The SRI-led Trauma Pod, developed for DARPA as a next-generation mobile robotic surgery platform for the military. Credit: SRI International via Wikimedia Commons.

This week, the “144” study on adverse events in robotic surgery made the rounds in media outlets such as the BBC, NBC and the MIT Tech review. An impressive group of researchers found that between January 2000 and December 2013, we have records of robotic surgical procedures killing 144 people, injuring 1,391, and involving 8,061 malfunctions – this is out of a total of more than 1.7 million procedures performed in that time period.

This is of course serious news, raises important questions, and adds to a heated conversation on the safety and effectiveness of robotic surgical procedures. A growing body of studies now argue these points (for example [1, 2, 3, 4]), and thankfully we’re starting to learn when and why this technology works or causes trouble. A lot seems to come down to how frequently a procedure is performed at a particular hospital and by a particular surgeon, for example. Hopefully this dialogue will mean safer and more effective robotic procedures as the years go on; this is generally the story with surgical technologies as they mature.

The intense focus on safety and effectiveness seems to be compromising training for new robotic surgeons.

It may also mean many fewer competent robotic surgeons, just as demand for robotic surgery soars. After two years of field research – observing close to a thousand hours of robotic and open surgery at five diverse hospitals and interviewing surgeons and residents from ten teaching hospitals around the country – I have found that the intense focus on safety and effectiveness seems to be compromising training for new robotic surgeons.  I should say from the outset that I’ve collected data only on surgery involving Intuitive’s da Vinci system, and primarily on urologic procedures. I did this to uncover best practice: the da Vinci is used far more frequently than any other robotic surgical system, and the field of urology has had – by far – the most experience with the system over the longest period of time.

Why is surgical learning compromised in surgery involving the da Vinci? Mainly because the technology allows a single surgeon to perform the bulk of each procedure. In traditional “open” surgery, senior surgeons need significant help from medical residents throughout. While I explore interesting exceptions in my research, at my sites this generally translated into a ten- to twenty-fold reduction in the time that medical residents got to spend actually doing surgery. It’s hard to hand the controls over to a rookie for long when you’re accountable for safety and quality in life-and-death work. Perhaps in response, Intuitive introduced a “driver’s ed” model that allows senior surgeons to digitally delegate surgical control to a trainee sitting in another control console. Interestingly, the presence of this model did not lead to notable increases in resident “console time” at my sites. Most often, residents watch procedures unfold on TV screens in the operating room, which likely has limited learning. As a chief of urology said to me, “Watching movies does not make you an actor.”

Most often, residents watch robotic surgical procedures unfold on TV screens in the operating room, which likely has limited learning.

This challenge adds fuel to a fire that was already burning: according to the AAMC, surgical demand is growing significantly faster than the supply of surgeons. The significant reduction in medical residents’ work weeks (from 120 hours to 80, mandated by the AAMC) has also decreased the number of surgeries that remaining medical residents get to perform. So “best practice” with the da Vinci has – on one key dimension – made it even harder for many students to get sufficient, safe, legitimate opportunities to learn how to use it.

So what are we to do about this problem? It looks as if every way forward involves challenging deeply held norms, laws and techniques related to surgical work. Change will likely require effort on multiple fronts simultaneously. It’s tempting to suggest that senior robotic surgeons should allow medical residents ten times the opportunity to struggle on their own during surgery… until you watch an expert remove potentially cancerous lymph nodes in ten minutes and on a different day watch a resident struggle to do the same in forty. Some surgeons allow this, but because of the legal, professional and quality assurance issues described above, many expert robotic surgeons won’t.

Robotic surgery will likely continue to expand in proportion to other methods, given that it allows fewer surgeons to perform surgery with less trauma to the patient.

One interesting option is to do nothing, because surgery itself is on the wane. A good number of the senior surgeons I speak with presume that new treatments – nanoparticles, nanobots and the like – will make cutting into patients less and less necessary. But many physicians and researchers expect that surgery isn’t going away for at least several decades, and that if anything, robotic surgery is going to continue to expand in proportion to other methods, given that it allows fewer surgeons to perform surgery with less trauma to the patient.

