What did 100 US-funded AI projects for older adults actually prioritise?
A peer-reviewed review of 100 NIA-funded pilots found a portfolio concentrated on dementia, home settings and classification. Explicit focus on underserved groups and user involvement was uncommon—but the study examined proposals, not completed clinical benefits.
The short version
- What happened
- Researchers coded 100 pilot applications funded from 2021 to 2024, examining intended users, care settings, AI tasks, technology maturity and whether users were explicitly involved in development.
- Who it affects
- Older adults and caregivers should not read a funded-project listing as proof that a product works. Ask what was tested, with whom, for how long, and what happens after an alert or error.
- What remains unproven
- The study analysed 100 funded applications, so it describes proposed work rather than completed clinical performance or patient benefit.
Keep in mind
The study analysed 100 funded applications, so it describes proposed work rather than completed clinical performance or patient benefit.
The accessible early-version article reports several results qualitatively as 'most' or 'few' without publishing numerical category totals; this report does not invent percentages.
Application wording may understate work that later occurred, while funding success and carefully written aims do not guarantee delivery.
The 2021–2024 sample is a subset of a programme that now lists 166 projects across five cohorts, so later awards may differ.
The analysis covers a US federal programme and should not be treated as a map of all AI-for-aging research worldwide.
No clinical comparator, pooled participant denominator, product accuracy measure, adverse-event rate or implementation outcome was reported.
Source analysis · 2 linked sources. Primary analysed source date: 2 November 2026.
Keep following the evidence
Join The Impact Brief founding list for selected AI reports, their effect on people and the evidence limits. Read a sample first.
Read the full report · 9 min The evidence, methods, implications and limitations
The direct finding: the portfolio was broad in technology but narrow in emphasis
A peer-reviewed analysis of 100 pilot projects funded through the US National Institute on Aging's Artificial Intelligence and Technology Collaboratories programme found a clear pattern in where early investment went. Most applications targeted dementia risk or dementia management, most were intended for home-based settings, and machine-learning classification was the dominant computational task. Early detection and remote monitoring appeared frequently. By contrast, the authors say few projects explicitly targeted underserved populations or described intended users taking part in design and development.
That is a useful map of research priorities, not a verdict on whether the technologies worked. The paper reviewed project applications funded between 2021 and 2024. Applications describe a proposed problem, users, method and pathway; they do not necessarily contain a completed trial, a deployed product or measured patient benefit. The review therefore identifies what a prominent public programme chose to explore and what applicants documented. It cannot tell an older adult that a dementia classifier is accurate, that home monitoring prevents harm or that a robot reduces caregiver burden.[1]
How the researchers turned project documents into a portfolio analysis
The team used mixed-methods document analysis. Its denominator was 100 funded pilot applications from three original AITC centres during 2021–2024. The coding form asked for a one- or two-sentence account of each technology, its main clinical or functional objective and its intended user. Coders could mark family and paid caregivers, older adults, prescribers, nurses, administrators, health-IT staff, public-health staff and insurers. Care settings included living alone or with family at home, outpatient and inpatient services, post-acute care, long-term care, assisted living and public-health departments.
For the technology itself, the form separated AI tasks such as text or image generation, classification, segmentation, object detection, regression, unsupervised learning and reinforcement learning. It also recorded named methods—including machine learning, neural networks, deep learning, natural-language processing, human–AI interaction and speech processing—without instructing coders to infer an approach that the application did not state. Technology maturity was classified at application time as academic research, a pre-market product or a developed product. This structure makes the analysis more inspectable, while still depending on what applicants wrote and how consistently coders interpreted it.[1]
The distinction that matters
What the 100-project review can establish
The analysis describes funded intentions. Each step below marks a boundary between a proposal and evidence of benefit.
Documents
Code 100 funded applications from 2021–2024
Map stated users, settings, methods and maturity
Portfolio
Compare recurring priorities and omissions
Dementia, home use and classification dominate the authors' summary
Not tested
No pooled clinical comparison or outcome evaluation
Cannot establish accuracy, safety, access or health benefit
Next evidence
Link proposals to completed studies and real-world use
Test inclusion, usability, harms and patient outcomes
The inclusion finding matters because older adults are not one population
The study's concern about underserved populations comes from an explicit field in the abstraction form. Coders were to record a sociodemographic focus only when an application named a group defined by race or ethnicity, income, geography, disability, language or another characteristic. Silence was not treated as evidence of inclusion. That is a demanding but appropriate distinction: a project can say its tool is for older adults while overlooking differences in internet access, language, housing, hearing, vision, mobility, cognitive impairment, trust and the availability of family or paid care.
