A research vision · Prof. Runhe Huang · Hosei University
WARM AI
Rooted in Humanity, Advancing Human Well‑Being
Four human moments
A child left alone in a parked car. An older adult falls in the bathroom. A patient urgently needs medicine. A student suffers bullying in silence.
These are more than problems to solve. They are human moments calling for understanding, compassion, and care.
Moment 01 / 04 · Child & student safety
A child left alone in a parked car.
A 60 GHz radar can sense a child left in the cabin, in a moment where every minute counts.
Sensing
Moment 02 / 04 · Elderly support
An older adult falls in the bathroom.
Millimetre-wave radar recognises falls and unusual activity without a camera in the room. In one study it told six bathroom actions apart with 96.1% accuracy.
Sensing
Moment 03 / 04 · Spatial intelligence & robotics
A patient urgently needs medicine.
A home robot that understands objects and spaces can find the right medicine and bring it safely.
Sensing
Moment 04 / 04 · Child & student safety
A student suffers bullying in silence.
Speech, activity and emotion signals can support early intervention, with privacy and human oversight at the centre.
Sensing
The vision
WARM AI is a research vision that looks beyond what AI can accomplish to how its assistance is experienced by the people it serves.
Its ultimate aspiration is not to make machines more human, but to make their intelligence more humane.
- What can AI do?Capability
- Who is still unsafe?Vulnerability
- What support restores dignity?Human need
- How do we deploy responsibly?Trust
Vulnerable-first AI is not a narrow application area. It is a design philosophy: choose problems by human consequence, not by technical convenience.
The WARM AI framework
Five interconnected dimensions bring human values into intelligent systems.
- Care and ComfortHelp that arrives in time and eases the moment.
- Empathy and ConnectionUnderstanding how people feel, and keeping them close to those who care.
- Protection and FairnessSafeguards for those most at risk, without surveillance or bias.
- Dignity and EmpowermentPrivacy and independence by design, so people stay in charge of their lives.
- Safety and ResilienceDependable sensing and response when every second counts.
Why vulnerable-first, and why now
- 1.3Bpeople experience significant disability, about one in six of us.
- 2.1Bpeople will be aged 60 or older by 2050.
- 57Mpeople were living with dementia in 2021, with nearly 10 million new cases every year.
AI progress should be measured not only by accuracy or profit, but by whether it reduces avoidable suffering and expands human autonomy.
Sources: World Health Organization fact sheetsDisability and healthAgeing and healthDementia
Seven principles for vulnerable-first AI
These principles convert good intentions into deployable research practice.
- Dignity firstAvoid deficit-based or stigmatising design.
- Privacy by architectureMinimum sensing, local processing, privacy-preserving data.
- Reliability over noveltyField robustness, uncertainty, fallback modes.
- Human-in-the-loopCaregivers, clinicians, teachers, families, users.
- Equity & accessLow-cost devices, multilingual interfaces, offline modes.
- Explainability for actionClear reasons, not just model confidence.
- Evidence & accountabilityBenchmarks, ethics review, incident reporting.
From the lab
Bathroom safety without cameras
A practical activity-recognition pipeline for bathroom-like rooms: radar point clouds, DBSCAN and KDE denoising, temporal modelling, and an alert for review.
A child left in a car
60 GHz FMCW radar for in-cabin sensing: detecting a child left in a vehicle, an urgent, time-critical scenario.
Find my medicine and bring it safely
An augmented scene graph of the home, LLM planning grounded in that model, and a robot that asks for help when it is uncertain.
Who becomes safer, more independent, more included, or more understood because this AI exists?
Let us build a global discussion and development community.
contact@warm-ai.org
Send email