Recognition
Can they see what's happening?
The ability to recognise when and how AI systems are influencing the decision context — including subtle framing effects, authority cues, and information filtering.
Our Methodology
How we measure what others assume
The Discernment Framework
Most organisations assume their people can exercise meaningful oversight of AI systems. We don't assume — we measure. Our methodology examines four interconnected dimensions of human discernment capacity.
Evaluation
Can they assess what they see?
The capacity to evaluate AI outputs critically — distinguishing between well-reasoned recommendations and confidently-stated errors, between genuine insights and systematic biases.
Agency
Can they act on their assessment?
The practical ability to override, modify, or reject AI recommendations when warranted — including psychological readiness to disagree with an "expert" system.
Accountability
Can they take responsibility?
Clear chains of responsibility for decisions made with AI involvement — knowing who decided what, why, and how to trace decisions back to accountable humans.
How we assess
The Discernment Snapshot uses three complementary methods.
Structured Interview
Guided conversation exploring how decisions actually happen in your context. Where does AI input come in? How is it weighted? What triggers override decisions? What's documented and why?
Mini-Simulation
Brief scenario exercise that surfaces real responses to AI influence under controlled conditions. Not a test — a diagnostic. How do you actually react when the system is confident and you're uncertain?
Method Demonstration
Live walkthrough of how we measure discernment — showing the concrete, reproducible chain from observation to assessment to recommendation. Proof that this isn't subjective opinion.
What makes this different
Not AI training
Training gives knowledge. Knowledge doesn't guarantee capacity. We measure whether people can apply what they know under real conditions.
Instead: capacity assessment
We test actual response patterns, not theoretical understanding. Can they recognise, evaluate, and act — not just recite principles?
Not compliance audit
Compliance checks whether you have policies. It doesn't test whether people can follow them when the AI sounds authoritative.
Instead: operational truth
We examine what happens in practice, not what the documentation says should happen. The gap between policy and behaviour is where risk lives.
Not generic assessment
Abstract frameworks don't capture context-specific vulnerabilities. "AI oversight" looks different in every decision environment.
Instead: context-specific
We assess one bounded decision context at a time. The output reflects your specific situation, not generic best practices.
Theoretical foundation
The Discernment methodology draws on established research in cognitive science, human factors, and automation psychology — adapted for the specific challenges of superintelligent systems.
Automation bias research
Decades of evidence on how humans over-rely on automated systems, particularly when those systems are presented as expert or authoritative.
Decision support systems
Understanding of how advisory systems shape human judgment — and what conditions enable humans to maintain independent evaluation.
Cognitive load theory
Recognition that meaningful oversight requires cognitive resources. Overloaded humans default to system recommendations.
Organisational psychology
How responsibility diffuses in complex systems, and what structures preserve clear accountability chains.
See it in action
The Discernment Snapshot applies this methodology to one concrete decision context.
Learn about the Snapshot