I am not, it should be acknowledged, an expert on the merits of frontier AI models. <\/p>\n
I am, however, a reasonably seasoned observer of government ministers proclaiming solemnly that they have just made a very prudent investment of taxpayers\u2019 money, on the most favourable of terms, as AI minister Evan Solomon did this past weekend<\/a>.<\/p>\n Sometimes it works out: the Trans Mountain pipeline expansion may yet turn out to have been a profitable investment by Ottawa.<\/p>\n Often, it does not. <\/p>\n Does anyone have confidence that the $50 billion or so in loans and subsidies provided by federal and provincial governments to create an EV supply chain<\/a> will prove to be money well spent?<\/p>\n Since time immemorial, governments have proclaimed that they couldn\u2019t afford not to inject public money into commercial projects: be it the Sturgeon refinery in Alberta, the Bombardier C-series jet in Quebec, or a sports car factory in New Brunswick<\/a>. The ability of governments to pick winners has an extremely checkered history.<\/p>\n And so we come to the deal to which Solomon was referring: the $1.6-billion federal government equity investment in Canadian AI company, Cohere<\/a>.<\/p>\n The Toronto-based company builds secure AI systems for businesses and governments. Ottawa\u2019s involvement will allow Cohere to scale up and compete in the global AI market, Solomon said. <\/p>\n \t\tEditor\u2019s Picks\t<\/p>\n Cohere aims to raise $7 billion from the feds and other investors like Germany\u2019s Schwarz Group, chipmaker NVIDIA and Bell Ventures, the venture capital arm of BCE. <\/p>\n Solomon would not disclose the commercial terms, but the new money values Cohere at nearly $36 billion, making the government\u2019s stake equate to just short of five per cent. <\/p>\n The move gives Canada \u201ca seat at the table,\u201d Solomon said, so that the potential upside will not benefit someone else.<\/p>\n It will keep Cohere headquartered in Canada and require the company to invest $1.7 billion in computing infrastructure in this country. <\/p>\n It all sounds too good to be true. Let\u2019s hope it isn\u2019t, because the stakes are high. <\/p>\n Canada has, quite reasonably, focused on its digital sovereignty in recent times, keen to ensure it is free from coercion as AI is incorporated into all areas of modern life. <\/p>\n There have been calls by people such as Jim Balsillie, the former chair of Research in Motion and the current chair of the Council of Canadian Innovators, for more robust economic nationalism. In particular, he has urged the government to make investments in key strategic industries like AI.<\/p>\n Yet, he says he doesn\u2019t have a good feeling about the Cohere investment. <\/p>\n \u201cThere are red flags all over this one and it raises the prospect of surprises of rent-seeking behaviour,\u201d he said. <\/p>\n \u201cUnfortunately, it undermines the imperative for strategic government investments.\u201d <\/p>\n Concerns in the broader tech community extend from Cohere\u2019s performance, to the tightly knit (perhaps too tightly knit) Canadian tech C suites. For example, Patrick Pichette, the head of the new federal organization Digital Transformation Canada, is a partner in Inovia Capital, an early investor in Cohere, and he sits on the AI company\u2019s board. <\/p>\n Dominic Barton, the chair of the Invest in Canada federal agency, is also a partner in Radical Ventures, an investor in Cohere. <\/p>\n Neither may have been involved in the financing round, but the optics are not good and potentially both could benefit from Cohere\u2019s enhanced valuation. <\/p>\n Canadians should probably be worried less about Cohere\u2019s success than its potential to fail. <\/p>\n Prime Minister Mark Carney has described Cohere as \u201cworld-leading.\u201d<\/p>\n Yet, the Artificial Analysis Intelligence Index<\/a> has Cohere\u2019s \u201cCommand A\u201d large language model scoring poorly against its peers.<\/p>\n The company would no doubt claim that it is unfair to rank a smaller and more targeted model against the large-scale general models such as Gemini, Claude and GPT-6.<\/p>\n But this is the company that Canada has now explicitly crowned as its national champion. <\/p>\n Sovereign AI means the ability to develop and govern AI using its own infrastructure. <\/p>\n In a report that came out earlier this year \u2014 Sovereignty By Design<\/a> \u2014 written by the Munk School\u2019s Sean Mullin and Jaxson Khan, it was recognized that Cohere offers an important hedge against foreign dependency.<\/p>\n But it noted that it is impossible for it to compete across all the dimensions on which foundation models are now judged. <\/p>\n Cohere has built a niche supplying large organizations, but consumer use of American models is creating a dependence that enterprise and government procurement policies will not be able to address.<\/p>\n Was the Cohere investment the best use of $1.6 billion, if the goal is freedom from coercion? The Munk report suggested that sovereign cloud infrastructure, owned and operated in Canada, should be the biggest strategic priority, to ensure Canadian data stays in this country.<\/p>\n Another alternative to explicitly declaring Cohere a national champion and attempting to ensure it has resources and government support to remain competitive, would be the pursuit of a multinational frontier AI partnership with like-minded allies, the report suggests,. <\/p>\n It is clear that the government\u2019s chosen course is a much bigger gamble than the minister would have you believe. <\/p>\n Canada could be relying on the AI version of the East German Trabant car, while the Americans cruise past us in the equivalent of Audi 80s.<\/p>\n jivison@criffel.ca<\/a><\/p>\n Twitter.com\/IvisonJ<\/p>\n \t\t\t\t