Extending AI capability across an integrated services platform
Sierron provides AI and modernization advisory in support of MD THINK, the Maryland Total Human Services Integrated Network, focused on where artificial intelligence contributes value on a platform serving multiple agencies and programs.
Program context
MD THINK, the Maryland Total Human Services Integrated Network, is a cloud based platform delivering integrated health and human services across state agencies. Announced in 2017 and funded initially through a federal grant of approximately $195 million, the platform consolidates services spanning the Department of Human Services with health, juvenile services, and labor programs.
Within five years the program completed modernization and statewide deployment of its human services applications, enabling decommissioning of the legacy mainframe environment. The platform is cloud native, built on containerization and microservices, and provides a consolidated view of individuals receiving benefits across programs that previously maintained separate records. Operational responsibility for the platform has more recently shifted toward the Department of Information Technology.
Challenge
An integrated platform changes the artificial intelligence question. On separate legacy systems, the constraint is that data cannot be brought together. On an integrated platform, data can be brought together, which makes the governing question what may be brought together, for which purpose, and under whose authority.
MD THINK spans programs with different statutory bases and different confidentiality regimes. Child welfare records, benefits eligibility data, juvenile services information, and health data carry distinct handling requirements. Consolidation that is appropriate for one purpose may be prohibited for another, and the boundaries are legal rather than technical.
A second consideration follows from the platform’s success. Where applications are modernized and data is consolidated, the practical barriers to deploying artificial intelligence are lower than in most state environments. Lower barriers accelerate adoption, which makes governance a prerequisite rather than a subsequent step.
Sierron’s role
Sierron leads artificial intelligence approach and governance for platform based services, identifying use cases, defining data use boundaries across programs, and aligning to state AI governance requirements.
Approach
Sierron begins with permissible use rather than technical feasibility. On an integrated platform the determining questions are which program’s data may inform which program’s decision, what authority supports cross program use, and where consolidation that is technically straightforward is legally constrained. These questions are answered with program counsel and policy staff, not with architecture.
Use case identification concentrates on work internal to caseworkers and program staff, where volume is high, procedure is documented, review precedes consequence, and operational baselines already exist. Case triage, document intake, records classification, and correspondence handling meet these conditions across multiple programs on the platform, which means a capability proven in one program may transfer to another. Transfer is a genuine advantage of platform architecture and one of the stronger arguments for beginning at the operational layer.
Access control requires specific design attention. Application layer controls assume users reach data through an application enforcing program specific rules. Retrieval based approaches on a consolidated platform can surface content across program boundaries to a user authorized for only one of them. Authorization must be enforceable at retrieval time and must reflect the confidentiality regime governing each source.
Sierron documents, in advance and in writing, which determinations affecting individuals and families will not be delegated to or influenced by automated systems. In human services this is a consequential boundary. Determinations affecting child welfare, benefits eligibility, and family services carry direct effect on people with limited capacity to absorb error, and the boundary belongs in place before capability exists.
Governance aligns to the NIST AI Risk Management Framework’s Govern, Map, Measure, and Manage functions and to Maryland AI Enablement Strategy priorities, supporting consistency as sector specific frameworks contemplated by the state strategy are developed.
Outcomes to date
The engagement is ongoing. Work delivered to date includes:
- Identification of permissible cross program data use boundaries ahead of capability design
- Prioritization of operational use cases with transferability across programs on the platform
- Retrieval time authorization requirements reflecting program specific confidentiality regimes
- Documented determinations affecting individuals and families that remain human decisions
- Framework alignment to NIST AI Risk Management Framework structure and state AI strategy priorities