Native explainability
Explanations are generated from the decision process itself, not approximated after the fact.
Hybrid Intelligence
Learning and reasoning in one auditable architecture.
Hybrid Intelligence is UMNAI's approach to building decision systems that are adaptive, explainable and governable.
It combines neural learning, symbolic reasoning and causal understanding, so decisions are not just predicted; they can be inspected, challenged and improved.
The architecture
AI that learns from data and reasons with rules.
Pattern-based AI is powerful but hard to explain. Rule-based systems are transparent but rigid. Hybrid Intelligence brings both together: it learns from data, connects those patterns with human knowledge and policy, and reasons through cause and effect to produce decisions that can be explained.
Three components carry the work. XNNs learn patterns from data, the Symbolic Hypergraph links logic and causality, and ESMs turn the result into explanations.
- XNN
Explainable Neural Networks
XNNs decompose complex predictions into modules and partitions, exposing the logic and attributions behind each decision. - SHG
Symbolic Hypergraph
The hypergraph connects features, rules, constraints, causal relationships and human knowledge through hyperedges the system can reason over. - ESM
Explanation Structure Models
ESMs organise XNN outputs into structured explanations, global, local and contextual, that different audiences can understand.
Why it matters
What makes it different.
Causal reasoning
The system can support “why,” “what if” and “what would change the outcome” questions.
Human knowledge injection
Experts can add, review and update rules without treating the system as an untouchable black box.
Evidence chains
Decision artefacts retain the context needed for audit, assurance and governance.
From induction to monitoring
Adapt without losing control.
Hybrid Intelligence supports the full lifecycle of a decision engine, preserving traceability while performance improves.
01 / 03Once
Induce
Bootstrap the decision engine, learning structure from data and codified domain knowledge.
02 / 03Continuously
Monitor
Watch live behaviour: drift, out-of-distribution inputs, explanation stability and rule coverage.
03 / 03On signal
Refine
Update affected components in place when monitoring surfaces a real change, then hand back to Monitor.
Go deeper
Explore the research behind Hybrid Intelligence.
Read more about the technical foundations, safety model and need for a different AI architecture.