Artificial intelligence is increasingly treated as an expected part of modern software. Sentinel has deliberately not incorporated AI into its production surveillance or outbreak-response workflows.
This is not a rejection of AI. It is a decision about sequencing and value. Sentinel is still building out the deterministic methods needed for surveillance and response, and those methods currently offer more practical value than adding probabilistic behaviour to the product.
Deterministic by design
Public health software needs to behave predictably. Given the same information and configuration, Sentinel should produce the same result. A user should be able to identify the rule that was applied, understand why an outcome occurred and reproduce it during review.
This principle informs features such as machine-readable case definitions, survey mappings and calculations, laboratory-message processing and task generation. The logic is explicit and can be tested against known scenarios.
Probabilistic systems answer a different kind of question. Their output may vary, and the reasons for a particular result may be harder to inspect. That does not make them without value, but it creates additional requirements for validation, governance, monitoring and human oversight.
Complete the transparent methods first
Sentinel's immediate priority is to complete and strengthen the deterministic foundation: structured records, configurable data collection, reproducible classification, reviewable workflows, reporting and controlled public health actions.
There is substantial work still to do within that model. Adding AI before those methods are mature could make the platform more complex without solving the most important operational problems. Operational scalability comes first from making supported changes explicit, configurable and repeatable.
Where AI may have a future role
Probabilistic methods may have a place where they address a clearly defined problem and perform better than a transparent alternative. Possible uses could include assisting with narrative review, suggesting structured information for confirmation, drafting summaries or highlighting records that may require attention.
Any future use would need to support rather than obscure the surveillance process. AI-generated output should be clearly identified, appropriately reviewed and evaluated against a defined standard. Decisions affecting classification, investigation or public health action should not become untraceable simply because a model has been introduced.
A potential future Sentinel application
Emerging technologies may offer opportunities to improve the reach, accessibility and scalability of public health interventions. Conversational AI is a particularly promising application, potentially enabling automated case interviews, contact tracing and delivery of personalised public health advice through familiar communication platforms such as WhatsApp or SMS.
Multilingual capabilities could improve access for populations who do not speak the predominant local language, while natural language processing could interpret narrative responses and translate them into structured surveillance data. Such systems could substantially expand contact tracing capacity during an emergency while allowing public health practitioners to concentrate on complex investigations and individuals requiring additional support.
However, their implementation would require careful evaluation of accuracy, privacy, public acceptability and the reliability of information collected, with clear arrangements for human oversight and escalation. Integration with established case-management systems would be essential to ensure that information collected translates into coordinated public health action.
Conversation
Public health unit
Hello, I'm contacting you from the public health unit. Would you prefer to continue in English or another language?
Individual
Español, por favor.
Public health unit
Por supuesto. ¿Necesita ayuda para permanecer en casa de forma segura?
Individual
Sí, vivo con mi abuela y no tengo otra habitación.
Proposed structured output to Sentinel
This also complements Sentinel's planned task-barrier functionality. An AI interview could identify an individual's difficulty complying with isolation advice, populate the relevant structured fields and initiate a reviewed referral workflow.
The important architectural distinction is that the AI would provide a conversational interface to Sentinel's deterministic workflows. The underlying case management, permissions, referral eligibility and reporting would remain governed by explicit rules rather than unrestricted AI decisions.
AI in development and testing
AI has already been useful during Sentinel's development, particularly for regression testing and the generation of synthetic test data. It can help create varied mock patient records and HL7 messages that exercise the platform's workflows without using real patient information.
This includes testing patient matching, validating incoming laboratory messages, applying configured disease mappings and case definitions, detecting regressions, and checking that expected tasks or public health interventions are created.
In this role, AI helps test the deterministic system. It expands the range of scenarios that can be exercised and makes it easier to repeat those tests as the software changes. It is not making operational decisions inside the deployed product.
A deliberate sequence
Sentinel's approach is to build transparent and testable methods first, then consider probabilistic methods where there is a specific benefit that can be measured and governed.
AI may become part of the platform in the future. If it does, it should be introduced because it improves a defined surveillance workflow—not because AI is expected to appear in every product.