How Can Boards Adopt AI While Maintaining Diversity of Thought?
Picture a board meeting where one director, the one with a computer science background from a decade at a tech firm, fields every AI-related question. The rest of the table nods along, relieved someone else is carrying it. That board looks AI-ready on paper. In practice, it has concentrated its judgement in a single person whose comfort with the technology far exceeds that of the rest of the board.
The instinct to solve AI readiness this way is understandable. Appointing an expert feels like progress, and it is, as far as it goes. But governing AI well is a board-wide capability, not a delegated one, and the difference matters more than most boards currently admit.
Enough literacy to govern is different from technical fluency. Capability needs to be spread across the table, not concentrated in one seat, and it needs to keep developing as the technology does. That only works, though, if it doesn't cost the board the thing that made oversight worth having in the first place: people who see a problem differently enough to challenge each other.
In this article:
Why AI readiness is more than appointing one expert
Should every board have an AI expert? Having one helps, measurably. Boards with even a single director fluent in AI governance are 2.8 times better at spotting the technology's risks early, according to National Association of Corporate Directors research. The problem is what happens next: most boards stop there, treating one appointment as the finished project instead of the starting point.
The gap this leaves is stark. Eighty-three per cent of S&P 500 boards have formally named AI a material risk to the business, yet just 2.7 per cent of the directors sitting on those boards hold any disclosed AI expertise, according to The Conference Board's 2026 analysis. Spencer Stuart's most recent index found technology experience among new S&P 500 directors nearly tripling since 2021, up to 46 per cent of new seats filled. Boards are hiring for this. They aren't yet spreading it.
A single expert seat has a structural weakness that has nothing to do with the expert's competence. Remove that director from the room, and the rest of the board has no independent basis to challenge an AI-related recommendation. Worse, the other directors can stop trying without anyone noticing: if someone else is clearly the AI person, everyone else's incentive to stay sharp on the subject erodes. The board ends up with expertise in one seat and a widening comprehension gap in every other one.
Building collective AI capability
How should boards build AI capability, then, if not by hiring their way out of the problem? The honest answer is slower and less satisfying than a single appointment: shared understanding, built deliberately, reviewed often.
AI literacy versus technical expertise
Literacy and technical expertise are not the same skill, and boards that conflate them end up either over-investing in training nobody needed or under-investing in the judgement that matters most.
A director does not need to understand how a model is trained to ask a sharp question about what it was trained on, what its known failure modes are, or why a particular recommendation should be trusted. That's the level of literacy a board needs: enough to interrogate AI-assisted material with the same rigour applied to a set of accounts, without every director becoming a data scientist. Technical depth belongs with specialists the board can call on. Literacy belongs with everyone at the table.
Continuous learning
AI capability has a shelf life, and boards that treat a single training session as complete are working from a picture that goes stale within a year, sometimes faster.
What keeps pace looks less like a course and more like a habit: a standing point on the calendar to refresh understanding, external briefings that stay current instead of getting recycled, and a genuine expectation that directors bring questions from their own use of AI tools back into the boardroom. Reviewing board capability on the same rhythm as reviewing risk appetite or strategy keeps it from falling behind the technology it's meant to oversee.
Board-wide capability
Concentrating AI oversight in one person creates a single point of failure, and the fix isn’t another expert seat. It’s making sure enough of the board can do the job so that, when the person in the room changes, the board’s capacity to challenge AI-assisted material doesn’t change with them.
Practically, that means directors doing more of their own preparation instead of pushing it upward: using AI tools themselves to interrogate board papers before a meeting, testing an unfamiliar assumption, flagging which sections of a long paper deserve closer reading, and arriving at the discussion with an independently formed view, formed before anyone hears what the resident expert thinks.
As Mike Soutar put it on the Boardroom Confidential podcast, coming into the room with a clear point of view changes how you participate once you are there: “I’m very clear about my point of view […] and that allows me to not be trying to do the work in the room. It allows me to be curious and open to what other people have got to say.”
A board where every director has done that preparation argues differently, and better, than one where eleven people are waiting for the twelfth to speak. Our guide, 5 Practical Ways to Build an AI-Ready Board, covers the governance habits behind this in more depth.
