Kineta H. Tober is a European cyberpsychology MSc student at Université Paris Cité, Universidade Lusófona, the University of Barcelona, and the Université du Québec en Outaouais (UQO). Her focus is on voice-based AI and digital mental health, with research conducted at long-term and transitional care facilities of Bruyère Health, Ottawa, in collaboration with UQO.

 Every morning, Margaret tells me about her friend Tommy. They were eight. The treehouse, the secret code, the promise to stay in touch.

Then she looks at me with sudden urgency and says, “I really need to call him today. I need to see what he’s up to.”

I have heard this roughly 30 times. I still listen. But I am a researcher, not a frontline worker, and I get to leave.

The personal support worker helping Margaret with breakfast has heard it too. She smiles, nods, moves on. Eleven other residents before lunch.

Here is what Margaret does not remember, and what the worker has no time to explain. Tommy has been gone for 20 years. Even so, every morning, Margaret wakes with the same animated need to reach the person who once made her feel most herself.

The math behind that moment is not an anecdote. Ontario will need more than 50,000 additional personal support workers by 2032, according to government figures obtained by The Canadian Press.

And in 2025, 90 percent of long-term care homes surveyed by the Ontario Long-Term Care Association reported difficulty filling registered practical nurse shifts.

Recruitment alone will not close that gap. As staffing thins, people living with dementia lose more than memory. They lose conversational partners, an audience, the sense that someone has time for their stories.

I study voice-based conversational artificial intelligence. In plain terms: software you talk to out loud, the way you would talk to a person. No typing, no menus, no computer skills, which suits people whose other abilities are fading.

I came to Canada, joining a dedicated team to compare two ways of meeting an AI companion: through a virtual reality headset that places you in a shared space, and through an ordinary desktop monitor.

Across Bruyere’s long-term care facilities, we measured which better creates a sense of presence, engagement and ease of use. Data collection is now complete, and the project is currently in the data analysis phase.

What stays with me is not only the technology. It is what people said to it.

Participants told the companion things they no longer felt comfortable telling busy staff. One afternoon, in a calm virtual park with birdsong in the background, a woman sang her wedding song to the AI-powered virtual companion. She had not sung it in years.

Afterward she told us she rarely opens up like that, because the people around her are too stretched to truly listen. One moment proves nothing on its own. But it shows what the absence of listening costs.

Now imagine that patience turned toward Margaret. A well-designed AI companion would not rush her on the 31^st telling. It could ask about the treehouse, remember the secret code, and let the story grow richer each time she returns to it.

Some worry about dehumanization, and they are right to. A companion must supplement human relationships, never substitute for them.

That means clear consent processes for people whose thinking has changed, honesty that the device is not a person, and privacy rules that treat voice recordings as sensitive health information. It also means openly debating a harder question: how a companion should respond when Margaret asks to call a man who died 20 years ago.

A speaker cannot hold a hand. It (still) cannot read a room. The risk of making a machine feel too human is real, and so is the risk of leaving someone alone.

Where Canada’s research and innovation system fails people with dementia

Here is where Canada’s research and innovation system fails people like Margaret: we fund promising work, then strand it in pilot mode.

Since 2019, more than $400 million has flowed into dementia initiatives through the Public Health Agency of Canada and the Canadian Institutes of Health Research.

Yet the fund closest to daily life, the $20-million Dementia Community Investment, directly reached just 8,820 individuals between 2019 and 2022, out of the roughly 770,000 Canadians living with dementia. That is about one percent.

Federal projects have reached Canadians hundreds of millions of times through campaigns and resources, but impressions are not relationships. Many pilots showed promise. Most never got the chance to scale.

The gap between ambition and allocation is just as old. In 2018, the Alzheimer Society of Canada recommended $150 million over five years to fully fund a national dementia strategy. Budget 2019 committed $50 million.

And this is not an abstract future. The personal support workers staffing these homes in 2035 are in high school now. The policymakers who will decide whether voice companions become standard care or remain a curiosity are interns in Ottawa this summer.

Three changes would alter the trajectory.

First, fund research built with people living with dementia, not just for them. Participatory research means they help shape the systems from the start, in the design room rather than only the testing room. Margaret’s Tommy story is not a glitch to be redirected; it is the specification.

Second, fix adoption, not just invention. Conversational AI carries real, recurring operating costs, and care homes face tight budgets, setup challenges and understandable skepticism. They need procurement frameworks and staff training that let them pilot low-risk assistive technologies without treating each one as a one-off experiment.

Third, demand accountability in deployment. In our study, staff called the companion “the future” and “a cool project,” but a pilot is not a policy. The real question is not whether a voice companion can free up 20 minutes of a worker’s morning; it is whether those minutes go to the resident or to the task list.

So here are the next steps. Ottawa should dedicate a defined share of the next phase of dementia funding to scaling the pilots that worked, with participatory design as a condition of the money. Provinces should build procurement pathways and training programs for low-risk assistive technologies in long-term care. And every funded deployment should report where the time it saved actually went.

Margaret will wake tomorrow needing to call Tommy. No machine can give him back. But an unconditionally patient virtual companion can ask about the treehouse on the 32^nd telling, exactly as if it were the first.

The rest of us face a simpler question than she does: whether we will build a care system, and an innovation system, that has time to listen.

Note: Names and identifying details of residents have been changed.