Human-Centered AI: How to Evaluate AI for Human Flourishing

The moment we ask, “What should this AI system do?”, we step into philosophy. AI is not only about speed or accuracy. It is about impact, human flourishing, and meaningful outcomes. So the real question is not simply whether AI works, but whether AI helps humans flourish. This is where human-centered evaluation becomes necessary.

The capability perspective

Philosopher Martha Nussbaum, alongside a growing number of scholars across the humanities and social sciences, has significantly developed the capability approach. Despite philosophical disagreements about how the capability approach should best be understood, it is broadly used as a conceptual framework for normative questions concerning individual well-being, social arrangements, and proposals for social change. At its core, the capability approach directs attention toward what people are genuinely able to do and be—including opportunities to be educated, move freely, participate in society, maintain supportive relationships, and pursue meaningful forms of life.

Martha Nussbaum’s capability approach provides a useful lens for thinking about technology. Technology should expand what people are able to do and be rather than merely optimizing efficiency. In this sense, AI should not only automate decisions. It should support human judgment, dignity, agency, and meaningful participation. When we evaluate AI through this lens, we move beyond performance metrics and begin asking deeper questions:

Does this AI expand human capability or replace it?
Does it improve judgment or weaken it?
Does it support meaningful decision-making?
Does it create dependency or empowerment?

Beyond performance metrics

AI systems increasingly shape education, healthcare, hiring, finance, and everyday decision-making. If these systems are trained on biased data, optimized only for efficiency, or deployed without sufficient transparency, they can reduce human agency rather than enhance it. Evaluating AI for human flourishing therefore requires a broader framework.

1. Alignment with human goals

Does the AI actually solve the user’s problem, or does it merely produce plausible-sounding outputs? Is the system useful within real decision-making contexts?

2. Bias and fairness

Does the model reproduce or introduce structural bias? Which individuals or groups may be advantaged or disadvantaged by its decisions?

3. Transparency and explainability

Can users understand how the AI reaches its conclusions? Can important outputs be explained, questioned, and evaluated?

4. Preservation of human judgment

Does the AI help users become better practitioners, or does it encourage them to outsource judgment? Does it strengthen or weaken decision autonomy?

5. Reliability and trust

Does the system hallucinate or generate unsupported claims? Can its outputs be verified against reliable sources?

Evaluating human-centered AI

Existing methods for evaluating AI already move in this direction. These include:

Human-in-the-loop review
Red teaming and stress-testing for failure and harm
Comparative evaluation across models
Independent verification of outputs
Bias auditing and fairness checks

These approaches align closely with the capability perspective: technology should enhance human well-being, support meaningful participation, and preserve agency.

Consider, for example, a health-tech startup using AI to prioritize patient care. A purely performance-driven algorithm might maximize throughput while unintentionally neglecting patients with complex conditions, particularly when relevant information is missing or poorly represented in the training data. A human-centered approach changes what we evaluate. Instead of optimizing only for efficiency, the system would also need to account for patient complexity, fairness, and clinician judgment. The goal becomes not simply faster decision-making, but more responsible decision support.

From intelligent automation to human-centered intelligence

The same principle applies across organizations. Whether we are designing AI tools, conducting research, or building decision systems, the goal should not simply be intelligent automation. It should be human-centered intelligence.

Evaluating AI for human flourishing therefore means asking:

Does this system improve human capability?
Does it support meaningful decision-making?
Does it respect dignity and context?
Does it increase trust and accountability?
Does it help individuals and organizations make better decisions?

These questions reposition AI from a purely technical tool toward a decision-support system grounded in human values.

For professional enquiries, you can reach me at contact@aderayoadelanwa.com.