MassAITC Gets $24.5 Million to Give Aging-at-Home AI Some Adult Supervision

Massachusetts has found a respectable use for its habit of turning every practical question into a research consortium. Give an algorithm a wristwatch, introduce it to a caregiver, and insist that it explain what, exactly, it has accomplished before anyone orders the conference tote bags.

This week, UMass Amherst and Mass General Brigham announced a $24,499,836 National Institute on Aging renewal for the Massachusetts AI and Technology Center for Connected Care in Aging & Alzheimer’s Disease, mercifully shortened to MassAITC. The award supports five more years, following approximately $20 million for 2021–2026.

The center, housed at UMass Amherst, has funded 53 projects, most industry-led or involving industry-academic partnerships. Its renewed scope includes supporting older adults in both homes and care communities, with a national pilot-partner network and mentoring support.

My verdict: a meaningful investment in the unglamorous middle of health technology. Sensors and language models can produce information. Finding out whether that information helps someone live better requires rather more work than giving the dashboard a calming shade of blue.

The Mass Pike is part of the architecture

This is a statewide story with a substantive Boston connection. The collaboration combines UMass computing and nursing expertise with Mass General Brigham. Co-directors include UMass professors Benjamin Marlin and Deepak Ganesan and Mass General Brigham and Harvard Medical School researcher Sudeshna Das.

That pairing matters because building a classifier and designing a usable care intervention are different jobs. One asks whether a computer can recognize a pattern. The other asks whether the pattern means anything, whether someone can respond, and whether the whole arrangement survives Tuesday.

The center’s facilities and implementation-site directory makes the practical appeal clearer: access to research cohorts, recruitment help, measurement facilities and places to test technologies. Its national partners include the University of Pittsburgh’s Healthy Home Lab, which uses an actual house more than a century old.

A splendidly appropriate opponent. Any technology aspiring to work in Massachusetts should eventually have to negotiate a house whose floor plan predates the concept of a reliable wireless signal.

As we observed after Boston Tech Week’s marathon of institutional enthusiasm, this region is most interesting when the work gets specific. Here, the useful connections are between engineering, clinical knowledge and the people who must live with the product.

The watch has data. The household has context.

Consider an earlier MassAITC pilot involving the University of Chicago and Adiona Health, described in the center’s Apple Watch agitation-monitoring project report. It examines whether wrist-motion measurements can help identify agitation in people with dementia. A custom app collects sensor readings alongside episodes reported by caregivers.

The reported recruitment total was 17 patient-caregiver pairs, with seven participants withdrawing. The team established a workable collection process, but the page says full analysis is pending. That supports feasibility; it does not establish a clinically validated warning system.

The distinction is the whole engineering problem. Movement data are measurements, not explanations. A useful system must connect a signal to the situation accurately enough that someone can act on it. Otherwise, the family receives another notification and the machine receives another flattering engagement metric.

For future evaluations, I would want to see missed episodes, false alarms, time spent responding and whether the device remains usable over months. Those are proposed tests of value, not results this pilot has already delivered. They also explain why research infrastructure deserves attention: it can help expose a product’s weak assumptions before those assumptions become somebody else’s daily chores.

The chatbot needed a calculator and a little humility

Another earlier project offers unusually useful candor. MassAITC’s Kinto caregiver-support pilot account says the team used a conventional Python tool to anonymize 162 caregiver datasets after finding large language models unreliable for that task.

It also reports deploying an AI-assisted caregiver-strain assessment in Kinto’s production system, with checks against historical coach notes. The account describes generally useful assessments but identifies arithmetic mistakes and insufficient sensitivity to recent changes in caregivers’ lives. A resource-recommendation system used a curated collection of dementia-care materials.

These are historical pilot findings, not products launched with this week’s grant. Nor does that project account establish that the software improved health outcomes in a controlled trial.

Still, there is something encouraging about an AI project admitting where ordinary software worked better. The industry occasionally behaves as though using a calculator would be a betrayal of the revolution. Here, the objective is helping a human coach, which gives the team permission to choose tools according to whether they work.

Our health AI guide explores the wider scramble to mediate medical information. This is a narrower and more testable proposition: can software help someone identify a need and find relevant support without losing the context that made the need urgent?

Please do not disrupt the living room

The center’s published stakeholder-engagement approach calls for involving older adults, caregivers and clinicians in needs assessments and design. That sounds bureaucratic until you imagine the alternative: a technically exquisite device nobody wants to wear, maintain or explain to visiting relatives.

Consent, access to intimate household data, charging, installation and ongoing support should all count in the product evaluation. So should the question of who pays once a pilot ends. A grant can fund experimentation; it cannot, by itself, make the resulting service affordable or establish a durable business.

We encountered a related challenge in brainjo’s attempt to turn a headset into therapy homework: the hardware becomes interesting when it has a defined job. The same standard should apply here. More data are worthwhile only when they support a better decision or reduce a burden.

Readers outside Massachusetts should care because the central challenge travels easily. A system that needs unusually patient researchers standing nearby is not ready for ordinary households. National testing partners could help reveal which ideas travel and which merely look excellent within commuting distance of their inventors.

MassAITC’s renewal is a serious bet on doing that homework. It has existing projects, documented limitations and a useful place to apply technical ambition. The next five years should be judged by what becomes dependable, usable and helpful—not by how many devices learn to describe themselves as intelligent.

Boston and Amherst have five more years to make the machines earn their place in the room. Preferably without giving the caregiver another password to reset.