The implementation gap is a data gap
Ambassador Burhan Gafoor opened by moving the diagnosis one step back. The familiar explanations for why the SDGs are off track, financing gaps and capacity gaps, sit on top of something more basic: governments frequently cannot see the problem they are meant to be solving.
He went further than most delegates are willing to, questioning whether the global indicator framework itself, with its lagging and largely unreportable series, is still fit for the purpose it was designed for, and suggesting it may need revision rather than defence.
Making official data AI-ready
The organizing idea of the session was that the risk of AI in public statistics is not that it replaces statisticians but that it routes around them. If a language model cannot read and interpret authoritative data, it will answer from something else, confidently. The proposed remedy is to make official statistics machine-readable and machine-understandable, so that the model has somewhere trustworthy to go.
Singapore’s Chief Statistician set out four conditions for that shift: governance, interoperability, investment in capabilities, and trust. The last, he argued, is not a soft addendum but the load-bearing one, since a statistical system that citizens will not feed with their data cannot produce the statistics anyone needs.
Rwanda’s Deputy Permanent Representative made the same argument from the other direction, describing surveys and censuses as indispensable but no longer sufficient on their own, and insisting that judgement stay in human hands.
The counting angle
The sharpest moment came from the floor. A delegate from Guatemala pointed out that the satellite and AI tools being celebrated rest on data sources that underrepresent indigenous and rural populations, which are precisely the places with the largest development gaps. The tool inherits the gap and then hides it behind a national average.
That is the same failure mode identified at this Forum’s AI for social inclusion side event a day earlier: a community that is undercounted in the statistics is undercounted in the training data, and the system built on top treats it as absent. The responses offered here were institutional rather than technical. Singapore described sandboxing its public statistics chatbot so that it answers only from internal authoritative databases and says the data is unavailable rather than inventing it, and pointed to its AI Verify toolkit for checking whether deployed services return accurate, verifiable data.
The most consequential announcement was quieter. A new expert group on the AI readiness of official data and statistics, established at the 57th session of the UN Statistical Commission in March 2026, will be hosted by the National Institute of Statistics of Rwanda. Its brief is to write the good practice that national statistical offices will follow, with explicit attention to least developed countries and small island developing states. Where the Global Dialogue on AI Governance in Geneva set the rules for AI, this group decides whether the data underneath those rules can be trusted.
Why it matters for the SDGs
The event maps to SDG 17 (data, statistics and capacity), with SDG 9 (digital infrastructure) and SDG 16 (institutions and trust) close behind. It is the clearest statement this Forum produced of SDGCounting’s own premise: measurement is not downstream of implementation, it is a precondition for it. The tension the session left unresolved is that the same wave of digitalization that could finally close the SDG data gap will widen it wherever governance, capacity and trust are not built in first.
Watch & read
- UN Web TV, recording of the side event (9 July 2026).
- UN Statistics Division, and the Data for Now initiative.
- AI for social inclusion, the companion HLPF side event on who gets counted.
- UN Global Dialogue on AI Governance · Full HLPF 2026 coverage.
Quotations are lightly edited from an automated (Otter.ai) transcript of the UN Web TV recording and should be read as close paraphrase. The transcript garbled most speaker names beyond reliable reconstruction, so apart from Ambassador Gafoor, whose name and title were confirmed against public records, speakers are cited by role and institution. Figures quoted by panellists were not independently verified and are omitted where the transcript was unreliable.