AI Orientation Observatory

1. A shared measurement substrate, not just shared principles. The world has more than fifty AI principle documents and almost no comparable indicators. Success means the Dialogue endorses or commissions a working group to deliver by 2027 the following: a standardized, multi-dimensional Conscious Care Index that scores AI systems, policies and research on a common scale (care orientation, diversity, governance, geographic equity, extraction risk, inclusion and human agency). Without measurement, every subsequent commitment is unverifiable. 2. Operational equity, not aspirational equity. Success means concrete commitments that move resources, not just rhetoric: a Global South care-AI capacity fund tied to the domains where developing countries' research already leads (healthcare, social impact, agriculture); machine-readable disclosure of gender, founder and monetization data on regulated AI; and an explicit recognition that the care premium observed in sub-Saharan African AI research (≈30–35% care-orientation vs 26.5% global average) is a comparative advantage to be invested in, not a deficit to be patched. 3. A standing evidence layer for the Dialogue itself. The Dialogue will be returning every two years. Success means it commits, by July 2026, to maintain or sponsor an open, public observatory of the global AI ecosystem: orientation, geography, policy gaps, research-product gaps, that all 193 Member States can interrogate. This converts the Dialogue from an episodic forum into a continuously informed governance loop, anchored in real data rather than the loudest delegations. If the first Dialogue produces a measurement track, an equity track and an evidence layer, every future session has a foundation to build on. Without them, 2027 will repeat 2026.

• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Interoperability of governance approaches
• Protection and promotion of human rights
• Transparency, accountability, and human oversight

These four are where evidence most clearly identifies underinvestment. Implications of AI (cultural, social, economic): Across our analysis of 67,433 AI applications and 30,100 companies, the global AI Conscious Care Index sits at 39.5/100: utility-dominant, with weak inclusion (26.9) and near-zero geographic equity. The cultural and economic implications of AI are downstream of what AI is built for. Until value orientation is a first-order metric, every other governance debate addresses symptoms. Capacity-building: Our Research-Product Gap shows Healthcare carries a 40.32× research-to-product ratio - vast public-good research that has not become accessible tools. Sub-Saharan African research demonstrates a measurable care premium (South Africa 34.6%, Nigeria 33.9% vs 26.5% global average). Capacity-building must follow these signals: invest where Global South strength already lies, rather than replicating Northern product paths. Transparency, accountability and human oversight: Of 387 tracked AI policies, average CII is 60.3 - frameworks are inconsistently scored on rights, equity and enforcement. Of 870 extraction-oriented apps, only 2 use participatory governance framing. Citizens cannot hold accountable what they cannot see. Disclosure of CII profiles, founder diversity, monetization model and governance framing should be a baseline. Open-source, open data, open models: The Dialogue's authority depends on its evidence base being public and contestable. We treat open observatories as digital public infrastructure: an open dashboard of the global AI ecosystem (1,005,937 papers, 175 countries, 387 policies, 30,100 companies) is a model for how UN-aligned, machine-readable, openly-licensed evidence can power policy across Member States. While the other three themes matter, these four are where measurement, equity and openness compound across all of them.

Three cross-cutting issues are missing or insufficiently visible in the seven themes: 1. Value orientation as governance infrastructure. The seven themes treat AI as a single technology to be made safe, accessible and rights-respecting. But our data shows AI is not one thing - it is a portfolio of orientations (care, utility, creative, extraction) with radically different societal effects. Governance without an explicit value-orientation lens regulates the symptom, not the cause. Adding a measurement track - a UN-endorsed Conscious Care Index, would let every existing theme be evaluated on a comparable scale. 2. The policy-innovation gap. Cross-referencing 688 OECD AI policy papers against 69,675 commercial AI products across 16 industries reveals systematic mis-targeting: Media & Content Creation is +48% over-watched, Legal +32.6%, HR +32.4%, while Developer Tools & Infrastructure (the substrate of the next decade's AI) is -25.2% under-watched. Compounding domains (developer tooling, agentic infrastructure, foundation-model supply chains) are invisible to policy. The Dialogue should treat this gap itself as a governance failure mode worth a dedicated work-stream. 3. Anticipatory / forecasting governance. All seven themes are reactive. None ask Member States to project 2-, 5- and 10-year capability horizons before legislating. We propose a "Possibility Engine" obligation: before any major AI bill, governments publish a forward-looking equity, care and risk projection, ensuring policy is designed for what AI will be, not what it was when the bill was tabled. Cross-cutting recommendation: adding a fifth track - Measurement, value orientation and anticipatory governance as the connective tissue across the existing four clusters. Otherwise, the Dialogue produces parallel rhetoric without comparable, forward-looking evidence.

