A question at the heart of designing AI for social good is whether the system replaces human connection or makes human connection count more.
The answer matters because the people least well served by technology are often those who most need human guidance, and for whom it is scarcest. Refugees navigating unfamiliar bureaucracies. Newcomers learning to work in a language and culture they do not yet know. People building livelihoods in systems designed by and for those who already know the rules. By the end of 2024, UNHCR counted more than 120 million people forcibly displaced worldwide. Each country that receives them runs integration systems under heavy strain.
These people do not need chatbots that answer frequently asked questions. They need systems that meet them where they are, help them prepare for the interactions that matter, and stretch limited human expertise.
Why information alone fails
Most technology for newcomers delivers information. Websites list requirements. PDFs explain procedures. Portals provide forms. The assumption is that access to information is enough. It fails for four reasons.
Information is not understanding. A page explaining business registration in formal administrative language is accurate but useless to someone who reads that language at a basic level and cannot tell what is essential and what is optional.
Systems assume knowledge newcomers lack. A form asking about VAT registration assumes the person knows what VAT is and why it matters. A requirement for proof of address assumes they know which documents count.
Language barriers compound everything. Someone with intermediate English navigating systems in Dutch faces extra effort at every step. They may understand the words but miss the implications.
Human expertise is scarce. Caseworkers, advisers and volunteers serve large populations with limited time. An adviser who could spend an hour refining someone's business plan instead spends it explaining what a chamber of commerce is.
The result is a system that provides too much information and too little support.
Integration is increasingly local
In the Netherlands, the Civic Integration Act, which took effect in 2022, made municipalities responsible for guiding newcomers through integration, with personal plans tailored to each person. That design is more humane and more demanding. Each municipality and its partner organisations now need to understand individual situations, in several languages, at scale, with the same limited staff.
Preparation as the design principle
A better approach reverses the flow. Instead of providing information and hoping people can act on it, the system helps people prepare for the moments when human guidance is available, so that guidance can be used for what only people can do.
In practice, preparation means:
Before an appointment: helping the person understand what the meeting is for, what to bring and what questions they might ask, in their own language.
Understanding a procedure: explaining steps in plain language, checking understanding, and distinguishing what is required from what is optional.
Organising their situation: helping them gather and describe their documents, experience, and goals so the adviser starts with a clear picture.
After an appointment: recapping what was agreed and what happens next.
The adviser then spends the appointment on judgement, encouragement and the decisions that a person needs.
Why preparation works
Albert Bandura's research on self-efficacy shows that people act more effectively when they believe they can succeed, and that belief grows through small, successful steps. Arriving at an appointment prepared, understanding what is happening and knowing what to ask, is exactly such a step. It changes the person's experience from being processed to taking part.
Amartya Sen's capability approach offers the deeper frame. Development, he argued, is about expanding what people can do and be. A system that helps someone register a business, understand a contract or prepare for a job interview expands their capabilities. A system that only tells them where the forms are does not.
Designing for dignity
Designing for vulnerable people raises the bar on every decision:
Plain language, in the person's own language, checked for understanding rather than assumed.
No pretence: the system says clearly that it is an AI and what it can and cannot do.
Clear limits: legal status, asylum decisions and anything with serious consequences stay with qualified people.
Privacy by design: minimal data, held securely, with the person in control of what is shared.
A person always reachable, especially when someone is distressed.
The EU AI Act treats AI used by public authorities in migration, asylum and border control as high-risk, with strict duties. A preparation tool used by support organisations sits outside those decisions, but should be designed to the same standard of care.
What success looks like
The measure is not efficiency alone. It is whether people arrive prepared, understand what is happening to them, and achieve what they came for, whether a registered business, a job or a stable home. And whether advisers spend more of their time on the conversations where they make a difference.
Questions for organisations supporting newcomers
Where does adviser time go on explanations that preparation could cover?
In which languages do the people you serve need support?
What must always stay with a qualified person?
How will you know whether people understood, not only whether they were told?
How is personal data minimised and protected?
How Mitochondria's agentic AI systems support integration work
The approach. Mitochondria's research with organisations supporting newcomers in Europe starts by mapping how support happens: where time goes, where preparation gaps cause friction, and which conversations people need. Systems are designed for plain language, in the person's own language, with clear limits and a person always reachable.
What it enables. Newcomers prepare for appointments, understand procedures and organise their situation before they meet an adviser. Advisers start from a clear picture and spend their time on judgement and encouragement.
The impact. Scarce human expertise reaches more people and counts for more, and newcomers take part in their own integration with confidence and dignity.
Read more about responsible AI at Mitochondria.
Sources
UNHCR (2025). Global Trends: Forced Displacement in 2024.
Wet inburgering 2021 (Netherlands Civic Integration Act, in force from 1 January 2022).
Bandura, A. (1977). Self-efficacy: toward a unifying theory of behavioural change. Psychological Review, 84(2).
Sen, A. (1999). Development as Freedom.
Regulation (EU) 2024/1689 (AI Act), Annex III on migration, asylum and border control.
—
Mitochondria builds agentic AI for evidence-led core operations. Research-led and model-agnostic. ISO/IEC 27001:2022 certified, with controls aligned to the GDPR, UK GDPR, India's DPDP Act and Australia's Privacy Act. Amsterdam and Pune. Request a working session →