The authors describe explicit targeting of underserved groups as uncommon, but the accessible early-version article does not publish a numerical breakdown for that category. The responsible interpretation is therefore qualitative. It would be wrong to convert 'few' into a made-up percentage or to assume the remaining projects excluded people. Some applications may have involved diverse participants without making that focus prominent in the material reviewed. The result still identifies a documentation and design problem: public funders cannot assess whether inequity is being addressed when target populations and participation are left implicit.[1]
User involvement was coded only when the proposal said what people would do
The review treated user involvement as more than naming beneficiaries. A project was coded positively only if it explicitly described older adults, caregivers, clinicians or other intended users taking part in design, development or testing—for example through co-design sessions or usability work. The authors found that few projects documented this kind of involvement. That gap is particularly consequential for tools used in homes, where a design must fit routines, fluctuating health, shared devices, informal care relationships and the practical limits of broadband, charging, maintenance and technical support.
Co-design does not prove that a product is safe or effective, and a proposal can involve users poorly. But absence of a stated route for user feedback raises the risk that technical success is defined too narrowly. A classifier may distinguish records in a dataset while producing advice that people cannot understand. A monitoring system may detect a signal while creating false alarms that exhaust a caregiver. A conversational interface may sound reassuring while failing someone with hearing loss or limited English. Those are testable design and service questions, not cosmetic preferences.[1]
What the study did—and did not—measure
There was no clinical comparator in this portfolio review because it was not an intervention study. The measured unit was the application, and the outputs were coded characteristics: intended users and settings, AI type, named technology, maturity, involvement, problem and delivery form. The paper does not aggregate patient numbers across pilots, compare an AI service with usual care, or report sensitivity, false-alarm rates, hospital use, quality of life, independence or caregiver outcomes. The 100-project denominator describes the analysed portfolio—not 100 completed products and not 100 trials.
The official AITC directory currently lists 166 projects across five cohorts, so the paper's sample is a historical subset rather than the entire programme as it exists in November 2026. That difference is not an error: the researchers set a 2021–2024 window and analysed the material available for those projects. It does mean the findings should not automatically be applied to later awards, new centres or changes made after an application. A follow-up could test whether newer cohorts show more inclusive recruitment, stronger co-design and more attention to real-world implementation.[1][2]
What this means for older adults, carers, clinicians and funders
For older adults and families, the immediate lesson is to ask what stage a technology has reached. A funded pilot may be a concept, prototype or early validation exercise. It should not be presented as established care merely because it uses AI or has federal support. People considering a tool need evidence about the specific population, the consequences of errors, who receives alerts, what data leave the home, how consent changes with cognitive impairment, and whether a human service remains available when the technology is unsuitable.
Clinicians and care organisations should also distinguish detection from benefit. Finding a possible change earlier can help only if the signal is reliable, the next action is appropriate and services have capacity to respond. False positives can create anxiety and extra appointments; false negatives can offer reassurance when care is needed. Funders can use the review prospectively by requiring applicants to name intended users and underserved groups, set out a participation plan, define measurable human outcomes and explain how a pilot could be evaluated in the conditions where older adults actually live.[1][2]
What would change the assessment
A stronger portfolio assessment would report category counts with uncertainty about ambiguous documents, describe how many coders reviewed each application and quantify agreement between them. It would connect proposed aims to later milestones: completed prototypes, peer-reviewed results, recruitment, withdrawals, adverse events, usability, regulatory status and continued use after funding. Linking awards to spending and follow-on investment could show whether certain needs receive only small exploratory grants while others progress toward implementation. Public reporting should also distinguish academic, nonprofit and commercial teams and disclose relevant interests.
For any individual technology, the evidence threshold depends on its claim. A low-risk reminder still needs accessibility and privacy testing. A system that influences diagnosis, medication or urgent care needs external validation, appropriate comparators, prespecified endpoints, subgroup performance, calibration and prospective evidence about patient consequences. The portfolio review's most defensible conclusion is therefore modest but important: early US public investment created many AI-for-aging experiments, yet the documented portfolio leaned toward dementia, home monitoring and classification while leaving visible room for broader needs, explicit inclusion and genuine participation by the people expected to live with the tools.[1][2]
Watch the background, then check the evidence
Illustrative photo · not a video frame
The Royal Society · 44 min 28 sec · Background explainer
Is this the AI future we want? | The Royal Society
YouTube loads only when you choose; playback starts when you press play.
