Elon Musk recently claimed (see the interview here) that Optimus will be a better surgeon than any human within three to five years, at scale (and it’s pointless going to medical school).

So there we are, medical students can all go home. Surgeons can retrain as baristas, although, given Optimus still appears to be working its way up to making a cup of coffee reliably, perhaps even that plan needs a little validation. I am being deliberately provocative of course and (hopefully) so is Elon.

The difference is that last week I was at the Hamlyn Symposium on Medical Robotics, listening to people who actually understand robotic surgery. Surgeons, engineers, researchers and developers who are not short of ambition, but who also know how difficult and complex achieving robotic surgery is.  Nobody directly addressed Elon’s remarks, but I thought I’d take a few moments to think about them in the context of what I heard last week.

The hard problems are not simply about accuracy

I am guessing the difference here is that we are comparing apples with pears, Elon is describing open surgery which he imagines to be humans (or humanoid robots) standing around a patient where pinpoint accuracy is the ambition.

Of course surgery is not just moving a tool from A to B with impressive precision. Surgeons are not just biological CNC machines in scrubs. They feel, interpret, adapt and judge continuously. They use resistance, texture, tension, vibration, tissue deformation, tiny changes in force and countless other haptic cues that are very difficult to describe, let alone replicate.

A good surgeon knows when tissue feels wrong.  A robot does not, unless we have given it the sensing, interpretation and control systems to understand what ‘wrong’ means in that moment, in that patient’s messy bit of anatomy.

Even complex surgical systems such as the da Vinci 5 are still only beginning to address haptic feedback properly. That is not a criticism. It is an acknowledgement that this is difficult, very difficult.

Do we need humanoid robots in the OR?

There is also a more basic assumption hiding inside Elon’s comment: that the future of surgery still looks like open surgery, with a humanoid standing over a patient using human shaped hands.

Chris Fryer, CTO of Cambridge Medical Robotics, pointed out that open surgery still accounts for around 50% of procedures, but the direction of travel is towards minimally invasive surgery. For MIS, a humanoid robot makes very little sense. The whole point is that you are not standing over a large open incision with two hands in the body. You are working through ports, with instruments, imaging, navigation and carefully controlled access.

A humanoid form factor solves yesterday’s mental model of surgery, not necessarily tomorrow’s clinical workflow. Even in open surgery, why would we limit a robot surgeon to two arms?

Indeed why would we try to build a machine that copies the constraints of the human body in an environment where this is not necessary? If we are designing from first principles, why stop at two arms, two eyes and one tired pair of shoulders? Surgical robots are already specialist systems: multiple arms, endoscopes, instruments, imaging, navigation, working alongside the clinical teams.

The robot is part of a system of care, it is not a lone heroic machine. Humanoid robots are designed to be generalists, to fit where a human can and use human tools, great if you want to perform many different tasks, but surgical robots are very specific tools, honed for a tightly controlled environment.

Robots are stupid!

One comment from Hamlyn that stuck with me came from Professor Ferdinando Rodriguez y Baena (Professor of Medical Robotics and Engineering Codirector of the Hamlyn Centre for Robotic Surgery): “Medical robots are the stupidest robots we have”, because we need them to pass FDA regulations.

That line is funny but it’s true, medical robots are constrained for a reason. They operate in environments where failure is not just inconvenient, it can be catastrophic. The regulatory burden is not an irritating admin layer added at the end, it is part of what makes the device acceptable for use on actual humans.

Even if Optimus could perform surgery today, which it very much cannot, it would not be on the market at scale any time soon. Regulatory approval alone can take years, building clinical evidence takes time, human factors, manufacturing validation it all takes time.

Robots vs Regulations

This is before we even get into the delightful question of software upgrades on a physical AI surgeon.

How do you approve a surgical robot that keeps learning? How do you validate a major AI update? Who is responsible when performance changes after deployment? What counts as the same device, and what counts as a new one?

AI in medical devices already bumps into this problem in the realms of diagnostic tools, imagine the hurdle of proving the safety of a Class III robot working autonomously using a vast pool of evolving training data.

