Democratising AI: Towards Open, Decentralised AI Ecosystems

Review

The ORF volume on “Democratising AI” moves the conversation beyond access to chatbots and reframes democratisation as access, participation, and governance across the AI stack. In other words, it offers a strong “why” and a solid “what.”

That layered framing is a real strength. It forces us to ask where centralisation actually lives: data, models, compute, deployment, and standards.

The report also avoids the simplistic “open vs closed” binary. It treats openness as a spectrum and acknowledges hard realities: interoperability gaps, compute asymmetries, auditability challenges, incident response, and the political economy of concentration. For a Global South lens, this is strategically valuable.

But “decentralisation” is doing too much conceptual work. At different points it means distributed infrastructure, multi-stakeholder governance, reduced Big Tech concentration, open weights, and data ownership reform. These are not equivalent. Each addresses a different failure mode. Without a decision framework, decentralisation risks becoming an aspiration rather than an engineering choice.

The recommendations are directionally correct: standards, transparency, audit, regional models, and interoperability. Yet the operating model is underdeveloped. Who sets the standards? What does conformance look like? How is compliance verified? What are the incentives for adoption? Strategy is strong. Instrumentation is thin.

There is also limited analysis of re-centralisation by convenience. Even open ecosystems accumulate power through default SDKs, curated registries, cloud credits, and distribution hubs. Topology does not automatically produce democratic governance.

The work highlights the need to complete the following before we build an executable architecture:

• Define types of decentralisation and associated trade-offs.
• Introduce conformance profiles and testable baselines.
• Add a threat model annex mapping adversary goals to controls.
• Clarify sequencing and funding models.

Key Insight

AI democratisation only becomes operational when “decentralisation” is broken into testable design choices, conformance rules, and incentives rather than treated as a feel-good umbrella term.