Preparing Small Island States for AI
Thank you so very much, Chair. Honorable chairs, excellencies, distinguished delegates, participants all: Saint Vincent and the Grenadines appreciates this opportunity to share its ideas, and of course, we acknowledge the value of opportunities such as this one. We know what vulnerability looks like. We also know what resilience looks like, because we live it and survive it with every hurricane and volcanic eruption that has decimated our shores. Every AI conversation, including conversations relative to vulnerability and natural disasters, assumes the existence of a prerequisite upon which too few SIDS can act.
And by SIDS, of course, I mean small island developing states. The assumption is that we as SIDS countries can generate the right type of data so that AI can manipulate it meaningfully in service of our people. But here is the uncomfortable truth: the University of Malta suggests that 71% of SIDS lack quality data suitable for artificial intelligence. Half of all SIDS have no officially stated AI initiatives, and 43% have no data governance frameworks. The reality is that we do not collect proper data, and if we have it, we do not have it in a form that is easily accessible.
It is the daily reality of our public servants in Kingstown, Saint Vincent, who cannot run a workforce planning model because personnel records sit in filing cabinets across 11 ministries. It is a reality for nurses who cannot flag a patient for follow-up because the clinic's register never connects to the hospital's main system. It's the reality of a school principal in one of our Grenadine islands who cannot see whether a child who missed September schooling has ever re-enrolled, because the education database and the social protection database do not speak to each other.
Deploying agentic AI from a faulty data foundation would not yield intelligent results. Saint Vincent and the Grenadines therefore proposes a practical [unclear] approach to data readiness, centered on three domains where AI can deliver immediate public value if the data foundation is correctly laid. We see human resource management across government as being one of these areas. Our public service is made up of more than 7,000 civil servants. The HR data manages payroll in one system, vacation leave in another. Appraisals are done on paper; staff training is reflected on a spreadsheet that is not connected in any way with the other data.
AI, therefore, cannot optimize allocation, read the leaves of attrition, or find the right person for vacancies if it is asked to assess fragmented and unstructured data. What we need is a government-wide HR data standard, a common schema that defines our employee records in order to be able to conduct machine learning. Once this happens, we will have systems of record, payroll, and leave, for example, which will use a common set of definitions. This is not a technology challenge, but a data challenge, and it is entirely solvable, we think, within a year, as long as we all agree on the standard before buying any technology tools.
With the standard in place, we could begin to answer critical questions: Which programs are at risk of duplication? How should the ministry allocate its finite resources in the most efficient manner possible? In a small island state with finite resources, such questions, you can imagine, are existential. Everything for us is about equity, and every dollar we spend on a duplicated program is a dollar we do not spend where it can better be utilized to impact the lives of our people.
We think of coordination of programs in healthcare. The health system, again, faces the same fragmentation plaguing the civil service. There are thousands of maternity records, immunization logs, and other emergency department data that exist in silos, on paper, probably, without any structural linkages. When, in 2021, our volcano erupted, we learned how vital the need is to track displaced patients between facilities, to monitor medication supply chains, and deploy clinical staff with urgency. Without this structured data, we responded as humans with amazing resolve.
But you could imagine that with the right data and the assistance of AI, we would be able to deploy even better quality results. Data governance before AI deployment is therefore imperative. None of these possibilities become real unless we first build the governance framework. Saint Vincent and the Grenadines welcomed the Action Brief's emphasis on beginning with the public need and governing data and risk throughout the work. We would add, from our SIDS perspective, four principles that we feel must accompany any data readiness effort.
Data standardization must come before AI procurement. No vendor should define our data architecture. We define the schema; the tools adapt to it, not the reverse. Data interoperability must be mandated across ministries. Separated data is a governance risk. AI training on incomplete data generates incorrect data. Data sovereignty is key, as we cannot afford to be at the mercy of large corporations or jurisdictions we cannot regulate. The data must live where we govern it. And data capacity must be built alongside data infrastructure. Training public servants to generate, clean, structure, and govern data is the main goal. So what do we ask?
We propose that an AI for the World program supports SIDS-specific data readiness pilots that make AI possible and useful. We are prepared to nominate two public institutions, one in health, one in education, for the first cohort of this program, with defined service problems, committed project leads, and the political will to establish the data standards AI requires. But let us be honest about what we need from partners: not products, but preparation; not algorithms, but architectures. The AI revolution will allow citizens from all over our nation to self-generate the right data, structured, standardized, interoperable, and governed, so that AI can manipulate it in service of our nurses, our teachers, our administrators, and our people.
As we ask to be included in the future of AI, we insist also on being prepared for it. Thank you.