What TedAI Taught Me About the Future of Healthcare Intelligence
What TedAI Taught Me About the Future of Healthcare Intelligence
I've been to a lot of healthcare technology conferences. The kind where you sit through five panels on "digital transformation" and leave with a tote bag and a business card you'll never use. TedAI was not that.
I came away from the conference with three ideas I haven't been able to stop thinking about. Not because they were novel abstractions, but because each one had a direct — and in some cases, urgent — implication for how we build, deploy, and lead AI in healthcare. I want to share them here, not as a recap, but as the starting point for a conversation I think our industry needs to have.
1. We Are Still Living in the Architecture of a 2017 Breakthrough
One of the most quietly extraordinary moments of the conference was hearing from Llion Jones, one of the original co-authors of Attention Is All You Need — the 2017 paper that introduced the Transformer architecture and, effectively, laid the foundation for every large language model in existence today.
What struck me wasn't the technical detail. It was the reminder that the AI systems reshaping healthcare right now — the ambient clinical documentation tools, the diagnostic co-pilots, the patient communication layers — are all downstream of a single architectural breakthrough made less than a decade ago by a small team of researchers.
That's a remarkably short timeline for a technology now being embedded into clinical workflows. When we debate AI readiness in healthcare, we're not really debating a mature technology with decades of real-world evidence. We're debating whether to deploy something that is, at its core, still very young.
That humility matters. Not to slow us down — but to inform how we build governance around it.
2. Simplification Is the New Superpower — and Most Leaders Are Ignoring It
May Habib, CEO of Writer, delivered what may have been the most practically useful message of the entire conference: as AI capabilities explode in complexity, the most valuable leadership skill is not AI literacy. It's simplification.
The leaders who will define this era aren't the ones who can explain how a model works. They're the ones who can take overwhelming technical possibility and translate it into a clear, executable priority for their organization.
This resonates deeply with what I see in MedTech right now. Organizations are drowning in AI options — every vendor promising transformation, every pilot showing promising early results. The bottleneck isn't capability. It's clarity. The question isn't "what can AI do for us?" It's "what exactly are we trying to change, for whom, and how will we know it worked?"
That's a leadership question, not a technology question. And in my experience, it's the question most organizations are still struggling to answer before they start spending.
3. The Shift from Probabilistic to Deterministic AI — and Why Healthcare Should Be Paying Close Attention
The most forward-looking conversation at TedAI was a closing discussion hinting at what comes after today's probabilistic AI models. The suggestion — not yet a prediction, but a serious directional signal — is that the field is beginning to move toward more deterministic AI foundations: systems that don't just produce probable answers, but provable ones.
For most applications, the distinction is interesting. For healthcare, it is potentially everything.
Today's AI systems, including the most sophisticated diagnostic and clinical tools, operate probabilistically. They produce outputs that are likely correct — often impressively, reliably likely — but not provably correct. A radiologist reviewing an AI flag understands this. A regulatory pathway accounts for it. A liability framework tries to manage it.
But what happens when that changes? What happens when AI systems can provide outputs with verifiable logical guarantees rather than statistical confidence intervals?
The implications for clinical validation, FDA submission pathways, physician trust, and ultimately patient outcomes are significant. More deterministic AI won't arrive overnight, and the path there is genuinely hard. But the direction matters for how we're building infrastructure today.
Are we designing clinical AI systems, data architectures, and regulatory strategies that can evolve as the underlying technology shifts? Or are we building for the probabilistic present in ways that will have to be torn down and rebuilt when the paradigm changes?
The Uncomfortable Question I Left With
Here's what I couldn't stop thinking about on the flight home:
Most of the AI being deployed in healthcare today — including tools I've advised organizations to adopt — is being evaluated against a standard of good enough compared to current practice. Is it better than no AI? Is it better than the human doing this task without assistance?
Those are the right questions for now. But TedAI reminded me that we're building on top of a rapidly evolving foundation. The Transformer architecture that underpins today's clinical AI tools wasn't designed for healthcare. It was repurposed for it. And while repurposing has produced remarkable results, it also means we're operating with borrowed infrastructure — systems whose failure modes and limitations we are still learning to understand in clinical contexts.
The leaders I most respect in this space hold two things simultaneously: genuine optimism about what AI will enable in healthcare, and genuine intellectual humility about what we don't yet know.
That's not a reason to slow down. The cost of delayed AI adoption in healthcare is measured in real patient outcomes. But it is a reason to be deliberate about how we build, how we validate, and how we lead organizations through a transition that is not yet complete.
The frontier is extraordinary. I'd just rather we navigate it with our eyes open.
These are the conversations I'm genuinely having with operators, investors, and clinical leaders. If they resonate — or if you see it differently — I'd welcome the exchange.