Navigation is the easier part (and we haven’t solved that yet)

Professor Brad Nelson (Professor of Robotics and Intelligent Systems at ETH Zurich), made another point that needs to be called out: “Manipulation is harder than navigation.”

We have made real progress in AI navigation, self-driving systems can now operate autonomously in some limited environments (although how autonomously is certainly up for debate!). Driver assistance is however common. This shows that technologies once considered impossible can become normal.  But driving around San Francisco in a Waymo is not the same as performing brain surgery.

A road is complicated, but it is designed for vehicles. The human body was not designed for robots, anatomy varies, pathology evolves and tissue moves, bleeds, tears, swells and reacts. The plan changes and the unexpected is not an edge case in surgery. As Professor Nassir Navab (Technical University of Munich) pointed out “Our AIs have no idea about human physiology… 90% is not what you see”

I do accept, in principle, that a future AI robotic system could learn from more cases than any human surgeon. It could absorb patterns across anatomy, physiology, pathology, imaging and outcomes at a scale no individual person could match. That is not ridiculous, what is ridiculous is pretending we are nearly there because a humanoid robot can wave, fold a shirt or avoid falling over on stage.

Maybe a robot can lend a helping hand?

Professor Philip Chiu (Dean, Faculty of Medicine, The Chinese University of Hong Kong) made what felt like a much more plausible near-term point: perhaps a humanoid robot could act as an assistant. It might hand over tools, it might help with retraction or do controlled, bounded tasks within a well-defined workflow. That is interesting and potentially useful but is a long way from replacing the surgeon.

Are surgeons ready to be replaced by physical AI?

The surgeons I spoke to cannot see themselves being replaced any time soon and I agree with them. Although I also think we should be careful not to become luddites. It is very easy for experienced professionals to underestimate change, especially when that change will affect the next generation more than their own.

In one of the panel sessions at the symposium the consensus was that surgeons would not accept AI assistance (we are talking about simple automated steps like suturing), because they have the hundreds of hours of practice to make this easy for them. Here I disagree. A surgeon in their fifties may find the idea of AI assistance absurd, but their grandchildren may not. Let’s not forget that only 20 years ago the idea of driving assists were widely mocked, but they are pretty much standard these days.

But the gulf between today’s surgical robotics and “Optimus is better than any surgeon at scale” sits a mountain of dexterity, sensing, haptics, clinical judgement, team integration, regulation, economics, liability and trust.

Okay, but what about the ambition?

Elon’s version is a soundbite, possibly a stock price soundbite. But I don’t think the ambition of replacing surgeons with physical AI is unachievable, it’s just far further away than the tech billionaires would have us believe.  Should we do this or is it inevitable? Whether we want to go down this road is of course a different discussion. Where I do take issue with the learned surgeons is that people would rather speak to real people and that a human can show more empathy than an AI. In theory this should be true, but in reality (and I have had a lot of personal experience with this in recent years), I do not believe an AI (with all the training data in the world) could not fake empathy as well as a surgeon can, and on a bad day far better.  AI LLMs already do this very well and let’s be honest a doctor at the end of a brutal shift with a million other priorities and distractions is not always going to be at their best. Ais are already providing valuable diagnostic assistance, they can provide support, advice, maybe a steading hand.

The inevitable march of AI

At Hamlyn there were several demonstrations of AI automation of suturing tasks, clearly AI assistance is close at hand. So, let’s be honest, there is a place for AI in a surgical environment and while some of us might instinctively fight against this, the coming generations will see no issues here at all. Twenty years ago, many people could not imagine accepting a car that drove itself. Now driver assistance is normal, and younger people may well be more comfortable with surgical automation than today’s senior surgeons. However, if Tesla are not yet at the point of offering us a fully automated car that can drive anywhere in the world, then chances are they are a lot further away from providing a surgeon who is ready for any eventuality in the OR. So no, I would not cancel medical school just yet.

But while an OR staffed entirely by physical AI maybe be a long way off, automation of surgical processes maybe not that far away at all and there may even be a place for humanoid AIs in our hospitals sooner than we might think. Maybe new clinicians should start learning how to work with robots, because this is probably going to have a big impact on their careers.

I leave the final word to Professor Chiu, “AI plus human is better than human”