The AI That Costs Too Much

Marvella Johnson is 72 years old, a retired home health aide who lives in Memphis and receives $1,140 a month in Social Security. Her rent is $550 for a room in a shared house on the south side. After rent, she has $590. Her medications cost $85 a month after her Part D plan. Her food costs roughly $250. Transportation to her doctor, her pharmacy, and her church costs $40 to $60 a month depending on the price of gas and whether her neighbor Robert can drive her. What remains is between $195 and $215, depending on the month. That is not discretionary income. That is the margin between Marvella and an emergency she cannot absorb.

The personal AI ecosystem this publication has spent twelve series describing requires, at its minimum viable configuration, more money than Marvella has.

The Minimum Viable Ecosystem

#

The health monitoring AI from Series 1 requires a smartphone and a data plan. A basic smartphone costs $200 to $500. Monthly service with adequate data for health monitoring and daily check-ins costs $30 to $60. A health-monitoring wearable, the device that tracks heart rate, sleep, activity, and fall risk, costs $50 to $300 for the device plus $5 to $15 a month for the platform subscription. The ambient home monitoring from Series 3, which detects movement changes, sleep disruption, and environmental hazards, requires broadband internet at $50 to $80 a month.

The monthly carrying cost of the minimum viable personal AI ecosystem is $85 to $155. That is roughly half of Marvella’s margin. The device costs, amortized, add another $15 to $40 a month. The total: $100 to $195 a month for a system designed to protect the health and safety of a 72-year-old woman living alone.

Marvella does not own a smartphone. She does not have broadband. Her nearest library with public computer access is 1.4 miles from her house. It closes at 6 PM. She has never used a wearable device. She is not resistant to technology. She is outside its economic reach.

The ecosystem BML has described across twelve series is not a luxury product. It solves real problems for real people. Medication tracking that prevents the interactions nobody caught. Baseline monitoring that detects the change nobody noticed. Cognitive screening that catches the decline two years before the crisis. Fall prevention that keeps a woman in her home instead of a nursing facility. Every function matters. Every function costs money Marvella does not have.

What the Free Pathways Provide

#

Free and subsidized pathways to some of these functions exist. They are real. They are also insufficient, and honesty requires naming both.

The Lifeline program, a federal subsidy for phone and internet service for low-income households, covers a basic smartphone plan in most states. The subsidy is typically $9.25 per month, which can reduce a basic wireless plan to free or near-free. The program serves roughly 7 million households. It provides a phone with limited data. It does not provide the broadband that ambient home monitoring requires. It does not provide a wearable. It does not provide the platform subscriptions that connect the devices to the monitoring intelligence.

Library-based public computer access provides intermittent internet connectivity. Marvella can walk or ride to the library, use a computer for her allotted time, and access health information. She cannot run continuous health monitoring from a library computer that closes at 6 PM. The gap between intermittent access and continuous monitoring is the gap between looking something up and having someone watch over you.

Community health centers, funded through the Health Resources and Services Administration, provide primary care on a sliding fee scale regardless of ability to pay. They are Marvella’s primary healthcare access point. They do not provide personal health AI.

PACE programs, the Program of All-inclusive Care for the Elderly, offer comprehensive care coordination for qualifying low-income elders who meet nursing home eligibility criteria. PACE is available in fewer than 200 locations nationally. Memphis has one PACE provider. Marvella does not currently meet the functional eligibility criteria. PACE is real, important, and not available to most of the people who need it.

Each program addresses a piece of the access problem. None of them, alone or together, close the gap between what the AI ecosystem provides and what Marvella can access.

The Community Health Worker as Human AI

#

Marvella’s health AI is a man named Raymond. He is a community health worker employed by a federally qualified health center. He visits Marvella twice a month. He asks the questions the health monitoring AI would ask: How are you sleeping? Have you fallen? Are you taking your medications? Have you noticed any changes in how you feel? He checks her blood pressure with a cuff he carries in his bag. He reviews her medications. He coordinates with her primary care provider. He helps her navigate the benefits she qualifies for.

Raymond does what the AI does. He does it with a clipboard and a phone, on twice-monthly visits, for a caseload of fifty-four people.

The arithmetic of Raymond’s job is the arithmetic of the access gap. The AI checks in daily. Raymond checks in twice a month. The AI monitors continuously. Raymond monitors in the twenty minutes he can spend at each visit. The AI can process changes in real time and flag them to a clinician within hours. Raymond writes a note and calls the clinic when he gets back to his car.

