This is a wide-ranging interview with Craig Smith, MD, Chairman of the Department of Surgery and Surgeon-in-Chief at Columbia, about the many technological, professional, and cultural changes that are happening in surgery today and in the future.
The Future, Today
What is the single most futuristic thing we’re currently doing in surgery?
It depends on how you define “doing” because I would say tissue engineering and its applications, except that most of them are not quite in the doing stage. They're being tried here and there. It also depends somewhat on how you define tissue engineering. It's out there in heart valves, for example, in congenital and transplant. The stuff going on in labs is very exciting and shows the promise of substituting engineered tissues for the stuff we currently use that are much more successful long term. But it's still so futuristic, most of it’s not quite there yet.
The robotics applications are farther along. In the last couple of years, robotics has really started to take off in ways that it hadn't for 20 years. And it's not so much that we're doing different things with it, but they're finally able to do both standard and more advanced things that were out of the reach of robotics before, and it's likely to open up an acceleration in the underlying technology. We might finally get away from the era when only one company was surviving on the amount of robotics being done, which had a constraining effect on innovation of the platform. As it gets more successful, into more hands, and applied to more cases, the competitive pressures that draw in other companies and innovators will probably start to bear fruit.
I wish I could say thatnAI was that thing, but it would be a little bit like tissue engineering, but even more so. It's sort of teetering on the brink of application but not quite in application.
Tissue Engineering: Dream or Reality?
Do you have any predictions about the wider application of engineered tissues?
Well, on the smaller scale, where it's probably a little bit easier to envision, it could be very transformative for skin replacements and valve replacements. Small units, simple structures. But what they're doing in some labs is building organs. And if they can do that, it could leapfrog xenotransplant by creating organs that aren't even immunogenic [meaning they would not trigger an immune response] in the conventional sense. At least in theory, it should be easier to avoid immunogenicity in lab-built organs than it is to genetically engineer immunogenicity out of pigs.
You would be starting with the person's own cells, right?
Yes, potentially, or on a neutral scaffold that can be populated with the patient's cells. That's probably the most popular approach right now. Some labs are trying to create lungs that way. We may have to start from more simple and unitary replacements like skin, then move on to organs by building on that experience. However, manufacturing organs, like the genome and precision medicine, have been dreams for a long time, and that horizon might keep receding in front of us.
Artificial Intelligence Has a Way to Go
Have there been any applications of AI in surgery, or medicine at large, that seem particularly promising to you?
Well, specifically in surgery, no. Although, as it is with medicine in general, people see the potential. I've ended up following AI pretty closely because it's the question everybody asks. My guess is AI is more likely to have major impact on the so-called cognitive specialties long before it impacts the non-cognitive spinal-reflex specialties like surgery.
In cognitive areas like radiology, the data is digital to begin with; it begins as zeros and ones. You could say the systems that generate the information now are doing humans a favor by converting the data into analog pictures that we can understand. The underlying data is purely digital. So, to have an AI manipulate something that was always digital seems like an easier leap than converting analog to digital and manipulating that.
Many kinds of diagnostics in medicine rely on following an algorithm—rule this out, then rule that out; if A, then B, then if it’s not B-prime, it’s C, etc. Humans get very good at that from practice, but an AI will probably be able to do it much faster and more compulsively while eliminating more of the uncertainty.
Will an AI ever have the intuitive gifts of a great internist? It may not need to if it can plow through the algorithm at light speed. The AI might be able to skip the intuition step. Of course, my assumption that AI will have major impact soon in those kinds of applications requires setting aside the increasing concerns about its fidelity and reliability, and that's a whole fascinating area of uncertainty.
The ethical questions about AI are fascinating. Have you seen any progress stemming from those discussions yet?
I attended a seminar at the downtown campus on the pitfalls of AI. The morning was all science, the whole afternoon was on ethics, sociology, and those kinds of issues. I thought the afternoon would be the most interesting, but that turned out to be a bit of a snooze. It was a bunch of typical pointy-headed intellectuals talking about lofty stuff, but the science in the morning was quite interesting. That’s how I got hooked on this blog from a Professor at Princeton named Arvind Narayanan. In the seminar Narayanan was very good at explaining these things at a human level. The ethics are an interesting part of it, but they're not as directly connected to the accuracy of AI, which I think may be a bigger hurdle up front.
When he spoke at the seminar, maybe six months ago, the error rate that seemed baked in was something like 12 to 15 percent. That’s the rate at which AI engines are still confabulating, making stuff up.
Wow.
About a year ago, I was playing around with ChatGPT in preparation for a lecture I give every year. I asked it some general questions, and it gave me these beautiful text answers that sound a bit like an AI but are quite reasonable in quality.
