct some of the divisions over IT that have plagued the organization.

Perhaps with cross-cutting, general-purpose capabilities such as AI, there is nobody in Canada who can lead radical change except the “Level Zeros.” The thinking within DND/CAF may well be that trying to govern AI is like trying to govern electricity, and this may prove to be the correct interpretation. However, it seems equally likely that the “balkanized” character of DND/CAF’s organization means that devolving power and responsibilities to the L1s was likely to be the outcome no matter what, and it is not necessarily aligned with good practices. The question of internal governance and organization for AI within DND/CAF is therefore an extremely difficult one.

5 Funding Defense AI in Canada

The Government of Canada has spent lavishly to develop a thriving homegrown AI ecosystem. As mentioned earlier, DND’s Defence Research and Development Canada (DRDC) is the primary delivery agent for defense science and technology investments, and their main vehicle has been the Innovation for Defence Excellence and Security (IDEaS) program. The government has committed to investing CAD85M per year for 20 years into IDEaS, significantly exceeding the funding for their civilian and commercial Pan-Canadian Artificial Intelligence Strategy. IDEaS allows for five funding mechanisms to assist Canadian “innovators” in addressing defense issues: competitive projects up to CAD1.2M; “Innovation Networks” of up to CAD$1.5M; contests on approved topics; sandboxes for field testing; and test drives for high-readiness ideas.

The key advantage of IDEaS initiatives is that they individually involve quite small expenditures. In Canada, the Treasury Board has a CAD10M threshold for its Organizational Project Management Capacity level, meaning that a project valued less than CAD10M “will be exempt from much of the oversight and rigour” of the standard procurement system and does not need to prepare a project complexity and risk assessment. This is an extremely important point. AI projects valued at less than CAD10M are quite easy; anything beyond that becomes an order of magnitude more difficult (Bedley 2021: 39). The IDEaS initiative is vital because of the procurement context that it has been designed to bypass.

5.1 Canada’s Procurement Disaster

Canada’s DND has, for over a decade, been effectively disabled from procuring major technological systems. This is part of the wider, slow-motion disaster that is Canadian defense procurement. Beginning in 2008, when significant new capital investments occurred, major capital projects became “jammed up” and have never become unstuck. 70% of procurement contracts for Canada are now overdue or seriously delayed. The procurement system requires the achievement of concord between three different government departments. Making matters worse, Canada’s Treasury Board (rightly) deems DND a fiscally risky institution and has forced the adoption of manufacturing industry best practices and standards on the military procurement process, meaning that all projects must follow the same Project Approval Directive (PAD). This process is also designed to prioritize investment in the Canadian defense industry over actually building the capabilities of the armed forces. A recent study showed that the average length of information technology-related procurement projects in DND was 9.6 years, with some IT projects open for over 16 years.

The PAD can bypass much of its own process when DND decides there is an Urgent Operational Requirement (UOR) that streamlines procurement to address short-term operational deficiencies. However, the UOR approval requirements “demand that the project directly affect combat operations and contribute to a life-saving capability.” This makes the UOR ideal for CAF operational requirements but there is no chance that such a mechanism can be used to help with the broader digital transformation.

There are recent signs that DND/CAF is hoping to create new, more expansive relationships with industry partners to speed up procurement. This shows promise, but will require significant shifts in mindset and the ceding of some elements of control of process by DND/CAF (Fawcett 2023).

The AI Strategy cites the need to improve the procurement process to support the development and acquisition of AI as a critical element of its plan, but this appears to be a forlorn hope. Procurement in Canada will not be fixed soon. Even if it was, money in DND/CAF flows towards improving tactical capabilities that allow the CAF’s commands to maximize their fitness for operations on the near horizon, not towards more general, longer-term, or enterprise-wide capabilities. And so many of DND/CAF’s aspirations for defense AI are likely going to have to cost CAD10M or less each if they want to be realized this decade.

