Ideas
I've worked on AI since 2010 — years before ChatGPT — and I've never given the same talk about AI twice. But underneath the talks, a few convictions haven't moved. These are the ideas I keep bringing into rooms of leaders, whether they run agencies, companies, foundations, or universities.
Leadership is what fills the permissive space.
Regulation will lag. Structures will allow. The option space of AI — like that of any keystone technology — is enormous, for good and for ill. In that space, the scarce resource isn't compute or talent; it's judgment: the wisdom and restraint to ask, before anything else, what's the problem we want to solve for? Plenty of capable people will tell you the principles are the friction — the thing to engineer around on the way to the metric. They have it backwards. The choosing enacts values, whether or not anyone names them. This is the leadership challenge of our time.
The most consequential decisions stay human.
Every framework helps, and every framework leaves a residual. Sometimes a system produces an answer no one has seen before — the move no human would have played — and in the moment you cannot know whether you are looking at brilliance or at noise. That is the novel-or-noise problem, and no amount of documentation decides it for you. What remains is a human judgment call. Leaders who understand that keep the most consequential decisions where they belong.
Most AI ethics questions are old questions in new clothes.
There's a market hard at work selling leaders the idea that AI ethics is so unprecedented they're helpless without a priesthood. I don't buy it. While the framing may have changed, whether an algorithm or policy should treat everyone the same or correct for unequal starting points, for example, isn't a new question — leaders have wrestled with versions of it for generations. A lot of what you need to meet this moment, you already have.
Who benefits?
My standing test is this: expand the benefits so they're broadly shared, and reduce the harms. It's simple to say — all the work is in the contextual details, and the sharpest version of the question is always who benefits.
Freedom is only as real as the options in front of you.
It's the oldest question in leadership: who gets to choose — and how. The aim is expanding the freedom people have in shaping their own life. But freedom is thin where opportunity is thin, and a technology that hands more choices to people who already had plenty hasn't expanded much of anything. That's what the standing test is really asking. Widening the circle of people with real options is the work, and it does not happen on its own. Someone decides.
Leadership across boundaries.
In twenty-five years advising leaders — my teams ran AI trainings for audiences from the European Parliament to the White House and Congress, from Fortune 50 CEOs to philanthropists, executives from healthcare to education and the arts — the pattern I trust most is cross-sector: nonprofit executives learning from government, healthcare professionals from universities, and back again. Every engagement feeds the next room. Expertise is diffuse. No single room holds enough of it, and assembling those rooms is not secondary work — it's the cross-sector learning and leadership we need to meet the moment.
People are capable of so much more than they realize.
It's how I end my courses and my team farewells, and I mean it literary: awareness of capacity is the first step to agency. AI is fundamentally an unleashing of human creativity — for good and for ill. So the leadership question is never just what the tools can do. It's the older one: given this context, how are we going to make the world a better place?
Views expressed here are John's own, in his personal capacity — he does not speak for Berkeley or the Fung Institute.