The AI Probe Rule
Three probe questions that intervene between an AI’s contradiction and the leader’s reflex, turning disagreement into information rather than threat or verdict.
Probe-Before-Dismiss-or-Defer is the practice that intervenes when AI returns a view that contradicts the leader’s. Two reflex moves arrive faster than the merit evaluation that should govern engagement: the dismiss reflex protects the existing view; the defer reflex outsources the judgement. The probe is the short, structured run that goes in between. This was named and refined through ChangeSchool’s work with senior leaders across our executive education programmes, as part of the modular blocks of our Judgement and Governance curriculum.
​
The three probe questions
Run them in order. Five to ten minutes if the leader is honest with themselves.
​
Probe-Before-Dismiss-or-Defer, a three-path decision tree. When AI returns a contradicting view, the leader has three routes: dismiss (protect the existing view), defer (outsource the judgement), or probe (run three questions before deciding). The probe path forks into Q1 evidence, Q2 framing, Q3 unstated premise, then converges on a named verdict that records which path was taken and why.
​
One. What evidence is the AI weighting that I am not? Ask the AI directly to name the two or three pieces of evidence its conclusion most depends on. Then ask whether each is one the leader had on the table. AI retrieves patterns from a wider reading without the leader’s filters; the disagreement may be sitting on evidence the leader has access to but has not foregrounded.
​
Two. Is the framing the AI is using one I have considered and rejected, or one I have not seen? A disagreement can come from the same evidence read through a different frame. Ask the AI to name its frame in one sentence. Compare to yours. The question is whether the framing choice has been made consciously, with the alternative in front of the leader, or by default.
​
Three. Does the disagreement reveal an unstated premise of mine? The hardest of the three. The AI may be exposing an assumption so natural to the leader it has dropped below conscious view (we will always have access to this client; senior associates will absorb the change without leaving). When the leader can name the premise, they can decide whether to endorse it. When they cannot, the premise is running the decision without their consent.
​
The probe is not a commitment to change the leader’s mind. It is a commitment to know what the leader is doing when they don’t.
​
Genesis
The probe sits within Andrew Likierman’s account of trust as one of the six elements of judgement (Judgement at Work, Profile Books, 2025): the discipline of identifying who and what to rely on for honest, relevant input, and of evaluating an argument on its merit rather than on the authority of its source. The probe extends the same discipline to AI as an interlocutor whose authority is unstable and whose disagreement still carries information. Underneath sit three frames built for human disagreement: Daniel Kahneman (2011, Thinking, Fast and Slow) on adversarial collaboration, the disagreement is the most concentrated source of information about what neither party has yet seen; Daniel Dennett’s steelman move (Intuition Pumps, 2013), which is to restate the opposing argument in a form the opponent would recognise as fair, before disagreeing; and Roger Martin’s integrative thinking (The Opposable Mind, 2007), which is to hold the two opposing models in mind without prematurely collapsing into either.
​
Why it matters now
Two pulls drive the dismiss reflex. Identity protection, the senior leader’s professional self is built on judgement; taking seriously a contradiction from a tool admits the judgement might not be self-sufficient. Frame protection, the leader has spent six weeks getting to a frame; a contradiction through a different frame is a tax on the work already done. Two pulls drive the defer reflex. Authority confusion, AI is right often enough on bounded factual questions that the authority creeps across into questions where its answer is one credible view among several, not the answer. Cognitive offload, holding the tension is hard work; picking a side is easier. All four pulls are escapes from the holding move; the probe keeps the leader inside it.
​
The moves
Steelman the AI’s argument before responding. Type the strongest restatement of the AI’s view; only then type the response. Run the three probe questions, in order, typing answers even briefly. Decide on the verdict and name the reason, I have considered the AI’s frame and rejected it because… The named reason is the difference between a judgement and a reflex.
​
How ChangeSchool applies it with executives
We run senior cohorts through the Disagreement Probe: each leader brings a current decision where AI has returned a contradicting view, runs the three probe questions in front of a peer, and reads back the named verdict. The exercise reliably surfaces which of the three questions most often catches the leader out.
​
The discipline
A disagreement log kept as the leader goes: what the AI said, what was decided, why. Re-read monthly to surface which probe question most often catches the leader out. A second-opinion booking on consequential decisions: before any decision above a threshold, the AI gets a structured pass with the explicit instruction to disagree, the institutional cousin of Likierman’s (2025) devil’s-advocate practice. A premise-naming practice once a fortnight: the leader writes three premises their decision depends on, asks the AI to identify three premises in their framing, and compares.
Run for a quarter, the three habits convert single-decision probes into a standing record of where the leader’s frame holds up to contradiction and where it does not. The log is the audit trail of the practice.
​
​
​
‘When AI disagrees, the disagreement is the data; the probe is what turns it into a decision you can name.’
​
Viren Lall, Managing Director,
ChangeSchool LDN (2026).
virenlall.com/ai-probe-rule
​
​
​
AI for Leaders.
Executive Education that changes practice.
Viren Lall is Managing Director of ChangeSchool LDN, a London-based executive education partner. ChangeSchool specialises in AI for senior-leader development, winning the EFMD Global Excellence in Practice Award in 2023 and 2025, with programmes in 39 countries.
​
Since April 2024, ChangeSchool LDN has been designing and delivering mindset shifts through Executive Education Programmes across sectors such as deep tech, manufacturing, and education, for business owners, governance professionals, and senior leaders. Leaders gain AI fluency, protect decision quality, spot value creation opportunities, and foster human-centric AI use. AI capability for senior leaders is also a core element and a constant spine of our Open Programmes for Chief Digital Officers, Chief Operating Officers, and Chief People Officers, delivered by our partner business schools.
Some of our clients include the Royal Academy of Engineering, Education and Training Foundation, and the UK Government's Meet Smart programme.
For speaking, programme, or partnership enquiries, get in touch with him through ChangeSchool LDN.