6 Fielding and Operating Defense AI

There are few instances available in the unclassified realm of defense AI systems being fielded and operated within DND/CAF. Most projects are still being researched, under development, being tested, or finding limited tactical or business applications.

A few defense AI initiatives are certainly underway, as detailed by Canada’s Open Government transparency site for procurement. These included a noncompetitive contract for “artificial intelligence / inference systems (R&D)” awarded to IMRSV Data Labs in Ottawa in December 2021. No tender description is available, but products such as IMRSV’s “Anvil Crucible” intelligence fusion platform use machine learning to automate data analysis tasks such as establishing relationships between entities and making predictions for key variables (IMRSV Data Labs 2022). This is almost certainly the software that has been used by the CAF’s Joint Targeting Intelligence Centre, and was employed in support of the evacuation of Afghan personnel after the fall of Kabul in 2021, discussed below (Department of National Defence 2024: 29).

Operations primacy continues to shape Canada’s investment priorities. In its recent Modernization Strategy, for instance, the Canadian Army decided to prioritize the modernization of their tactical C4ISR (command, control, communications, computers, intelligence, surveillance, reconnaissance) capabilities, while putting off the investments needed to support a transformation into a “digital army” until a minimum 2025–2030 timeframe. Their stated assumption is that “modernization efforts must be undertaken concurrent to force employment on operations – there will be no pause” (Gonthier 2022a: 2–3; Canadian Army 2020: 26). This is a textbook application of the operations primacy assumption. The C4ISR capabilities are certainly important and are likely to involve AI components; several ISR-related projects are going through the IDEaS process at present. However, these will be narrow tactical applications of the technologies, embedded within existing organizational stovepipes and without any wider integration or governance.

There are a few other AI systems that are now in service, or will be shortly, with DND/CAF today. The RCN has a large contract with Kraken Robotic Systems Inc., out of Newfoundland, to provide remote mine hunting and mine disposal equipment, and an additional contract for underwater sound equipment (Department of National Defence 2022b). This acquisition includes autonomous underwater vehicles and the use of Kraken’s AquaPix synthetic aperture sonar that allows for embedded automatic target recognition and data exfiltration (Kraken Robotics Inc. 2022). Since 2016, DND has awarded large contracts to IBM Canada to provide both the Canadian Forces Health Services Group and the Canadian Institute for Military and Veterans Health Research with the data infrastructure and cognitive computing capabilities to conduct advanced “big data” analytics related to healthcare research for service members (Bélanger and Cramm 2016). The Calian Group consulting firm has also been awarded large sums for data remediation and marking of serially managed material, a project to make more DND assets machine-readable that has been ongoing, with stops and starts, since 2016 (Department of National Defence 2020).

One use case publicized by DND/CAF as an effective application of AI, however, is in fact an indictment of the organization’s current data ecosystem. The AI Strategy cites the example of an AI tool (probably Anvil Crucible) which maps networked relationships. It reads:

In 2021, CAF was presented with an urgent request from Immigration, Refugees and Citizenship Canada (IRCC) for the names of Afghan personnel who had worked for Canada and now needed evacuation. This data existed, but as large quantities of paper files that would take dozens of people hundreds of hours to review manually. With permission from JITC and support from the vendor, the team spent a weekend scanning the documents and used the tool to extract thousands of names for IRCC. (Department of National Defence 2024: 29)

While the story suggests that applying an “agile AI-based solution” was a major accomplishment, the problem it solved was entirely manufactured by the failure of basic digitization and stove-piped information within DND/CAF. Perhaps this points the way towards what might be the best possible use of defense AI for Canada: rescuing the organization from its own poor practices.