Operating globally with research grounded across 175 countries, 30,100 AI companies and 387 AI policies, we observe distinct regional patterns that reshape what "urgent action" should mean. Regional realities (from our live data): Africa: high-care, low-capacity. Sub-Saharan AI research carries a 33% care orientation (vs 26.5% global avg), South Africa 34.6%, Nigeria 33.9%. Yet only 888 papers across the continent and Healthcare's 40.32× research-to-product ratio mean care-led research is stranded. Asia-Pacific: volume without orientation. 268,024 papers but only 24.3% care (China 21.4%, South Korea 28.7%, India 30.8%). Highest absolute output, lowest average orientation, governance must steer scale toward purpose. Latin America & Caribbean: promising, under-resourced. 30.4% care orientation (Brazil 32.4%) but only 1,021 research papers and 25 tracked policies. Policy frameworks score well (avg CII 64.8) — under-investment, not under-ambition. Western Europe & Other: mixed. 31% research care, but extraction is geographically concentrated: of 870 extraction-oriented apps, 49% are U.S.-built, with Germany (101), UK (72) and Canada (24) following. International bodies (UN/OECD/EU) average CII 93.7: strongest framing globally, but enforcement reach is the weakest link. Cross-cutting impact: Mis-targeted regulation: OECD attention is concentrated where AI is visible (Media +48% over-watched), not where it compounds (Developer Tools −25.2% under-watched). Care premium going to waste in Africa and parts of Asia for lack of capacity infrastructure. Transparency floor too low - only 2 of 870 extraction apps adopt participatory governance. Opportunities within reach: a standardized Conscious Care Index, a Global South care-AI capacity track, and open observatories as digital public infrastructure, turning regional asymmetries into directed investment, not abstract regret.

The Dialogue's unique value is not to create yet another framework — it is to make every existing framework comparable, contestable and consequential. Our analysis of 387 tracked AI policies shows international bodies score highest on framing (avg CII 93.7) but weakest on enforcement reach. The Dialogue can fill exactly that delta. We see four roles the Dialogue is uniquely positioned to play: 1. Convener of a measurement substrate. Endorse or commission a UN-aligned Conscious Care Index so the OECD AI Principles, EU AI Act, UNESCO Recommendation, ISO/IEC 42001 and African Union AI strategies can be compared on one scale. Today they cannot. Without a common index, "international cooperation" is fifty parallel monologues. 2. Standing evidence layer. With the Dialogue returning every two years, it should anchor a continuously-maintained, openly-licensed observatory of the global AI ecosystem (orientation, geography, policy gaps, research-product gaps) accessible to all 193 Member States. Cooperation requires shared facts updated faster than two-year cycles. 3. Equity broker. The Dialogue is the only forum where Africa, LAC and small island states sit at the same table as foundation-model labs. It can broker directed cooperation e.g., capital-rich states co-investing in Healthcare and Social Impact AI built by Global South innovators (where research-to-product ratios reach 40.32× and 13.86×) rather than re-licensing Northern products back. 4. Anticipatory governance hub. Most existing initiatives regulate retrospectively. The Dialogue can normalize forward-looking obligations, capability horizons, equity projections, care-impact forecasts to enable cooperation that is designed for what AI will be, not what it was. Ideally hence, the Dialogue's role should not to be one more table; but rather to make all the other tables interoperable.

