The qualities that help leaders reimagine institutions can also prevent them from seeing what their institutions have become
Trust, curiosity and optimism sound like exactly the qualities leaders need in the age of AI.
Charlie Nunn, chief executive of Lloyds Banking Group, recently discussed all three in a McKinsey interview about institutional longevity, technological change and the future of financial services. His argument is an attractive one: enduring organisations survive because they keep learning, keep reinventing themselves and retain the confidence to imagine something better.
I agree.
But these qualities have two very different forms.
They can open an institution to reality. Or they can help an institution tell a more persuasive story about itself.
The difference is whether the institution remains willing to discover that it may be wrong.
Trust must be testable
Trust is fundamental to banking. Customers hand over their money, their personal data and often some of the most consequential decisions of their lives. A bank cannot function without trust.
But institutional trust is too often treated as an inherited asset: something established by history, scale, brand recognition or the authority of senior leaders.
In reality, trust is not what an institution says about itself. It is what its processes allow other people to verify.
This matters when customers present carefully evidenced correspondence about serious process failures and receive no substantive reply. A bank may continue to speak convincingly about customer outcomes, service and trust. But the unanswered evidence creates a gap between the institutional narrative and the citizen’s lived experience.
AI can widen that gap.
An AI system used inside a bank will inevitably be shaped by the bank’s data, policies, risk classifications, legal constraints and approved language. That does not make the system dishonest. But it may make the system incapable of seeing beyond the institution’s own frame.
The AI can become highly intelligent within a reality defined by the organisation deploying it.
We call this captured AI:
AI whose permitted reality is bounded by the interests, data and assumptions of the institution deploying it.
Captured AI does not need to fabricate information. It may summarise the bank’s records accurately, apply its procedures consistently and communicate its established position fluently. Yet it can still reproduce an injustice if the records are incomplete, the procedures are defective or the established position is wrong.
Trust in the age of AI must therefore become testable. Citizens need to know what evidence was considered, what assumptions were made, where judgment entered the process and how a conclusion can be challenged.
Without those rights, greater intelligence may simply produce more sophisticated opacity.
Curiosity must cross the boundary
Curiosity is rightly celebrated as an organisational virtue. Leaders want employees to experiment, learn and ask better questions.
But institutional curiosity often stops at the institutional boundary.
The organisation becomes curious about how to improve efficiency, reduce costs, personalise products and manage risk. It may be much less curious about evidence that challenges its account of what happened.
That is bounded curiosity: exploration is encouraged, provided it does not disturb the assumptions protecting the organisation.
Genuine curiosity is more demanding. It asks:
- What if the customer possesses evidence our records do not contain?
- What if the complaint is not an isolated service failure, but evidence of a process failure?
- What if our AI has learned from decisions that were themselves institutionally biased?
- What information would cause us to change our position?
- Who has the authority to say that the system is wrong?
An institution cannot claim to be learning rapidly if it only learns from information it already controls.
The real test of an intelligent institution is not how quickly it processes confirming evidence. It is how seriously it treats inconvenient evidence from outside itself.
Optimism must leave room for correction
Optimism matters too. Institutions cannot reinvent themselves if their leaders believe decline is inevitable. Technology does create opportunities to widen access, improve services and help more people make sense of their financial lives.
But optimism becomes dangerous when aspiration is allowed to stand in for achievement.
Saying that AI will deepen trust is not evidence that trust has deepened. Saying that technology will improve customer outcomes does not show which outcomes improved, for whom, or according to whose definition.
Responsible optimism is not passive confidence in the institution. It is confidence that the institution can face contrary evidence, correct itself and emerge stronger.
That requires a different kind of leadership question.
Not merely:
How can AI help us deliver our strategy?
But:
How can AI help us discover when our strategy, records or processes are failing the people they were designed to serve?
The first question makes AI an instrument of institutional intention. The second makes it part of an institutional conscience.
Three kinds of AI
The public debate often presents a false choice between trusting institutional AI and opposing it.
There is a third possibility.
Captured AI begins with the institution’s frame and protects its established conclusion.
Adversarial AI begins with the critic’s conclusion and assembles the strongest case against the institution.
Agency AI begins with neither. It helps the citizen identify what is evidenced, what is inferred, what remains unknown and what must be asked next.
That distinction matters because replacing institutional bias with automatic hostility does not restore human agency. It merely changes whose conclusion the technology is designed to defend.
This is why we developed the BIG Checker around a simple public assurance:
Challenge the claim. Don’t manufacture the case.
And a deeper mission:
It does not decide whom to believe. It helps the citizen see what belief would require.
The purpose is not to act as a truth machine. It is to help people recognise the basis of a claim, the options omitted, the gap between evidence and conclusion, the emotional framing being used and the authority on which they are being asked to rely.
Reimagining the relationship, not just the institution
The most important AI transformation in financial services may not be faster processes, more personalised marketing or automated access to something described as advice.
It may be a change in the distribution of intelligence.
Historically, institutions held the records, employed the experts, controlled the processes and wrote the final account. The individual entered the system as a customer, claimant or complainant, but rarely as an equal participant in establishing what had happened.
AI can reinforce that asymmetry. Or it can begin to correct it.
Citizens can now use technology to organise their own evidence, reconstruct chronologies, identify contradictions, test institutional narratives and prepare informed questions. They can retain ownership of their records and grant access to professionals when specialist help becomes necessary.
That is not anti-institutional. Institutions remain essential. Expertise remains valuable. Human judgment remains indispensable.
But the relationship changes.
The citizen no longer has to arrive empty-handed and ask the institution to explain the institution’s version of events.
The leadership test
Trust, curiosity and optimism really can help leaders reimagine institutions in the age of AI.
But only if all three extend beyond the institution itself.
Trust must give citizens the means to verify and challenge.
Curiosity must include evidence that threatens the institutional account.
Optimism must mean confidence in correction, not confidence that the organisation is already right.
The ultimate question is not whether an institution uses AI. Nor even whether its AI is powerful, efficient or safe according to its own controls.
It is whether the institution’s intelligence remains open to realities it does not own.
Because the defining risk of captured AI is not that the machine knows too little.
It is that the machine can learn everything except that the institution may be wrong.
The BIG Checker is a free agency-preserving narrative analysis tool from the Academy of Life Planning. It helps users examine the basis, options, evidential gaps, emotional framing, authority and agency within financial, health and public-interest communications.
Academy of Life Planning 