Raymond is good at his job. He catches things. He has sent Marvella to the emergency department twice in the past year, both times for medication issues that a daily AI check-in might have caught earlier. The system he works in is not designed to let him do what the technology would let him do. His caseload is too large, his visits too infrequent, and his tools too limited for the work that needs doing.

The most realistic near-term path to serving Marvella is not giving her a smartphone and a wearable. It is giving Raymond an AI backend. A system that helps him prepare for visits based on the patient’s recent pharmacy data and clinical records. A system that generates the questions he should ask based on changes since his last visit. A system that lets him extend his capacity between visits through automated check-in calls that Marvella can answer on the landline she already has. The AI does not replace Raymond. It makes Raymond’s fifty-four-person caseload manageable in ways it currently is not.

The Editorial Standard This Piece Establishes

#

This publication has reviewed paid solutions across twelve series. Medication tracking platforms with subscription fees. Wearable devices with monthly costs. Home monitoring systems with broadband requirements. Each review described what the technology does, what it costs, and how well it works. The reviews were honest. They were also written from the assumption that the reader could pay.

That assumption ends here. Every paid solution BML reviews must include a free or low-cost pathway presented alongside it, or the publication has written for the comfortable and called it universal. This standard applies retroactively across every series already published and forward to every future piece. Where a free pathway does not exist, the publication says so directly. Where a subsidized option exists but is insufficient, the publication names the gap.

A publication about AI for aging adults that does not account for the 23 percent of Americans over 65 who live on less than $1,500 a month has described a product category, not a solution.

What Offline-Capable AI Changes

#

The near-term technology path that most directly addresses Marvella’s situation is AI that works without continuous connectivity. Health monitoring tools designed to function on basic smartphones with minimal data plans, syncing when connectivity is available and running degraded-but-functional monitoring in between. Medication reminders that work offline. Health check-in protocols that operate over voice calls rather than apps. Fall detection through a basic wearable that alerts a preset number without requiring broadband.

These tools are technically achievable. Some are in development. The barrier is not engineering. The barrier is funding. Building an AI health monitoring system for a population that cannot pay $100 a month requires someone other than the end user to fund it. That someone is either the government, through Medicaid managed care or public health funding, or the healthcare system, through value-based contracts that recognize that keeping Marvella out of the emergency department saves money that more than covers the cost of the monitoring.

The economic argument is straightforward. One emergency department visit costs $2,000 to $5,000. One hospital admission costs $10,000 to $30,000. A year of AI health monitoring costs $1,200 to $2,400. The math works for everyone except the person who has to pay the $1,200 out of a $590-per-month margin. The system that would save money in the aggregate is unaffordable at the individual level for the people who would benefit from it most.

The Policy Response

#

Universal broadband as a public utility would change the access picture for Marvella more than any product design decision. Broadband infrastructure policy has been debated at the federal level for over a decade. Rural and low-income communities remain underserved. The digital divide for Americans over 65 is wider than for any other age group, with roughly 25 percent of adults over 65 lacking home broadband access.

Medicaid managed care plans that cover AI health companions as a care coordination benefit would change the picture further. Some states are beginning to include remote patient monitoring in their Medicaid benefit structures. The coverage is narrow, the qualifying criteria are restrictive, and the benefit typically covers the monitoring device but not the smartphone, the data plan, or the broadband that the device requires.

These are policy choices. They are not engineering problems. The technology to serve Marvella exists or is close to existing. The decision to fund it for her has not been made.

Raymond, Next Week

#

Raymond visits Marvella on the fifteenth. He will park on her street, walk to the side door of the shared house, and sit at the kitchen table where she has set out two glasses of water because she knows he is always thirsty after his morning rounds. He will ask the questions. He will take her blood pressure. He will review the medications she keeps in a plastic organizer on the counter. He will write his notes. He will drive to the next visit. He has fifty-three more people to see before the month ends.

The AI that could support him is in development. The backend that could prepare him for each visit based on pharmacy data and clinical records is technically achievable. The system that could call Marvella on her landline between visits and ask the questions Raymond asks in person is not a research problem. It is a funding problem.

Marvella is waiting for something that is being built. That is not the same as waiting for something that is coming. Built means someone decided to build it. Coming means someone decided to bring it to her. The first decision has been made. The second has not.