Build a parallel, community-rooted AI research institution

Stand up an independent research institute outside the corporate AI labs and the large foundation-policy think tanks, structured so the research agenda is set by the communities most affected by AI rather than by funders or executives. The form is small, distributed, and explicitly counter-institutional: a thin permanent staff, a flexible network of community-embedded researchers, fiscal sponsorship or non-profit incorporation, philanthropic but not Big-Tech funding, and a publication discipline that prioritises affected-community readers alongside academic ones.

An actor chooses this strategy when the existing research apparatus — corporate labs and policy think tanks — is structurally incapable of producing the kind of knowledge the movement needs, because its funding and career incentives point elsewhere. A parallel institute solves the agenda-setting problem at the source: researchers are paid to ask the questions affected communities raise rather than the questions a sponsor will fund.

It trades off scale and durability for independence. Operating outside the dominant funding stream means perpetual fundraising pressure, fragile career paths for the researchers it employs, and a small footprint relative to the institutions it is countering. The strategy depends on a philanthropic ecosystem willing to subsidise non-aligned research — a contingent, revocable condition.

Verdict — good strategy where the community-rooted criterion is structural

The strategy earns its keep where two conditions hold together: the institute is founded around a community-rooted research agenda rather than being foundation-spawned or accreted onto a pre-existing NGO, and the analytical output is paired with substrate-building for affected communities rather than producing only papers and conferencing. The DAIR Institute clears both bars cleanly — the team distributed across 18+ countries with 20+ languages, the Data Workers' Inquiry's WIRM methodology running with nineteen worker-researchers, and the institutional substrate that anchored the Data Labellers Association's Nairobi launch are the strategy in its strongest form. Coding Rights clears the founding criterion and is the strategy's clearest Global-South feminist-tech expression, with the Oracle for Transfeminist Technologies as workshop substrate that has circulated outside its origin community.

The strategy is weaker — but not absent — where the institute is foundation-spawned (Ada Lovelace, Nuffield-established 2018) or grafted onto a long-running portfolio (FMA's 2023 pivot into AI work after a 1987 founding and a 1997 ICT pivot). These adopters do real participatory and policy work, but the "parallel" framing has to be read more loosely: Ada Lovelace as a research-advocacy hybrid that uses participatory deliberation as method, FMA as a long-running digital-rights NGO with AI as one strand among privacy, gender-and-ICT, and democracy-and-ICT. Both clear the spirit of the strategy without sitting at its centre; both should be linked, with the verdict marking the gradient rather than smoothing it.

Two further conditions sharpen the verdict. First, the strategy is structurally exposed to philanthropic-ecology shifts. The Ford / MacArthur / Open Society / Nuffield / Hivos / Luminate cluster underwrites most adopters, and a constriction in that cluster contracts the strategy simultaneously across multiple institutes. The dependency is real, named in DAIR's own founding framing as a structural prerequisite rather than a stylistic preference, and exposes the strategy to political shifts (US right-wing pressure on US foundations, geopolitical re-orientation around Global-South funding) that no single institute can hedge against alone. Second, the strategy produces analysis and substrate, not yet alternative systems. Sefala's spatial-apartheid satellite-imagery work and Ernest Mwebaze's cassava-disease diagnostic work in Sub-Saharan Africa are the strategy's closest approach to deployed alternative AI applications, and both remain prototype research outputs rather than infrastructure communities use. The counter-read in the frontmatter is the load-bearing one to watch as the strategy ages.

Even with these conditions, the strategic case for it is robust. The strategy has produced the methodologies (WIRM; the Oracle deck; the Participatory Data Stewardship framework), the framings (DAIR's "AI is not inevitable, its harms are preventable"; Coding Rights's data colonialism frame from Oppressive A.I.; Ada Lovelace's deliberative-evidence-into-policy frame), and the worker-led substrate that the rest of the movement is now drawing on — and it has done so on a four-year runway (DAIR) and on much longer ones (FMA since 1987 and AI work since 2023, Ada Lovelace since 2018, Coding Rights since 2015). Good strategy; sharp cases.

Ecology

The strategy is reinforced by empirical audit and expose — DAIR carries both strategies, and the institute provides the standing infrastructure for sustained audit work that single-paper research lines cannot. It is reinforced by counter-narrative framing — every adopter uses the institute platform to issue framings that contest dominant AI discourse, and the institute form is what carries those framings into press cycles and policy briefings with durability rather than as one-off interventions. It feeds organise the AI supply-chain workers — DAIR's Data Workers' Inquiry is the most concrete current case of an institute building substrate for worker-led organising, with the Data Labellers Association's Nairobi launch as the visible output and the African Content Moderators Union sitting in the same network of institutional supporters. It overlaps participatory deliberation as policy input at the Ada Lovelace pole, where the two strategies run as twin commitments and the Citizens' Biometrics Council sits inside both. And it shares register with survivor-led testimony as evidence — Coding Rights's Not My A.I. case studies and DAIR's Data Workers' Inquiry produce community-narrative outputs that overlap that strategy's register without being structured as testimony per se.

Adjacent but distinct: the Algorithmic Justice League and AlgorithmWatch carry the algorithmic-accountability research posture but are oriented to audit-and-advocacy more than to the parallel-research-institute form per se. The strategy's distinctive contribution to the corpus is the standing institutional substrate — distinct from a one-off project, distinct from a campaign, distinct from a research collaboration — that compounds over years and produces the methodologies, framings, and worker-led substrate the rest of the movement draws on.