The AI governance ecosystem is dense but disconnected. The Dialogue should connect, not duplicate, this existing scaffolding: Standards & principle frameworks: OECD AI Principles, UNESCO Recommendation on the Ethics of AI, EU AI Act, NIST AI Risk Management Framework, ISO/IEC 42001, Council of Europe AI Convention, African Union Continental AI Strategy, ASEAN Guide on AI Governance and Ethics, GPAI, Bletchley/Seoul/Paris AI Safety Summit declarations. UN system initiatives: the AI Advisory Body, ITU AI for Good, UNESCO ROAM-X, OHCHR's B-Tech project, UNCTAD on data governance, UNDP's AI Hub for Sustainable Development, ITU/WIPO standardization tracks. Capacity & access: World Bank Digital Development Partnership, UNICEF AI for Children, Smart Africa AI Strategy, IDB's fAIr LAC, Indian Digital Public Infrastructure stack. Independent observatories & evidence layers: Stanford HAI AI Index, OECD.AI Policy Observatory, Mozilla Foundation Data Futures Lab, Partnership on AI, civil society observatories (including ours) cataloguing 1,005,937 papers across 175 countries. Added value the Dialogue can uniquely bring: > A measurement bridge: Today these initiatives use incompatible taxonomies. The Dialogue can endorse a Conscious Care Index that translates across them — turning fifty principle documents into one comparable scale. > An equity-weighted evidence base: Most existing indices are Northern-funded and Northern-framed. The Dialogue can ensure the Global South care premium (33% in sub-Saharan Africa vs 26.5% global) and research-product gap dynamics inform every cooperation track. > A 193-Member-State legitimacy floor: OECD speaks for 38 countries; the EU AI Act binds 27. Only the Dialogue carries universal legitimacy as a precondition for any future binding instrument. > Continuity: Existing initiatives are episodic (annual indices, biennial summits). The Dialogue can demand a standing observatory fed by these initiatives, transforming snapshots into a live governance loop.

Each stakeholder type holds a different kind of evidence. The Dialogue's structure should make those evidence types complementary, not competing. Member States bring policy reality. Contribute country-level CII baselines, machine-readable disclosure of regulatory frameworks, and named focal points for the standing observatory. Of 387 tracked AI policies, average CII is 60.3 — a baseline every state can report against. Academia & research community bring the long arc. Contribute peer-reviewed evidence on equity, value-orientation and capability horizons. Among 1,005,937 mapped AI papers, the gap between research production (Asia-Pacific 268,024 papers) and care orientation (24.3% there vs 33% in Africa) is a research question begging an institutional response. Civil society & affected communities bring lived consequence. Without their voice, the Dialogue regulates AI as engineers describe it, not as citizens experience it. Their inputs should weight oversight, accountability and rights design. Private sector bring deployment data. Contribute disclosure on monetization model, founder diversity, governance framing, the four signals our Observatory tracks across 30,100 companies. Today only 2 of 870 extraction-oriented apps adopt participatory governance; mandatory contribution closes that gap. Technical community bring testable specifications. Open-source models, evaluation suites, benchmark equity audits and reproducible measurement pipelines. International organizations & UN system bring scaffolding. ITU, UNESCO, OECD, OHCHR each hold a piece of the puzzle. Recommended structural features: Three-track design at every session: Measurement, Equity, Anticipatory Governance, so all four thematic clusters are evaluated against the same crosscutting lens. Mandatory regional rapporteurs from each of the five UN regions, ensuring no cluster's outputs reflect only one region's priorities.

The data is unambiguous about who is missing. 1. Sub-Saharan Africa & low-income states. Sub-Saharan Africa publishes AI research at 33% care orientation (vs 26.5% global) (a comparative advantage) yet the entire continent contributes only 888 papers in our dataset of 1,005,937. The voice with the highest care premium has the smallest microphone. 2. Latin America & the Caribbean. 1,021 papers, 25 tracked policies, 30.4% care orientation present at the principle table, absent at the funding table. 3. Small Island Developing States and least-developed countries. Often subsumed under "Global South" but face distinct AI risks (climate-AI, ocean-AI, sovereignty-AI) that need named representation. 4. Women founders & non-binary technologists. Extraction-oriented AI shows disproportionately lower female founder representation; participatory governance framings appear in only 2 of 870 extraction apps. Without representation in production, rights design downstream is decided by the absent. 5. Indigenous communities & non-dominant linguistic groups. Linguistic and cultural implications named in §4(c) of the Resolution will remain rhetorical unless these communities co-design evaluation criteria for what "respectful AI" means in their context. 6. Workers, informal economy actors & affected publics. Labor impacts of AI are debated about workers, rarely with them. Same for gig and informal economy workers in LAC and Africa. 7. Children and future generations. Whose AI environment is being built today, but who have no formal seat. Inclusion mechanisms: Travel-and-time bursaries so participation isn't gated by funding. Named regional rapporteurs from each UN region with veto-protected speaking time. Affected-community advisory panels for each thematic cluster. Multilingual contribution channels (written, audio, video) not English-only office forms. Pre-Dialogue listening tours in under-represented regions, results published openly.

