Practical guide · Responsible AI
Using AI to Build Technology for Good
The right use of AI begins with a useful outcome for people. In a care setting, that might mean helping a team find the right record or prepare a clearer update while leaving judgment, relationships, and accountability with people.
Start with the job, then choose the tool
A feature earns its place when it makes a specific task easier or more reliable for the people doing it. Before choosing a model, write down the user, the task, the intended benefit, and the cost of an error. Compare an AI approach with simpler options such as clearer forms, search, or a deterministic rule. NIST's AI Risk Management Framework calls for teams to map the context and risks of an AI system, measure them, and manage them throughout its life—not just at launch. Read the NIST framework.
Give AI a bounded role
Useful assistance
Summarise a long record for review, suggest a draft update, retrieve a relevant policy, or flag information that a person may need to check.
Human responsibility
Verify facts, decide what action is appropriate, speak with the affected person, and correct the record. The interface should make review and override possible.
This division reflects the WHO's guidance on human autonomy in health AI. It is a design principle for HoloCare-like products, not a claim that any particular HoloCare feature has passed a clinical evaluation.
Design around the people affected
In early learning, children and families may be affected even when the direct user is an educator. UNICEF's 2025 guidance calls for child-centred AI to protect children's data and privacy, support fairness and inclusion, and serve their best interests. For a product team, that means asking what information is necessary, explaining when AI is involved, and checking whether the feature works for people with different needs and levels of digital access. Read UNICEF's guidance.
Aged care and family tools also need clear boundaries around sensitive records and consent. The OECD AI Principles connect trustworthy AI with privacy, transparency, robustness, and accountability. These are useful questions for product design; they do not replace local legal or sector requirements.
Test the whole workflow
A good demo is a starting point, not evidence that a feature helps in practice. Test representative tasks with the people who will use the tool. Include incomplete records, ambiguous language, different accents or languages where relevant, and cases where the system should say it cannot answer. Check whether a person can notice an error, fix it quickly, and continue work when AI is unavailable.
After release, review reported errors and changes in use. Assign an owner who can pause or remove a feature if it causes harm. NIST describes risk management as a continuing activity across the system life cycle, with governance running through mapping, measurement, and management.
Five questions before release
- What problem does a person have, and how would we know it has improved?
- Would a simpler rule, search tool, or better workflow solve it just as well?
- What data is genuinely needed, and who can see or remove it?
- What happens when the model is wrong, uncertain, unavailable, or out of date?
- Who can review, correct, override, and stop the feature?
What this means for HoloCare
HoloCare develops software across care, learning, and service settings. A responsible direction is to use AI to help teams handle information and coordination while preserving a clear path for people to verify, decide, and respond. The examples in this guide illustrate design choices; they are not a feature list or a statement of certified safety, accuracy, or compliance.
This is general product-development guidance, not clinical, legal, or regulatory advice.
Sources
- AI Risk Management Framework CoreUS National Institute of Standards and Technology · Published 2023 · Accessed 2026-10-23
- Ethics and governance of artificial intelligence for healthWorld Health Organization · Published 28 June 2021 · Accessed 2026-10-23
- Guidance on AI and children, version 3.0UNICEF Innocenti · Published December 2025 · Accessed 2026-10-23
- OECD AI PrinciplesOrganisation for Economic Co-operation and Development · Published Updated May 2024 · Accessed 2026-10-23