Skip to content
Search

Latest Stories

Follow Us:
Top Stories

Will Generative AI Robots Replace Surgeons?

How AI-driven surgical robots may reshape medicine, challenge regulators, and test patient trust in the operating room.

Opinion

Will Generative AI Robots Replace Surgeons?

Generative AI and surgical robotics are advancing toward autonomous surgery, raising new questions about safety, regulation, payment models, and trust.

Getty Images, Luis Alvarez

In medicine’s history, the best technologies didn’t just improve clinical practice. They turned traditional medicine on its head.

For example, advances like CT, MRI, and ultrasound machines did more than merely improve diagnostic accuracy. They diminished the importance of the physical exam and the physicians who excelled at it.


Now, an even more radical upheaval is headed for the operating room. Generative AI and surgical robotics are advancing so quickly that procedures once done solely by surgeons may soon be performed autonomously by robots. How soon this happens depends on the willingness of (a) clinicians to support the transition, (b) regulators to approve the applications, and (c) patients to trust a machine with the scalpel.

How Generative AI Plus Robotics Would Perform Surgery

The idea of a robot performing autonomous surgery (that is, completing an operation without a human guiding the controls) sounds like science fiction. But the rapid rise of generative AI since ChatGPT’s debut in 2022 has made it possible.

Hundreds of millions of people use tools like ChatGPT, Gemini, and Claude, yet most have only a vague sense of how large language models work. Contrary to popular assumptions, these large language models do far more than “predict the next word.” If that were all they did, then GenAI tools would routinely produce incoherent paragraphs and nonsensical explanations. Instead, they consistently provide sophisticated reasoning, detailed plans, and expert-level summaries.

They accomplish this through imitation. In medicine, generative AI systems are trained on a massive corpora of medical textbooks, scientific journals, surgical videos, and clinical conversations. With billions of internal parameters, the model learns to mimic how humans solve diagnostic problems, interpret images, and execute procedural tasks. With more training, its responses increasingly resemble those of expert clinicians.

Today’s models can already describe the precise steps required to remove a gallbladder. But executing those steps requires two additional capabilities:

  • Training on thousands of real surgical cases to learn how expert surgeons perform operative procedures.
  • A physical mechanism capable of translating surgical steps into precise movements.

That’s where existing surgical robots come in.

Modern Surgical Robotics: The Missing Link

Over the past two decades, operative robots have allowed doctors to work through smaller incisions with enhanced visualization, increased precision, and tremor-free control.

A typical robotic procedure involves the surgeon sitting at a console, watching a high-definition video feed of the operative field. Then, using hand controls, the physician directs the robot’s arms. The robot carries out movements inside the patient with sub-millimeter accuracy.

For generative AI to operate autonomously, developers would provide information and video from tens of thousands of recorded procedures. The large language model would analyze the data coming from the operative cameras inside the patient and match it to the precise hand movements surgeons make at the console in response. Over time, the model would learn to reproduce the same stimulus-response patterns that expert surgeons use — just as it currently learns to generate accurate answers or create videos following verbal or written prompts.

This approach parallels how self-driving cars are trained. But unlike the chaos of city streets (where vehicles, cyclists, and pedestrians constantly move in unpredictable ways) an operating room is a controlled space, and human anatomy is more predictable than the external environment. Generative AI will have an easier time distinguishing anatomical structures than a self-driving car has when deciding whether an object leaving the curb is a scrap of paper, a rolling ball, or a child running into the street.

In the meantime, progress in robotics continues to accelerate. Elon Musk recently predicted that Tesla’s humanoid robot project Optimus would soon be able to perform “sophisticated medical procedures — perhaps things that humans can’t even do.”

Putting The Robotic Pieces Together

The building blocks for autonomous robotic surgery already exist. Whether it becomes reality in five years or 10 will depend less on technological progress and more on how quickly and effectively hospitals, surgeons, and technology companies collaborate to train these systems.

Three changes are needed now to prepare for that future.

A. Payment models must be updated

U.S. healthcare’s fee-for-service reimbursement system rewards higher volume, not superior clinical outcomes. Hospitals earn more when operations take longer. And when reimbursement is tied to each inpatient day, there is little financial incentive to schedule procedures after hours or on weekends.

Shifting to bundled payments (a single rate for an entire surgical episode) will financially reward hospitals that perform safe, efficient surgical procedures at night and on weekends when surgeons aren’t available. By eliminating clinical delays, patients will recover sooner, reducing the demand for inpatient beds while lowering costs.

B. Regulators will need to apply different approval standards

The FDA’s current evaluation framework for AI-enabled devices focuses on the specific data sets used to train an algorithm and the consistency of its outputs. That approach works for narrow AI tools (like those that read mammograms or flag suspicious skin lesions) that are trained on specific datasets.

