Sunny, a recent Apple TV+ limited series (now since cancelled!) starring Rashida Jones and set in a near-future Japan, follows a grief-stricken gaijin Suzie Sakamoto whose Japanese husband and son vanish in a plane crash. Balancing sci-fi, dark comedy, and thriller elements, Sunny (and Suzie’s foil, an eponymous house robot) explores grief, loneliness, and the complexities of human connection, while weaving in AI’s potential for both good and harm.
One of the many futuristic technologies we’re treated to in Sunny is the Babel fish-like instant-translation device that allow the American Suzie (who has never quite tried to adapt to life in Japan) to communicate with anyone speaking Japanese. In Star Trek, the universal translator seems fantastical (and, apparently still in experimental development around 2151); in Sunny, it looks like something from this year’s CES.
Real-time translation tech has come a long way in recent years thanks to AI, natural language processing (NLP), and machine translation (MT). Tools like Meta’s SeamlessM4T and Google’s AudioPaLM now handle direct speech-to-speech translation in over 100 languages, and DeepL lets me read reports from the le Gouvernement du Québec in English.
As someone who studies digital governance from a Canadian perspective, I think such translation technology—where language barriers dissolve seamlessly through AI intermediaries—is going to raise uncomfortable questions about what official bilingualism means in the Canadian context. Beyond the Canadian context, I think MT presents an interesting edge case for the provision of public services through AI and challenges principles of democratic pluralism.
Canada: bilingual by origin, but not comfortably bilingual
As someone who grew up in anglophone Canada and resisted (to my ongoing regret) the opportunities to learn French, let’s just say that MT technology is something I’ve been waiting for for some time (in addition to—owing to an egregious choice to not take grade 10 typing—being able to speak words into a microphone, and have them appear magically on my computer screen).
Canada’s French/English bilingualism is enshrined in its Constitution. Since 1867, both English and French have equal status in federal courts, the Parliament of Canada, and the Province of Quebec (though French has been that province’s sole official language since 1974). The 1982 Charter of Rights and Freedoms entrenched bilingualism, and the federal Official Languages Act supports the development of linguistic minority communities.
Beyond these legal foundations, bilingualism remains central to Canadian identity and governance. No aspiring national leader can hope to get far without being bilingual, and advancement in the public service or appointment to a high court can turn on your command of both languages.
Yet the current bilingual administration system suffers from a lack of capacity, with delays in the translation of important communications, failures to meet requirements to provide public services in both official languages, and the time and money spent on things like sending unilingual public servants to language school. Across Canada, about ¼ of the Canadian population speaks French as their first language, with about 18% of Canadians being French/English bilingual (to be fair, Francophones are five times more likely to be bilingual than Anglophones). Nonetheless, support for official bilingualism in Canada remains high, with approximately 78% of Canadians either strongly or somewhat supporting the policy.
Machine translation to the rescue?
Canada is by no means the only bilingual or multilingual country. Globally, 43% of the world's population is bilingual, while 17% are multilingual, and 55 countries maintaining official bilingual status, such as Canada, or multilingual such as Belgium and Switzerland. In the United States (which does not have an official language at the federal level), approximately 20% of the population are bilingual or multilingual.
Several countries and institutions are already leveraging MT to support their multilingual status and provide public services across languages. The European Commission's Directorate-General for Translation (DGT) has implemented AI-powered services including eTranslation, eSummary, and eBriefing tools. This allows for the issuing of communications across the EU's 24 official languages, enhancing multilingual communication efficiency. Croatia, Estonia, Iceland, Latvia, and Malta (co-funded by the Connecting Europe Facility) have partnered on the open-source National Language Technology Platform (NLTP) which offers governments and citizens access to free machine translation and multilingual services adapted to their respective languages and local public administration terminology.
For Canada's bilingual systems, the promise is equally compelling: MT technologies have the potential to significantly reshape the work of government translators and facilitate smoother operations in Canada's bilingual federal system through real-time translation, consistency in terminology, increased efficiency, improved accessibility, and providing public services to official language minority communities. The technology could also extend beyond French and English, offering vital support for First nations communities by supporting Indigenous language preservation. In a recent review of the potential for AI to contribute to the reform of public service delivery in Canada, I pointed toward MT as one way to help address the administrative challenge of Canadian bilingualism.
Former Clerk of the Privy Council (the head of the Canadian public service) Michael Wernick recently posed this provocative question in light of the advance of MT: “language models are getting you faster and more accurate translation of text and documents and simultaneous interpretation of conversations and meetings. Why would you need to train somebody to speak French or English if the technology is there for instant translation and interpretation?”
The Government of Canada is not ignoring this potential. Consider the Treasury Board's “Guide on the Use of Generative AI”, which includes translation as a potential AI application. While the Translation Bureau's own guidelines from April 2024 take a more nuanced stance—recommending machine translation only for non-protected, non-specialized texts where translation errors won't run afoul of Official Languages Act compliance—they do acknowledge that MT can support interpreters and enhancing communication efficiency. And informally, it’s apparent that federal public servants are using AI tools to reduce frictions in working with colleagues in their preferred language and improve their ability to get their work done regardless of the language. I certainly use MT to collaborate with Francophone colleagues, emailing seamlessly with fellow committee members that alternate between French and English (where my artificial bilingualism arguable keeps up with their real bilingualism).