7 Training for Defense AI

The Canadian Armed Forces are presently facing an existential threat in the form of a human resources crisis among uniformed personnel. As of December 2022, the CAF was 10,000 uniformed members short of its authorized strength, with a catastrophic 10% annual attrition rate (Hansen 2022). The Chief of the Defence Staff issued a directive on reconstitution was issued on 6 October 2022 that scaled back non-essential operations and activities to “recover and rebuild (reconstitute) the organization,” but noted that the personnel shortfalls had already “severely impacted the organization’s ability to deliver professional and collective training.” The directive reads that “we will need to make difficult choices about our readiness levels, capacity for sustained operations, as well as our level of commitment to all activities, while continuing to deliver strategic effects for the [Government of Canada]” (Eyre and Matthews 2022). A year later, things are no better. In late November 2023, the commander of the Royal Canadian Navy gave a brutally honest assessment that the RCN was in a “critical state” and would be unable to meet its readiness commitments for the next year and beyond. This is the result of a severe shortage of sailors and technicians, over 20% in many roles. The commander affirmed that the CAF recruitment wing has failed to meet its targets for more than 10 years, and that current attrition rates are unsustainable (Ritchie 2023).

What does reconstitution mean for defense AI? The DND/CAF AI Strategy highlights that “Talent and Training” requires its own line of effort: “We must identify and plan for our workforce needs, we must cultivate AI readiness amount our existing people, and we must find new ways to bring critical skills into the enterprise – and to retain and use them” (Department of National Defence 2024: 22). However, this will be an uphill struggle. Recent polling has found low and declining levels of enthusiasm among Canadians for joining the CAF, which is mirrored in disastrously low recruiting and high attrition rates. While this creates serious problems along the breadth of the organization, they pose special problems for nurturing an internal talent pool with fluency in the discipline of AI. If the values and principles of DND/CAF do not appeal to prospective servicemembers with skills in AI and digital technologies, then the “Talent and Training” conundrum is likely unsalvageable, because those skills will fetch an unmatchable premium on the open market in North America.

The AI Strategy says that it will meet the talent and training challenge first with a review of DND/CAF workforce needs for AI: “identify the skills, competencies, and personnel required to implement AI successfully. This must include not only subject matter experts in AI, but also staff whose roles support the AI lifecycle, including civilian and military leadership.” The AI Strategy also urges DND/CAF to identify priority AI workforce needs and either develop or procure training curricula to meet them: “This review should consider training needs at all levels, and the exploration of new options for both academic and professional training to ensure a talent pipeline for future needs.” Finally, it flags the urgent need to “Explore and identify processes to recruit and retain AI talent, and to utilize it where it is needed.” Some of the solutions the document offers are the creation of technical Reserves, short-term exchanges, and more flexible career pathways allowing for the attraction of tech-savvy talent above entry level in the CAF (Department of National Defence 2024: 22–23). These are all potentially viable paths, and the AI Strategy does a good job identifying problems and systemic barriers.

The barriers, however, are likely too many. The assumption of operations primacy has far-reaching consequences for the training and posting cycles. Promotion at the mid-level ranks is disproportionately determined by success in commanding operations (Hansen 2022). Personnel in non-operations career trajectories are unlikely to “have legs” in the CAF rewards system and are unofficially barred from holding the seniormost leadership positions within the organization (Kelley 2020). “While the CAF recognizes its own need for AI skills,” the DND/CAF AI Strategy confirms, “it often struggles to make use of those it already has. Members have described their specialization in AI and related fields as career-limiting and speak of having to choose between remaining within their technical field and [choosing] a career path that would lead to promotion” (Department of National Defence 2024: 22). It is therefore difficult to see how CAF will be able to train a reliable AI talent pipeline internally, particularly with uniformed servicemembers.

DND/CAF is already embroiled in the wicked problems of training and talent retention, which creates an unpromising milieu for harnessing defense AI. The special problems of talent retention in a hot tech market are aggravated by the CAF’s basic assumptions of operations primacy and unlimited job mobility discussed earlier. Neither hard-won technical expertise in a specific field, nor a focus on computer science occupations are rewarded within the CAF. Changing that will require an overhaul of the rewards and promotions system, and such an overhaul will meet fierce resistance from vested interests. At minimum, DND/CAF may have to place more reliance upon the civilian public servants on the DND side of the organization, which creates a different set of problems. DND/CAF may have no choice but to seek external partnerships and contracting for their AI needs, as the development of an internal talent pipeline poses, at present, a major problem without obvious solutions on the horizon.