It does not work for generative AI, which is trained on vast, multimodal information sources and personalized inputs. A more appropriate method would evaluate actual clinical performance. For AI-directed robotic surgery, expert surgeons would review anonymized operative recordings — some performed by humans, others by the AI — without knowing which is which. Approval would be granted only when the generative AI consistently matches the quality, safety, and outcomes achieved by expert clinicians.

C. Medical culture will have to evolve

Clinicians have long resisted technologies that threaten professional autonomy, judgment, or income. Autonomous robotic surgery will be no exception.

But rising economic pressure from the growing unaffordability of care, combined with the promise of safer and more consistent outcomes, will ultimately drive adoption.

Patients will hesitate at first. Technologies that take over tasks once performed exclusively by humans always generate concern. When ATMs were first introduced, many Americans worried their deposits might disappear. But as people gained experience and the systems proved reliable, trust grew, and the technology became routine.

Generative-AI-enabled surgical robots will follow a similar trajectory.


Robert Pearl, the author of “ChatGPT, MD,” teaches at both the Stanford University School of Medicine and the Stanford Graduate School of Business. He is a former CEO of The Permanente Medical Group.


Read More

Young Programmer Analyzing Code on Multiple Computer Screens in Office Environment

Automation makes government faster, but citizens need a right to escalation: a clear path to a human with the authority to review and reverse outcomes.

Witthaya Prasongsin/Getty Images

Better Government Requires Knowing When the System Should Stop

A digital public service can work perfectly for thousands of people and still fail the one person standing in front of an exceptional problem. That is the uncomfortable part of government automation. We tend to measure success by speed, completion rates and how many transactions no longer require a counter or a phone call. Those gains matter. But citizens usually discover the limits of a system at a very different moment: when the answer on the screen is wrong for their circumstances and there is no obvious person with the authority to stop, review or correct what happens next.

I have spent years thinking about operations, risk and control rooms, where automation is valuable precisely because time matters. Good systems reduce noise, organize information and help people act faster. Yet high-reliability environments also teach another lesson: speed is not the same as control. A process becomes safer not only because it can move quickly, but because someone knows when to interrupt it. Government should take that lesson seriously as more public decisions are supported, filtered or executed through digital systems.

Keep ReadingShow less
Written on a whiteboard: AI?
The letters AI and a question mark written in black marker on a whiteboard

A Democracy That Uses AI Agents Needs Permission Receipts

An organization can correct a faulty AI-generated paragraph in minutes. Repairing the damage after an automated system impersonates a trusted source, emails thousands of people, or changes public records can take much longer. As AI tools gain permission to act, civic institutions need to govern the authority they delegate with the same seriousness they give to the accuracy of an answer.

On October 8, OpenAI reported disrupting covert influence operations linked to Iran and Russia. One operation used seven fabricated journalist identities to place material in outside publications. OpenAI investigators identified almost 100 articles published or syndicated under those bylines across roughly a dozen outlets. The operators were humans using AI to refine articles and pitches. That distinction matters: this was deceptive human direction amplified by technology, rather than evidence of independent machine intentions.

Keep ReadingShow less
classroom, learning, teaching, class, students

AI literacy isn’t just about jobs. Learn why schools should teach students to question AI outputs, verify claims, and think critically as citizens.

Getty Images, Rafa Fernandez Torres

AI Is Shaping Democracy. Are We Preparing Students?

Almost everyone I know is talking about artificial intelligence. Teacher friends say it helps with administrative tasks, then describe their frustration when students use it to complete assignments. Principals wonder whether teachers are using it effectively. Colleagues past and present worry about how it will shape their students’ lives. And among parents in my kids’ circles, the chatter is mostly about jobs: Do we need AI literacy to stay relevant at work? What does AI mean for our children’s future careers? How much should we trust it?

These questions matter, but I worry we’re missing the plot. AI literacy isn’t only about preparing students to work with AI. It’s about preparing them to live in a democracy increasingly shaped by it.

Keep ReadingShow less
Judge's Gavel Hammer as a Symbol of Law and Order with Processor CPU AI Chip.

Anthropic warns of AI botnets seizing the web. A veteran trial attorney argues the solution is simple: unplug the internet and bring back an analog society.

Getty Images, Alexander Sikov

Let the Bots Have the Internet!

Anthropic CEO Dario Amodei has warned "that a swarm of AI agents soon ‘could be capable of taking over the entire internet with a persistent botnet.’”

Oh, perish the thought! I say, let’s unplug from and discontinue use of the internet. Let the bots roam uselessly on an unused internet and perhaps die a natural death there!

Keep ReadingShow less