Functional vs. cultural bilingualism
As MT becomes even more seamless, rapidly advancing towards the real-time translation seen in Sunny, bilingual (or multilingual) governments will be presented with a fundamental question: is bilingualism about the function of being able to communicate with someone in a different language, or does bilingualism have a deeper, cultural, value?
The cracks in the traditional approach to the bilingualism edifice founded on Canada’s origins have raised concerns from the country’s Commissioner of Official Languages about whether a machine approach to artificial bilingualism is sufficient for preserving real bilingualism. In calling for a modernized Official Languages Act, the Commissioner emphasizes the principle of “substantive equality” in service delivery between English and French, noting that services must be of equal quality in both languages, that substantive equality requires more than mere literal translation, and there needs to be cultural and contextual understanding—based on human interaction and cultural sensitivity —that AI systems may not fully replicate. That document’s emphasis on quality, human interaction, and substantive equality suggests that while MT might play a supporting role, it should not be seen as a primary solution for meeting Canada's official languages obligations. This caution reflects an understanding that language is more than just information transfer. While certainly wise, I fear the Commissioner is tilting at windmills.
Consider the broader philosophical argument that every technology diminishes the human ability it was meant to support, though we repeatedly ignore the warning signs. Plato in his Phaedrus (~370 BCE) recounted Socrates’ criticism of writing that it would weaken memory because people will rely on external symbols instead of internal recollection. Jacques Ellul argued in The Technological Society that once a technology is introduced, society reorganizes around it, causing prior skills to decline (consider phone-based maps and GPS, which have arguably decreased our natural navigation skills and spatial awareness). Or more recently, Sherry Turkle who argued that automation and digital communication tools are diminishing our interpersonal skills and emotional intelligence. And we are now seeing research that suggests that an over-reliance on GenAI might be weakening our critical thinking skills.
Advances in MT open up a fascinating paradox at the heart of Canada's bilingual identity: could the very tools meant to bridge language divides ultimately erode the incentive to maintain true bilingualism? Ask yourself this question if you are unilingual: why would you invest the time and effort in learning another language if a technology can make you functionally (though artificially) bilingual? Sure, there are many arguments in favour of learning a second language: bilingual individuals have improved problem-solving skills, memory, and executive function; socially, bilingualism fosters empathy and cultural understanding, strengthening communication in diverse settings; economically, multilingual professionals often earn higher salaries and access broader career opportunities; and bilingualism might even delay cognitive decline and neurodegenerative diseases like Alzheimer’s. But, if you’re like me and looking forward to an upcoming trip to Paris, you’ll spend a bit of time on your Babble app perfecting your “Je voudrais un café et un croissant, s’il vous plaît”, but will happily let an AR tool like Google Lens help to read any menu. While there are cognitive, interpersonal, and cultural benefits to being bilingual, in a world where MT translates perfectly between languages, the incentives to learn other languages will likely decrease.
And what of the fate of the world’s endangered languages? Of the world’s 8,325 languages that UNESCO tracks, only 65 are deemed “safe”, the remainder marked as endangered, vulnerable, or not in use. Canada has of over 70 distinct Indigenous languages across 12 unique language families spoken by First Nations, Métis, and Inuit communities. However, many Indigenous languages are not being passed down to younger generations, heightening the risk of language loss. The Indigenous Languages Act of 2019, which aims to support the revitalization and maintenance of Indigenous languages across Canada, makes no mention of the potential of MT focused instead on supporting real bilingualism. On the one hand, it will be a heartening to see young Indigenous people be able to ‘converse’ with their elders for the first time, even if mediated by MT. But whether MT will mean struggling and extinct Indigenous languages are revived, or that the declining number of Indigenous language speakers declines to zero even faster, is uncertain.
An analogy for other public services
Assuming the tides of technology are too powerful to resist or legislate against, the question of artificial bilingualism in Canada is an interesting edge case that illuminates where AI in support of public services can be useful but, surreptitiously, how such technologies are shaping the kind of society we are unwittingly building. Will an artificially bilingual Canada be a stronger nation because French and English will be able to communicate in real-time, or will part of Canada’s raison d’être be eroded, accelerating external threats to its viability as a country.
Beyond the niche use of MT, the application of AI to the machinery of government is proceeding at pace. One of the central premises of the Elon Musk-led “Department of Government Efficiency” (other than moving very fast and breaking many things) is that artificial intelligence and other applications of computer technology will be instrumental in accelerating the efficient delivery of government services and generally improving public administration.
If we do replace public servants with AI agents, and by extension replace the street-level bureaucrat with a personalized service agent for every citizen, we will solve one problem (i.e., the tendency of street-level bureaucrats to deviate from the consistency of centralized implementation) while possibly creating the corollary (i.e., the absence of situational nuance and creative problem solving). Being mindful of the value of human public servants—supported by, but not over-ridden by, AI technologies—would recognize that the goal of public administration is not merely efficient service delivery, but the preservation and enhancement of democratic pluralism through administrative practice.
As with Socrates and writing, few will heed warnings about the incremental advance of digital governance (and as a frequent flag-bearer in that parade, it would be inconsistent for me to argue against taking advantage of the potentials for AI to transform government administration). Yet I will argue for human-in-the-loop (HITL) checks and balances to preserve the intended democratic pluralism in our public services, ideally avoiding the tyranny of “sorry, computer said no”.