A Royal Society discussion with Madhumita Murgia and biomedical engineer Alison Noble on AI’s potential in healthcare, science and culture, recorded at the 30 June 2026 event.
Evidence check: A discussion of opportunities and risks, including future possibilities. It is not evidence that every application described is available or clinically proven.
Published 2026-07-01. Source checked 14 October 2026. Check captions on the publisher’s video page.
Check the healthcare evidencePhoto: Daniel Sone / National Cancer Institute / Unsplash · Licence
Continue the story
Read next
Continue exploring research funding: compare the finding and its evidence limits.
$20m renewed for AI and healthy ageing—what will patients actually see?
Johns Hopkins says a five-year US award will fund AI, sensors and digital-health research for older adults. The decision expands a research pipeline; it is not evidence that any tool improves diagnosis, independence or care.
Read the next reportExplore the full source and evidence record
What this means for people
- Older adults and caregivers should not read a funded-project listing as proof that a product works. Ask what was tested, with whom, for how long, and what happens after an alert or error.
- Funders and care services can make inclusion assessable by requiring named populations, accessible participation, practical home-use testing and outcomes that matter to older adults—not only model performance.
Global context
Population ageing, dementia care and shortages in formal and informal support affect many countries, but care systems, housing, language, connectivity, privacy law and public funding differ sharply. The US portfolio can reveal one influential funding pattern; it cannot establish that the same priorities or gaps apply in Asia, Africa, Europe, Latin America or the Middle East without comparable local analyses.
What the evidence does not yet show
- The study analysed 100 funded applications, so it describes proposed work rather than completed clinical performance or patient benefit.
- The accessible early-version article reports several results qualitatively as 'most' or 'few' without publishing numerical category totals; this report does not invent percentages.
- Application wording may understate work that later occurred, while funding success and carefully written aims do not guarantee delivery.
- The 2021–2024 sample is a subset of a programme that now lists 166 projects across five cohorts, so later awards may differ.
- The analysis covers a US federal programme and should not be treated as a map of all AI-for-aging research worldwide.
- No clinical comparator, pooled participant denominator, product accuracy measure, adverse-event rate or implementation outcome was reported.
What to watch next
- Numerical category tables and coder-agreement details in the final Version of Record
- Outcome reporting that links each funded pilot to completed studies, usability and adverse events
- Recruitment and performance for underserved racial, linguistic, rural, disabled and low-income groups
- Evidence that older adults and caregivers influence design choices rather than only test finished prototypes
- Prospective comparisons showing whether a specific tool improves outcomes over existing care
Living evidence record
Impact record IAI-0U1X977
Evidence stage
Studied
Confidence
Supported
Reporting basis
Source analysis
Independent or research support
Present
Record status
Monitoring
Last checked
3 November 2026
Source trail
2 direct sources across 2 source types.
People impact
Documented in this record.
Uncertainty
Limits and next checks are explicit.
Stages describe the evidence available—not whether a technology is good or bad. See the public method.
Related-source reporting disclosure
This record analyses 2 linked source records around the same underlying development. The extra records add method, date or context, but they do not by themselves constitute independent replication of every performance claim or predicted outcome.
Evidence trail
Sources used for this report
Links checked 3 November 2026
Research question, topics and regions
What the first four years of the US NIA Artificial Intelligence and Technology Collaboratories pilot portfolio prioritised and left underrepresented
Topics: Older adults · Dementia · Caregiving · Inclusive design · Health research · Research funding
Regions: United States · North America
This report is labelled source analysis. We summarise and analyse source material in our own words; company statements remain attributed claims until independently supported. Translated summaries preserve the meaning of the original source and link back to it. Read our editorial standards.
Reader commentary
What does this mean where you are?
Tell us what you have seen at work, in your community or in everyday life. A question is welcome too. No account is needed; we review notes before publishing them.
Explore commentary across the portal →Published reader notes
0No published reader notes yet. You can start the evidence-led discussion above.
Prefer a private correction or response? Contact the newsroom.