Why diversity of thought still matters
AI capability is supposed to sharpen board discussion, not replace the range of views that makes discussion worth having in the first place. That distinction gets lost the moment a confident, well-produced AI-generated recommendation starts functioning as the default answer nobody bothers to argue with.
There’s another kind of diversity at stake here too, and it’s easy to overlook. Put a recommendation from an AI tool in front of a board made up of former CFOs, and they’ll mostly ask about the numbers. A director who has run a customer service operation might ask who the model was trained on. Someone who has worked in a different market might notice that the assumptions don’t travel. None of that comes from being more technical. It comes from having lived and worked differently. Boards that look alike tend to miss the same things, and AI output, which arrives polished and pre-analysed, makes those shared blind spots harder to see. The more of the board’s thinking the technology does upfront, the more it matters who is in the room to question it.
Anna Marks, Chair of Deloitte's Global Board of Directors, has warned that boards routinely mistake silence for agreement. An absence of debate “implies agreement,” as she puts it, which is exactly how a weak bet sails through unchallenged and a strong one dies for want of anyone willing to champion it. AI makes that risk sharper, not smaller. A recommendation arriving with a confidence score attached is easy to mistake for a settled question instead of an opening position.
Researchers studying AI's effect on board decision-making have proposed a useful discipline for this. A recent paper in the journal AI & Society describes it as epistemic independence: boards deliberately cultivating divergent analysis, sometimes formally, so the group doesn't converge prematurely around whatever the AI output happens to say. The mechanism matters less than the habit. A Chair who actively invites the dissenting view, and treats a lone objection as data, not friction, stops AI-generated insight from becoming the room's only opinion by default.
It echoes a point made by Chair and NED Charlotte Valeur on the Boardroom Confidential podcast: “Disagreement is key for proper decision-making processes.” In an AI-enabled boardroom, preserving that disagreement becomes more important, not less.
Security is a useful test of whether this discipline is real. A well-configured AI system can flag that a risk sitting inside a recommendation needs review by a specific expert, a CISO assessing a data-exposure risk buried in a vendor contract, for instance, before the board proceeds. That handoff only works if the board treats the flagged expert's view as a genuine second opinion. Treated as a formality, it just delays a decision the room had already made. AI surfacing the need for a different perspective is only useful if the board is still willing to hear one. A related piece, “How AI Is Changing Boardroom Dynamics,” looks at how this shift is reshaping meetings more broadly.
The test of a board building real AI capability
Three questions cut through most of the noise. Could a majority of the board, not just one director, explain why an AI-assisted recommendation should or shouldn't be trusted? Has a director outside the board's designated AI expert ever pushed back on an AI-generated view and changed the outcome? And when AI flags a need for specialist input, does that specialist's view get heard, or does the board proceed anyway once the box is ticked? A board unsure about any of these has expertise in one seat and a widening gap in every other one.
How Sherpany supports board capability and better governance
Boards can build AI capability while preserving effective boardroom dynamics when the tools supporting individual preparation make challenge easier, not something one expert does on everyone else's behalf.
Sherpany's AI features help directors interrogate board papers on their own terms before a meeting, surfacing questions and flagging what warrants closer scrutiny, so every director arrives having formed an independent view instead of deferring to whoever speaks first on an AI-related item. That preparation is what turns a board discussion into an actual discussion, not a briefing one director gives to the rest.
None of this decides anything on a board's behalf, and none of it substitutes for the range of perspectives that make a board's challenge worth having. It gives every director, not just the one with the relevant background, the tools to show up ready to ask a hard question.
The Whole Board Has to Ask the Question
An AI-ready board isn't the one with the most technical expertise in the room; it's the one where enough directors can ask a sharp question, independently, that AI-generated material gets the same scrutiny as any other significant input to a decision.
That capability has to be built deliberately, spread across the table instead of parked in one seat, and renewed often enough to keep pace with a technology that won't hold still. None of it works without the thing boards had before AI arrived and still need now: directors who see problems differently enough to challenge each other, and a Chair who makes sure that challenge happens instead of assuming silence means everyone agrees.
If you would like to see how Sherpany supports board-wide AI capability without weakening the challenge that makes boards effective, book a free demo and find out how Sherpany can help.