Anbu Dangbeon
ANBU DANGBEON · PREVENTING LONELY DEATHS

Middle-aged people living alone, taking turns to check on each other

Anbu means "how are you"; dangbeon means "the person on duty." We make at-risk people the ones on duty rather than the ones being looked after. A 55-year-old man living alone checks on a 62-year-old in the same neighborhood once a week. He becomes someone who looks after another; the other opens the door because it is someone like him, not a social worker. When a pair goes quiet, the community center caseworker is alerted. People are connected to people, but the final safety net stays with a public caseworker.

The landlord finds out before the family does.

Why these people

Korea Ministry of Health and Welfare, 2024 Survey on Lonely Deaths (published November 2025)

3,924
Lonely deaths in 2024
up 7.2% on the year
54%
were men in their 50s and 60s, 2,117 people
66% if counting male deaths only
39%
had a basic welfare record in their final year, 1,462 people
The other 60.9% received none
43%
were first found by a landlord or building guard
Family: 26.6%

A policy blind spot

Korea's emergency safety service covers people aged 65 and over who live alone, plus people with disabilities at any age. A person in their 50s without a disability is not eligible. Statistics Korea counts about 2.24 million single-person households aged 40 to 59, sitting between youth and elderly programs.

Pride

This group refuses help. The moment they become "a person being cared for," they feel they have failed. A sensor, a camera, a robot in the home all fail for the same reason.

Suicide, or not?

Korean law defines a lonely death as dying alone "by suicide, illness or other causes" while socially isolated. Of the 2024 deaths, 13.4% (526) were classified as suicide. The survey does not break down the causes of the rest. What stands out is the age gradient: suicide accounts for 57.4% of deaths among people in their 20s and younger, 43.3% in their 30s, 25.7% in their 40s, and 13.5% in their 50s. Most middle-aged lonely deaths are not deaths by suicide.

These figures come from an analysis of 57,145 police criminal-justice records from 2024.

Where current approaches stop

A path reported again and again in journalism and research. Today's technology is crowded at the far end

Job loss · divorce · dischargeToday: nothing
Us: detect entry from crisis data
Severed ties · falling incomeToday: nothing
Us: pair up, give a role
Self-neglect · declining healthToday: welfare calls (only if answered)
Us: missed check-ins and conversation signals go to the caseworker
DeathToday: sensors, cameras
Never get this far
DiscoveryToday: screening models
39% had touched the system

Sensors and cameras find people quickly after death. Welfare calls reach only those willing to answer. Among those who died, 39.1% had a basic welfare record in their final year. Being in touch with the system was not enough. What is missing is the last step: someone knocking, and that person opening the door.

The community center pairs two men in their fifties who live alone. Check on each other once a week, they are told.

Both are offended. "Why me." Phone in hand, nothing to say. "Eaten?" "Yeah." Hang up. That is all.

Weeks pass. One day one of them does not pick up. The other stands in front of that door for the first time. Knock, knock.

Does pairing two at-risk people double the risk?

This is the first question anyone asks. Anbu Dangbeon is not a structure of two people

Layer 1 · The pair

A and B check on each other once a week. No housework, no nursing. They only check. They are not responsible for each other; they simply remember each other.

Layer 2 · AI

When check-ins stop, AI catches it. If one side goes silent, the other is asked to visit. If both go silent at once, that pair goes to the top of the queue. Two people falling quiet together is exactly the case AI reacts to most strongly.

Layer 3 · Public caseworker

Final responsibility for checking on a person stays with the community center caseworker. The duty pair does not replace the caseworker. It increases how often one caseworker can reach people.

In short: AI does not care for people. AI decides who should be connected to whom, and in what order. If a pair breaks, the link to the caseworker remains. Today a caseworker searches for at-risk people one by one, covering only as many as that allows. Adding duty pairs means the same caseworker receives signals more often and sooner.

What AI does, and what it cannot

It cannot make a relationship, give someone a job, or open a door. People do that

Four jobs for AI

  • Priority. Combine crisis signals from the public-private data platform, such as job loss, divorce and unpaid utility bills, to estimate where a person is on the path.
  • Pairing. Match by distance, daily rhythm and former occupation.
  • Alerts. Raise pairs whose check-ins stopped, and risk signals found in conversation logs, to the caseworker.
  • Summaries. Let one caseworker oversee hundreds of pairs.
Caseworker view (example)District A · 24 pairs
PairLast check-inStatusNext
A(55) ↔ B(62)2 days ago, 1-min callOK—
C(58) ↔ D(61)9 days agoMissedAsk C to visit
E(53) ↔ F(66)1 day ago, "went to hospital"Health signalCaseworker calls
G(60) ↔ H(57)14 days, both silentNowCaseworker visits

All names and figures are fictional. AI decides which door to knock on first, by whom, and with what words.

Will isolated men step up to look after someone?

It is a hypothesis. Three lines of evidence support it

Peer support research

In mental health, Korean studies repeatedly find that the person giving help gains self-confidence, self-esteem and social ties, not only the person receiving it.

Men's Sheds

Born in Australia, now across the UK and Ireland: a model for isolated men over 50. "Men don't talk face to face, they talk shoulder to shoulder." Studies show lower depression risk.

National plan

Korea's first Lonely Death Prevention Plan states that socially isolated households should be active agents of community connection, not just recipients. The direction exists, but across the government and local programmes and the major private services I surveyed, I found no structure that has at-risk people check on each other.

Plan and limits

Entry to the 2026 AI & Digital Social Problem Solving Challenge (Korea)

Plan

  • October: partner with one community center or welfare agency, fix data fields, test the hypothesis with field caseworkers
  • November: design and prototype three screens, pairing, check-in, caseworker, on the public-private AI and cloud platform
  • Metrics: check-in continuation rate, caseworker interventions, change in participants' weekly social contact

Limits and principles

  • We do not start with people outside the system. We start where a caseworker offers the role to someone they already know.
  • We never leave a pair on its own. When check-ins stop it goes to the caseworker, and a person makes the final check.
  • No cameras, no sensors. Personal data stays under consent and caseworker control.
  • Lonely death is not one country's problem. Japan appointed a minister for loneliness in 2021, the UK in 2018.