8 Conclusion

Technologies classified as AI are in use in Canada’s Department of National Defence and Canadian Armed Forces today, and their incorporation will continue at a modest pace in the years to come. However, the organization faces serious challenges to any kind of digital transformation or large-scale adoption of defense AI systems. These problems are primarily historical, cultural, and organizational, rather than purely technical. As a country, Canada is well-positioned to engage with the transformative effects of AI. The country’s armed forces, however, are not.

Historical and cultural trends within the Canadian military have left the institution heavily decentralized, and its L1 organizations (including the main force employers) are stovepiped from one another, strongly preferring to develop capabilities in isolation. DND/CAF’s enterprise-wide information management has been a disaster for three decades and the organization’s data stewardship is a constant source of scandal and embarrassment. Progress in AI is uneven across the organization. Governance appears to be an afterthought. While agile procurement is possible for minor projects, right now it is extraordinarily challenging for projects above a certain threshold. Closer partnerships and heavy reliance upon commercial interests and industry are likely the only meaningful way forward, and these partnerships have historically been harder to achieve and maintain in Canada than in the United States (Department of National Defence 2024: 26–28).

What will defense AI look like in Canada over the next 10 years? Barring major culture and organizational change, it will likely look something like this: small-scale AI projects, spread throughout the L1 siloes, with almost no cross-pollination between them. These AI systems will be focused on hyper-specific operational and tactical uses cases faced by the various commands. The amount of computing hardware used in training cutting-edge AI models has increased by a factor of 10 billion since 2010 and is doubling every 6 months. This growth wildly outstrips improvements in hardware, and so AI labs are making up the difference by buying more chips. Costs for training high-end AI models are already astronomical (Scharre 2023: 35–37). DND/CAF will not keep up, particularly with a cap of roughly CAD10M for timely technology purchases. If the organization wants cutting-edge AI models, the digital infrastructure, training data, and learning model architecture for them will need to be sourced from outside the organization as DND/CAF information management practices will need reform before they can be of use at scale. DND/CAF will probably also need outsourced engineers to run them, since the future of cultivating internal tech talent within the CAF looks very bleak. Barring major change, it is difficult to see how Canada can even nominally keep a hand in the defense AI game without extensive new public-private partnerships, the sort that generally run counter to how Canadian defense procurement traditionally works. In short, something fundamental must change before Canada makes any serious advance towards integrative defense AI across its military institutions. Realistically, many fundamental things must change, and it seems unlikely that they will change in a great hurry. The fact that it took the “Level Zeros” eighteen months to sign off on the excellent DND/CAF Artificial Intelligence Strategy is suggestive of what kind of priority AI can expect going forward in a department crippled by so many conflicting demands and crises.

This brings us back to the “teeth-and-tail” metaphor for operations primacy. Although AI technologies are broad and cross-cutting, enabling them as strategic assets will require transformational investment in what is derogatorily referred to as the “tail” of the organization. In a small defense force such as Canada’s, this will require uncomfortable trade-offs. The idea that new combat equipment should be (further) delayed or cancelled in favor of back-end computing capabilities or enterprise-wide information management will be anathema to the L1s. But the DND/CAF “animal” of the metaphor is in deep trouble, and its teeth are falling out on their own for lack of strength in the rest of the organization. If the global adoption of defense AI proceeds on its present course, Canada is going to be left well behind both adversaries and allies. The list of what must be done is daunting, and almost every point on it will be contested. But Canada will mostly likely continue to muddle through as it presently is, developing minor tactical AI-related capabilities while neglecting the more serious problems until it can do so no longer.