Work

SIMA — Co-shaping AI ethics with civil society

A survey of 67 civil-society organisations and interviews with 41 regional decision-makers, asking the same questions about AI from both ends of the policy table.

Organisation
Culture for Change — Civil society and AI policy
Engagement
2026
Our role
Research design, fieldwork and analysis — one of two researchers

Past

Where they were, and what we found

Σ.Η.μα was written to move civil society from observer to co-shaper of the rules governing AI. Seven work streams, fourteen months, three regions, a participatory Action Plan at the end. Research sat second on that list, and everything after it depended on the first two.

What the project inherited:

  • No number for what Greek cultural organisations knew about AI, what they used, or what they feared.
  • Whether the state’s commitment to co-design with civil society had reached the regional offices was unverified.
  • Two vocabularies over one subject: organisations described their work, the policy texts described systems, and nobody had put them in the same frame.

Present

The moment we were inside it

Two questions that could not share an instrument

Organisations needed a description of their situation: which tools they use, what they know about their rights under the AI Act, where the pressure is felt. Regions needed an assessment of intent: whether a policy exists, who owns it, whether civil society appears anywhere in it.

Run as one survey, both would have collapsed into opinion. Asked of only one side, the result that matters — that the two descriptions do not match — would have been invisible.

One design, not two studies

We treated the strands as a single design: convergent parallel mixed methods. The questionnaire and the interviews ran separately, integrated at interpretation, both anchored to the same three texts — the UNESCO Recommendation on the Ethics of AI, the EU AI Act and Greece’s Blueprint for AI Transformation.

The dimensions had to be identical on both sides. That is what turns a mismatch into a measurement. Organisations at low familiarity against decision-makers reporting no demand for it is a finding. Collected separately and written into two chapters, it is two anecdotes.

What the sample was allowed to claim

  • Define the population before sampling it — the ΟΚοιΠ definition, formal and informal, register-based, because the denominator is the claim.
  • Build the frame from the registry (roughly 900 organisations), the funder’s network and snowballing, and instrument it in Greek.
  • Interview 41 regional decision-makers across Attica, Central Macedonia and Crete against the same dimensions.

The cost was reach. A register-based frame is honest about who it did not hear from, and 67 responses are not the sector: 57% of them came from Attica. That limitation is on the first page of the report, not in a footnote.

Future

What we handed over

Delivered in July 2026: a 5,000-word report and a 250-word account of process, with the findings arranged so each one lands on a work stream that follows it — the empowerment workshop, the eight roundtables, the Action Plan. The instruments, the response sets and the analysis database went with it.

Culture for Change published the report in August 2026.

Today

Where it stands now

The study’s numbers, as it produced them: 67 organisations and 41 decision-makers. Of the organisations, 53 of 67 use AI tools at some level and 48 of 67 have no internal policy but want one. 38 of 64 say they know the EU AI Act not at all; 49 of 66 say the same of the national strategy.

Our contract ended with the report. The roundtables that follow are theirs to run, and the evidence is in the room now — which is what changes who gets consulted.

We are proud to have been one of the two researchers on it.

← All work