Capability Lens

Does AI widen what people can actually do and be?

Capability Lens turns Amartya Sen's capability approach into a working method, applies it to generative and agentic AI, and tests it on 44 real-world cases, from a farmer checking prices in her own dialect to an agent built to manipulate a lonely teenager. It ends in 42 recommendations and a theory of change, grounded in an evidence base of 872 screened interventions and 35 national AI strategies,, compared with risk-based AI regulation worldwide.

The scope is global. Where the study refers to AI law, it uses the EU AI Act's risk tiers as a reference model, because many other jurisdictions have drawn on them, not because the EU is the target.

What's inside

6stages from resources to agency
14canonical cases used to test the method
44AI use cases, urban and rural
29 → 8dimensions considered, spider-chart axes
5regulatory risk tiers mapped to every case
10findings
42recommendations in 10 action areas
33real-world programmes and laws matched to the recommendations
1theory of change linking actions to impact
872interventions screened in the evidence base (362 accepted)
35national AI strategies analysed
30+interactive charts, labs and simulators

Nine parts, one method

1 · For classrooms and newcomers

Learn the method

A 16-step interactive lesson that explains every symbol "like you're 5", then in math, then with the nuances.

  • Build functioning vectors, move money and rights sliders, flip conversion factors
  • See capability sets as dot clouds; compare freedom four ways
  • Agency map, fixed points, quiz and a sentence-to-symbols writer
Open the lesson
2 · For researchers

Formal model

The general, domain-agnostic formalization: primitives, stages, axioms, five propositions, remaining assumptions and open modelling decisions.

  • Reproduces Sen's 1985 formula as a special case
  • Validated on 14 cases from the capability literature
Read the model
3 · For analysts

AI and freedom

Where generative and agentic AI enter the formula, and whose options grow or shrink.

  • Persona lab with access gates, AI modes and a reliance slider
  • Delegation dial, jagged-frontier demo, positional arms race
  • Access policies compared across seven people
Run the models
4 · For everyone

Case atlas

44 cases scored on eight dimensions, with verdicts and regulatory risk tiers.

  • Filterable library with spider charts and side-by-side comparison
  • Heat map, theme radars, short- vs long-term and options vs agency plots
  • Legal tier vs capability verdict matrix
Explore the atlas
Evidence · For everyone

Evidence base

All 872 screened interventions, from the UN Activities on AI report, OECD.AI national policies, the practice scan and partner programmes, with the integrated analysis across every source.

  • Integrated analysis: every formula dimension, interventions, evidence strength and strategy commitments side by side
  • Who is changed: people, duty-bearers and builders, AI systems, institutions
  • Filter and search every item by source, entry point, area, evidence and tier
Open the evidence base
5 · For policymakers

Findings and actions

thirteen findings and 42 recommendations, with who should act and when.

  • Recommendation explorer by actor and timeframe
  • Thematic areas: risks reduced, benefits increased
  • What-if simulator and how far risk-based regulation covers each action group
See the recommendations
6 · For implementers and funders

Real-world practice

33 programmes, products and laws already widening capabilities or protecting agency, matched to the recommendations.

  • From Farmer.Chat and Rori to UK scam reimbursement and California's companion-chatbot law
  • Coverage map: 27 of 42 recommendations have real examples
  • Seven lessons from practice, and where practice is missing (agent governance)
See what works
7 · For strategy and evaluation

Theory of change

How the ten action areas turn into wider real freedom, with assumptions, evidence, loops, indicators and phasing.

  • Clickable pathway map from problems to impact
  • Pathways for five people, from a rural farmer to an older adult
  • Critical assumptions, indicators tied to the formula, three phases
Open the theory of change
8 · For governments and policy analysts

National AI strategies

35 strategies, policies and action plans from five regions, read through the capability formula.

  • What strategies commit to, by formula dimension and action area
  • Strategy-by-dimension heat map and country profiles
  • The agency blind spot: almost no commitments on who decides or on manipulation
Explore the strategies

Seven things we found

AI helps most by filling gaps people cannot bridge alone: language, literacy, sight, distance, missing experts. In the illustrative atlas, rural livelihoods score highest.
Getting AI is not the same as benefiting from it. 2.2 billion people are offline; others lack AI in their language or the skill to check it.
Many benefits are fragile. 12 of the 44 illustrative cases look good today but erode skills, relationships or independence later.
Agents trade control for convenience. People reach more but decide less; whose goals the agent serves decides the outcome.
A quarter of cases shrink freedom, mostly for people who never chose to use AI: scams, manipulation, deepfake abuse, automated screening.
National AI strategies have an agency blind spot. Across 35 strategies, 4 concrete commitments address who decides when AI is involved and 1 addresses manipulation; a third build governance and infrastructure.
Risk-based AI laws catch the worst harms, not the slow ones. 15 of the 44 illustrative cases rated risky carry only disclosure duties or none.

Quality and confidence

An internal audit aligned terms, numbers and categories across all tabs (see the crosswalk in the Formal model). Claims carry confidence labels: established (external studies), indicated (patterns in the evidence base), illustrative (the 44 atlas cases, built to explain the framework) and hypothesis (the theory of change). Open work: calibrating prominence scores, verified coding of non-English strategies, harm databases, missing regions, and fielding the survey.

Suggested paths

Teaching a class (2 × 50 min)
  1. Learn the method, steps 1–9
  2. Steps 10–15, then the quiz
  3. Pick three cases in the atlas and argue their scores
Briefing a policy team (30 min)
  1. This page, then the theory of change
  2. Findings and actions: the what-if simulator
  3. Regulatory fit and sequencing
Designing or funding a programme
  1. Real-world practice: lessons and coverage map
  2. Match your idea to a recommendation in Findings
  3. Check the matching atlas cases for risks
Research and critique
  1. Formal model and its open decisions
  2. AI and freedom models
  3. Atlas scores, to test and replace with field data