Traditional plenaries reward the most-resourced delegations. Geneva 2026 should pioneer formats that surface evidence and dissent. 1. Live-data sessions, not slide-decks. Open the Dialogue with a real-time projection from a public observatory - global AICCI score, regional care orientation, policy-innovation gaps, recent extraction-app counts. Speakers respond to live data, not pre-baked reports. 2. "Possibility Engine" scenario rooms. Each thematic cluster runs a structured 90-minute foresight session: 2-, 5- and 10-year capability horizons in that domain projected by AI experts; equity and care implications scored on a common scale; Member States respond. Converts the Dialogue from reactive to anticipatory. 3. Reverse panels. Affected communities (workers, civil society, Indigenous representatives, children's advocates) sit at the front; Member States and tech companies sit in the audience and respond. Inverts the standard hierarchy. 4. Evidence challenges. Stakeholders submit testable claims pre-Dialogue ("country X under-watches developer-tooling AI by 25%"). The Secretariat publishes a validation report. Disagreements are debated with the data on screen. 5. Cluster sprints, not plenaries. Replace one of the two days' plenary blocks with parallel 48-hour drafting sprints mixing stakeholder types: measurement specialists, civil society, Member States, technical community: co-authoring concrete deliverables (a CII v0.1, a policy-gap dashboard schema, a Global South capacity track funding template). 6. "Care Track" pre-meetings. A standing pre-Dialogue gathering of under-represented voices whose synthesized recommendations open each thematic session, ensuring the loudest delegation doesn't define the starting frame. 7. Open observability throughout. Every session live-streamed, transcribed, machine-translated into the six UN languages, and indexed against the standing observatory for tracking commitments made in Geneva for follow-through. The Dialogue's format should embody what it asks of AI itself: transparent, participatory, evidence-based, equitable.

Effective AI governance already exists in scattered form. The challenge is connection. Five categories of working examples: 1. Measurement & evidence platforms. >Stanford HAI AI Index: annual global indicators, publicly licensed. > OECD AI Policy Observatory: comparative policy data across 70+ countries. > AI Orientation Observatory: open dashboard of 1,005,937 papers, 175 countries, 387 policies and 30,100 companies, classified on a 7-dimension Conscious Care Index. Demonstrates that orientation, equity and policy-innovation gaps can be tracked in real time. 2. Standards-setting that bind. > EU AI Act: risk-tiered, justiciable, sector-aware. The first major framework with teeth. > ISO/IEC 42001 - first certifiable AI management system standard. > Council of Europe AI Convention: first international treaty with human-rights anchoring. 3. Equity-led capacity initiatives. > Smart Africa AI Strategy & AU Continental AI Strategy: Global South-led roadmaps. > fAIr LAC (Inter-American Development Bank): region-specific responsible-AI playbook. > UNICEF AI for Children: value-orientation embedded by design. > India's Digital Public Infrastructure stack: open-source, sovereign, scalable digital public goods, replicable across LDCs. 4. Transparency & accountability practice. > Mozilla's Data Futures Lab: open-source AI auditing. > Algorithmic Justice League: community-led bias audits. > Partnership on AI's transparency reporting framework: voluntary disclosure templates. > Hugging Face Model Cards / Dataset Cards: practical artifact-level disclosure. 5. Anticipatory governance. > UK AI Safety Institute & emerging US/Singapore/EU AISI network: capability-evaluation infrastructure. > GPAI working groups on future-of-work, climate AI. > Possibility Engine-style scenario forecasting (AI Orientation Observatory) structured 2-/5-/10-year capability projections with equity and care impact scoring. The Dialogue's contribution should be to interconnect them through a shared measurement substrate (the Conscious Care Index), a standing evidence layer, and 193-Member-State legitimacy.