And then I asked it to do what amounts to a reverse compound interest problem about salary—pretty simple math. I've done it myself, but it's a bunch of steps. I put the problem in the prompt and got an answer, but it just didn't smell right. So, I put the same prompt in two more times and got three different answers. Only one of them was right.
The wrong answers were way off. One used the wrong formula; one had errors in calculation—really bizarre. Math is turning out to be a significant weakness for AI when most people assume it would be exactly the opposite.
I would never have expected that. Are there speculations as to why?
The only explanation I’ve heard points to the training sets used by large language models (LLM), like ChatGPT. The LLMs do seem to be pretty successful at integrating all the vast inputs from Google that are a mixture of text, non-text images, data, whatever, and cooking it in some way that they can transfer into language.
I don't know whether it's a simpler problem or a more complex problem, but calculation doesn’t seem to benefit from the training set in the same way, maybe because it doesn't have to analyze a range of variabilities. It just has to know what to do and do it right. Again, this is my simple-minded interpretation of it, but there is some quirk in the way they're currently trained and managed that makes math a problem. And if that’s a problem, you might think, why wouldn't that worry us about every sub-element in AI?
Right, absolutely.
It makes me wonder if that difficulty with math explains some of the confabulation that goes on. The confabulation is fascinating because it goes off on these wild tangents and makes up references and everything, and it’s not easy to make up a credible reference.
How would it handle something like a complex proof?
That's a great question because I’ve heard more than one expert claim that the large language models will be very important for doing proofs in the future. That presents a conundrum—if an LLM can carry out proofs that are beyond human abilities, and we can't understand how the LLM does the proving, how can we accept the result? If AI reaches a point where it’s doing things we can't understand, how do we accept any of it?
One of the things Narayanan talks about is that the large language model variety of AI may be about to run out of training sets. First, because there is only so much data out there in the world– Google, libraries, all that. And they have already stripped those clean, so where do they go next? And then on top of that, there's increasing resistance to just letting companies developing AI grab the data and run. People are locking it up or making them pay. So, both those things are putting the brakes on training sets. That raises the question, why can't AI train on AI?
Right, is a training set reliable if it’s AI-generated?
Exactly. How will we know when it’s just reiterating a confabulation? If you go back to surgery and AI, my personal belief is that AI may be transformative in much of medicine, but it won't touch surgery until fairly late in the game. It will probably have a helpful and significant impact on elements of surgery like diagnostics, setting up structural preconceptions, 3D modeling, that kind of thing. It might facilitate big changes in some of the elements of an operation, parts that can be sort of roboticized and made more mechanically reproducible.
But surgery per se is so heterogeneous and hands-on—it’s hard to think of better example of something that’s hands-on. I think it may be a while before we have AI-programmed autonomous robots that you can wind up and put in the belly, and they go around and fix things.
The Humanity of Robotic Surgery
Even robotic surgery today requires the experience, skill, and even the artistry of the surgeon’s hands that are operating it. The robot is useless without the person behind it.
The robot is not a robot. It's a misnomer that’s been embraced because it gets everybody's attention, but what we call surgical robotics is what engineers call a controller-responder device.
The controller is where the machine turns the operator’s completely analog finger and wrist movements into digital. That data is transmitted to the responder, which turns it back into analog movements at the working end of the instrument. It’s all derived from and controlled by a completely analog human surgeon. There's nothing robotic about it because it’s not just loose to go do things. But “robot” does have more cache than “controller-responder.”
When I talk to robotic surgeons like Dr. Jason Hawksworth, he mentions that the robot's arms also still have some movement limitations. Surgeons develop ways around these rigid arm movements. How do you see functionality improving in the future?
That specific part of it has been the major limitation since the beginning, back when I was involved. And yes, the limitations are significant because that stick that goes through the chest wall presents an engineering problem. It can only have so many ranges of motion, so many degrees of freedom.
The most stunning thing about the robot is the stereoscopic camera, which gives true 3D vision, unlike anything you do in laparoscopy. The ability to get the camera right on top of what you're doing is better than the open view in many cases.
That visualization piece is great, but even if the view is perfect, you may not be able to get the instruments to rotate in just the right way or grasp delicate tissues gently enough. Twenty years of innovation have developed better instruments and smaller instruments with more degrees of freedom, making it possible to do things in smaller patients and more complicated structures.
Redefining the Practice of Surgery
How have these massive innovations in surgery, and technological applications across medicine impacted how you approach your practice?
Well, there are two ways I can answer that. In what I do as a cardiac surgeon, I think the innovation continues the way innovation usually has in medicine and life in general—through this almost undetectable process of incremental improvements. You look back 20 years later and can't recognize where you are because every year, there's some new element done better with something different. It's a thing I've called stealth innovation that has driven most of the improvements. The giant leaps are rare. TAVR was a giant leap and not beneficial to surgery in a sense, but those are the kinds of giant leaps that come along now and then and transform things.
