AI Doesn’t Fix Systems — It Exposes Them

The real transformation was not the nuclear reactor. It was the grid.

Most comparisons between artificial intelligence and nuclear power focus on risk and regulation. Those parallels are real, but they miss the bigger point.

A nuclear power plant is not magic. At its core, it is a steam turbine, a technology that has existed for more than a century. What changed was the energy source. Nuclear reactions made it possible to generate power at a scale and consistency older systems could not match. But that power only mattered once the system around it evolved.

To make nuclear viable, we had to redesign everything around it. Transmission networks expanded. Load balancing improved. New safety systems and monitoring had to be built and continuously enforced. Entire roles and disciplines emerged to manage a new kind of power that was both powerful and unfamiliar. The grid didn’t just require engineering. It required governance, regulation, accountability and trust to make that power safe and usable.

Artificial intelligence is following the same path. The models are not the system. They are the new power source.

Cardiologist Efstathia Andrikopoulou, MD, MBA, sees the problem clearly from inside clinical care. In our conversation, she put it simply: “Detection is not an outcome. We need detection, but detection means nothing unless there are clearly defined actions and a system designed to absorb the follow-up.”

That is the core failure in healthcare AI. The technology is not the bottleneck. What happens after is.

A model may flag disease, risk or deterioration. But without clear workflows, ownership and follow-up, nothing changes for the patient. A result sits in an inbox. A clinician may or may not see it. A patient receives information without context.

We often celebrate detection. We measure accuracy. We compare models. But we rarely ask the most important question: what happens next?

In many cases, the answer is nothing. Or something inconsistent.

That is why AI looks less like software and more like a new power source. Like nuclear energy, its value depends on whether the surrounding system can safely and reliably use what it produces.

The Constraint Is The Grid

Healthcare was not designed to absorb and act on this level of output. Workflows are fragmented, data is siloed, responsibility is often unclear and people navigating some of the most difficult moments in their lives are expected to make sense of it.

AI is producing more signals, but the systems expected to receive and act on them haven’t kept pace. Without a system to carry it forward, a signal turns into noise. And noise isn’t just inefficiency. It’s inconsistency, and in healthcare, inconsistency is risk.

This isn’t a deployment problem. It’s a systems problem. It means embedding AI into real workflows instead of adding it on the side, making it clear who is responsible for follow-up and measuring success based on outcomes rather than model performance.

What nuclear power required is often overlooked. It demanded new safety systems, including regulation, incident response and layered redundancy. It required long-term fuel handling and waste management. It created entirely new roles, from engineers to operators to regulators.

These changes were not made because nuclear energy was flawed. They were made because it was powerful.

Healthcare has not made this transition.

AI tools are introduced but not fully integrated, performance is often measured once or not at all, and when risk is flagged, ownership is often unclear.

We have built the reactor. We have not built the grid.

We are not limited by what AI can produce. We are limited by what our systems are built to absorb.

Regulate The Source. Enable The System

The nuclear analogy also helps clarify governance.

Nuclear energy is tightly regulated because the risks are real. But the goal of regulation is not to stop it. It is to make sure it is used safely.

Artificial intelligence requires the same balance.

We should be rigorous in how models are tested, approved and monitored. But we should avoid treating AI as something to contain. The goal is safe use, not suppression.

In the United States, this idea already exists in how we regulate technology. States focus on how tools are used in real settings, especially when safety is involved. The goal is not to ban the technology itself, but to make sure it is used responsibly and with clear accountability.

AI should follow the same path.

History offers a warning. Early failures shaped how the public sees nuclear power, and that perception still limits adoption today.

We should not make the same mistake with AI.

Infrastructure Is Policy

The most important AI policy decisions are not about the models. They are about the infrastructure that surrounds them.

The bipartisan AI-Ready Data Act introduced by Senators Ted Budd and Andy Kim reflects this shift by focusing on data quality, interoperability and accessibility, the foundations required for AI to function in the real world.

What this bill does is not attempt to regulate the model. It invests in the preconditions that allow models to work at all.

In healthcare, the issue is rarely whether a model can generate insight. It is whether the data exists in a form that allows that insight to be trusted, routed and acted on. When data is fragmented or inconsistent, even accurate outputs struggle to translate into decisions.

That is why the bipartisan nature of this bill stands out. Infrastructure is one of the few areas in AI where alignment is possible, because it avoids debates about restricting capability and instead focuses on enabling responsible use.

If nuclear power required building out the electrical grid, AI requires building out the data layer. Not just for development, but for validation, monitoring and action in real-world environments.

More Power Will Not Fix A Broken System

Artificial intelligence is a breakthrough, but breakthroughs don’t create impact on their own. Systems do.

We’re generating more intelligence than ever before, but the systems responsible for acting on it haven’t been redesigned. Applying a more powerful source to the same system won’t produce better outcomes.

More power will not fix a broken system. It will expose it.

The original version of this blog originally appeared on forbes.com and is reposted with permission.