Ask an AI tool if it's unsure about something. It won't tell you. It will give you an answer with the same confidence whether it's right or wrong. How are companies navigating that gap?
Last of three from the same summit. If you haven't already, paper one covers the incentive structures, paper two covers what this means for the customer experience side.
The AI Consensus
What struck me across this panel wasn't disagreement. It was how many leaders, independently, landed on the exact same framing. Douglas Abrams, CIO & VP of J&J's Innovative Medicine NA Business and Global Commercial Strategy Organization at Johnson & Johnson, said it plainly: "AI should absolutely amplify the human judgment, but never replace the human judgment."
The wins cited were concrete: ambient listening tools that give clinicians time back at the bedside, adopted so quickly that providers wouldn't give them up once rolled out, AI-assisted diagnostics, faster drug discovery. The risk cited was just as concrete. AI doesn't hedge. It gives its best answer with total confidence every time, whether that confidence is earned or not. In a clinical or educational setting, that's the difference between a tool that supports judgment and one that quietly erodes it.
The Unglamorous Prerequisite
Before any of that AI value shows up, there's a less exciting problem that has to get solved first: the data has to actually be usable.
Amy Perry, CEO of Banner Health, described consolidating 29 petabytes of data that had previously lived across somewhere between 16 and 60 fragmented platforms. Northwell has invested more than a billion dollars in digital infrastructure toward the same goal. Neither system led with AI. Both led with the unglamorous work of getting their own data into a state where AI could do anything useful with it at all.
Trust Is the Actual Bottleneck
The most interesting data point from the summit came from Michael Giuliano, President of Planetree International, citing research on what actually builds trust between an organization and the people it serves: relationships have the highest correlation to creating trust, higher than judgment, higher than consistency.
That reframes the AI adoption question from a technology problem to a relationship gap. The goal isn't to be the only source of information a patient or student trusts. It's to be the trusted source they return to after they've already gone looking elsewhere.
Jimmy St. Louis, Founder & CEO of Agentis Longevity, framed adoption as something earned in sequence: responsible execution first, trust with early adopters second, broader adoption after that. Skip the sequence and the technology doesn't matter.
There's a real equity risk underneath all of this. AI fluency isn't evenly distributed, and populations without access to it, or without comfort using it, risk being left further behind exactly as the systems around them get more efficient for everyone else.
What This Means for Healthcare Education
Most curricula still treat AI fluency and clinical judgment as separate tracks, one a tech skill bolted on, the other the real education. This summit suggests that split doesn't hold up much longer.
The graduates who will be most valuable aren't the ones who can operate an AI tool. They're the ones who can tell when the tool's confident answer is wrong, and who understand that their actual job is to be the trusted source a patient or client returns to, not to compete with the tool for the same territory.
Three questions to consider:
Is AI fluency taught alongside clinical or professional judgment, as one integrated skill, or does it still sit in its own separate module?
Are graduates being trained to catch a confident wrong answer, or only to use the tool that produced it?
Does the curriculum build the relationship and communication skills that research says actually drive trust, or does it assume trust follows automatically from technical competence?
If it's useful to work through what that alignment looks like for a specific program or portfolio company, I'd welcome a conversation about what we've found.
This is the last of three pieces from the same summit.