But it's constantly improving, and I'm sure the same is true of other specialties that aren't as technically complex. They have a slower pace, maybe, but it's still always going on. Of course, I’m very biased, but I think surgeons tend to be pretty inventive people. They're always thinking of new angles and keep developing these new subspecialties. In this department, a great example is Yuri Novitsky’s division focused on complex abdominal wall hernias.
Yes, absolutely. Dr. Novitsky is the Director of the Comprehensive Hernia Center.
I didn't even know that specialty existed until maybe ten years ago. He and a few others recognized that hernias were much more than just a simple operation done by all general surgeons. They came up with different ways to approach them and different materials to use. That gets back to tissue engineering. They use all kinds of fancy stuff to close the defects.
He and, I believe, Dina Podolsky figured out that they could use the robot to get the instruments inside, turn the camera up, and be looking at the inside of the abdominal wall rather than the outside. There are things that you could do from the inside that you can't do any other way. It’s a whole new field. And his group has gone from zero to 60 in a very short time. Most of the time our innovations are going from 40 to 60. We already have three surgeons, and soon to be four, devoting their careers to complex hernia repair.
That’s pretty amazing.
Acute care surgery (ACS), too. The acute care surgery group didn't exist when I took over the department. People who knew about that kind of thing educated me about it, and we recruited what is now like 8 or 10 people. They do emergency and urgent surgery, trauma, and ICU care, which all fit together in a fairly neat package.
Basically, ACS more expeditiously cleans up the stuff that used to be lying around waiting to be done after hours, like an inflamed gallbladder case. ACS is not as much of a light bulb moment as the abdominal wall group, but it's a new way to reorganize and improve the work we do. It’s been very beneficial, very popular, and meshes very well with one of the other cultural changes going on in medicine, the change in approach to work-life balance. But it’s surgeons who keep figuring this stuff out.
Culture Change as Innovation
Is the striving toward a work-life balance part of how the specialty of Acute Care Surgery was born?
I'm not sure the people who came up with the concept would argue that was why. I think it was probably driven more by the logic of the combination of niches than the fact that it would then become an attractive shift work option within surgery.
Most of surgery is not shift work, which gets to one of the other futuristic things going on in medicine, at least in my opinion: there are cultural headwinds aligned against what we traditionally considered to be professionalism, and the drive to shift work is one source of those headwinds. It's a change you see in medical school, in every step of residency, and then in the way specialties configure themselves.
One problem with pure shift work is that it interferes with continuity of care, which at least traditionally has been considered important in surgery and in other specialties. But specialties like anesthesia and ER have always been predominantly shift work, and they have been able to count on their native motivation and professional passion to make them productive. This issue is fresh in my mind also because all these pressures have been exacerbated here recently through a forced conversion to an algorithm-driven production-based compensation plan. This is a cultural innovation experiment that is being played out in real-time, and I don’t know how it’s going to end up.
That tension is certainly bubbling up everywhere, across industries, as the concept of work itself is changing rapidly.
For sure, and I hate to call it generational. Maybe my generation deserves a lot of the blame. There's a lot of blood on boomer’s hands for the things we believed and things we did in the sixties and seventies. But here we are, and in one gloomy worldview, medicine becomes a branch of social work or the ministry or something.
If a change like that happens, people will be less and less well-paid and differently motivated. Their motivations may be great and admirable, just like social workers and ministers, but different, and it's going to attract a very different cross-section of people. It’s also hard to know what that will do to innovation.
Well, on that note, let’s end this interview with a new one: what is your gloomiest prediction for the future of surgery?
As I just suggested, my gloomiest prediction is that medicine will be a branch of social work 25 years from now. I do think surgeons are still different, and there will always be a need for surgery. Humans are imperfect. Sometimes they're born imperfect, and sometimes they develop imperfections because of their environment, the things they eat, what they do, accidents, war, whatever. They're always being made imperfect in one way or another, and some of the solutions to those imperfections will always be surgical.
There may be less need for a certain kind of surgeon 50 years from now, but there will always be a need for somebody to fix these broken humans in ways that can’t be done by sitting on the couch with the psychiatrist or by taking pills and things like that. There will always be the need. So, I'd like to think that there's a kind of person who will be attracted to filling that need who is a little different than the person who's going to be a social worker. Surgery and social work are both noble service professions, but it's a different kind of service.
I'm not sure I'd want to be quoted as predicting that medical schools will be branches of social work or divinity school, but it's a powerful image.
Related:
- An Interview with Dr. Craig Smith, Heart Surgeon, Chair of Surgery, and Author of Nobility in Small Things
- Dr AI: Can ChatGPT Step Up To the Medical Challenge?
- State of the Union: Heart Care Today
- Overcoming Hernia after C-Section and Regaining Life as a New Mom