Ultra high-velocity AI-native leadership decision insights
Decisive AI was born out of the realization that government & industry leaders routinely make decisions with extremely sparse data that fails to inform them about the full spectrum of possibilities and the causal importance of irrationally large parametric combinations of options. They cannot ask and answer the hard, important, critical questions which will lead to the optimized outcomes. They are forced to focus on the questions their analytical infrastructure can resolve.
In general, trade spaces with more than 4-5 descriptive parameters are np-hard; i.e. they cannot be thoroughly data mined in an amount of time that is useful to humans. This forces analysts to use experience and intuition to construct small toy problems with only 2-3 independent parameters; even if they have access to powerful SaaS cloud compute that can process thousands of interventions (simulations) at a time, Design of Experiments (DOE), or other approaches. As the old saying goes, “they have to look for their glasses under the streetlights” because the balance of the street is computationally inaccessible.
We make the hard things easy…
Key IP: Collaboration
Decisive Models Marketplace
Model Based Systems Engineering (MBSE) Data Curation (SysML v2)
Unifying models ontology
Universal Models Protocol (UMP)
Rapid drag-n-drop analysis authoring
Key IP: Defeasible Logic
Battle Management is always complicated, e.g. “do not fire into a civilian zone, unless under direct fire, unless the target is a noncombatant.”
Classical first-order logic can handle the first rule “do not fire into a civilian zone” but completely fails to assess the multiple exceptions that commonly arise in combat.
DeFAb adjudicates natural language LLM BM instructions in real time, keeping pace with AI Battle Managers
Creates a natural language audit log of all decisions
Can be employed in both connected and disconnected (AI BM decisions made without human oversight) command situations
The DeFAb knowledge base currently includes: 89 strict rules, 32 defeasible rules, 27 defeaters; all audited by a polynomial time verifier. These cover restricted zones, friendly fire, target classification, self-defense, force ratio conservativity, and mission overrides.
Decisive AI solutions are built for leaders that struggle with…
Key IP: Causal Inference
Active Causal Experimentalist (ACE) Framework
World-leading best-in-class performance
Radical algorithmic training acceleration on low-cost NVIDIA H100 NVL GPUs
Optimizes 30-50 parameter trade spaces with trivial #s of interventions relative to Design of Experiments or full-grid
Node and parameter significance
The “shape” of parameter dependencies
Optimum parameter settings