We have recently seen striking progress on automating surgical tasks in lab settings. When I ask experienced robotic surgeons about the future of their work, many of them suggest that relatively routine portions of a given procedure might soon be automated. The system might move more slowly than they would, and the surgeon might need to orient the system to the landscape inside the patient, but a semi-autonomous system would allow a single surgeon to perform several procedures simultaneously. But while this solution could address the demand side of the equation, it’s not obvious how it would help surgeons in training.

Trainees in surgery are still expected to build embodied competence through “live” work.

Simulation has already played an important role in resident education – Intuitive has partnered with a number of organizations to produce relatively advanced software simulations that help build basic surgical competence. The trouble here is that – unlike most highly skilled, collective work such as sports, music or law enforcement – trainees in surgery are still expected to build embodied competence through “live” work. In order to get time on the simulator, residents at most institutions have to get less sleep or time with family – in other words, surgical education doesn’t build in sufficient time for offline practice. This changed in the field of aviation so it seems possible in the field of surgery, despite a strong norm for learning in live procedures.

A final option is telesurgery: if surgeons could operate on patients remotely, we could more efficiently deploy surgical expertise across geographically distributed demand. The original DARPA grant that contributed to the development of the da Vinci specified remote surgery on wounded soldiers as a key use case. Legal, regulatory and team-dynamics barriers loom large here. As of now, law and regulation make it difficult (often impossible) for physicians to do remote diagnosis and recommend treatment – let alone operate on a patient – across state lines in the US. Also, we have decades of research showing that geographically distributed teams underperform relative to their collocated counterparts. We should be wary of a surgeon in California trying to coordinate a surgery with a nurse and a surgical scrub at the bedside in Atlanta.

No matter how safe we make robotic surgical procedures, they will become a luxury available to a very few if we fail to address the sustainability of the practice. Some hybrid of the options above on a large scale will likely be required to address the constraints that “best practice” with these systems places on surgical training. Bringing this lens to the next wave of development in surgical technology – such as Google’s joint venture on robotic surgery with Johnson & Johnson – shouldn’t take much additional effort, and could extend the benefits of minimally invasive surgery to many more at decreased cost.

References

[1] Agarwal, Piyush K., Jesse Sammon, Akshay Bhandari, Ali Dabaja, Mireya Diaz, Stacey Dusik-Fenton, Ramgopal Satyanarayana, Andrea Simone, Quoc-Dien Trinh, and Brad Baize. 2011. “Safety Profile of Robot-Assisted Radical Prostatectomy: A Standardized Report of Complications in 3317 Patients.” European Urology 59 (5): 684–98.

[2] Chang, Steven L., Adam S. Kibel, James D. Brooks, and Benjamin I. Chung. 2014. “The Impact of Robotic Surgery on the Surgical Management of Prostate Cancer in the USA.” BJU International, September, n/a – n/a. doi:10.1111/bju.12850.

[3] Sammon, Jesse D., Firas Abdollah, Dane E. Klett, Daniel Pucheril, Akshay Sood, Quoc-Dien Trinh, and Mani Menon. 2015. “The Diminishing Returns of Robotic Diffusion: Complications Following Robot-Assisted Radical Prostatectomy.” BJU International, March, n/a – n/a. doi:10.1111/bju.13111.

[4] Trinh, Quoc-Dien, Jesse Sammon, Maxine Sun, Praful Ravi, Khurshid R Ghani, Marco Bianchi, Wooju Jeong, et al. 2012. “Perioperative Outcomes of Robot-Assisted Radical Prostatectomy Compared with Open Radical Prostatectomy: Results from the Nationwide Inpatient Sample.” European Urology 61 (4): 679–85. doi:10.1016/j.eururo.2011.12.027.

This post is part of Robohub’s Jobs Focus.

Recently, many of these explanations, predictions and proposed solutions focus on robots. Robots seem to have found their way into an increasing number of titles, photos and articles about our economic future, whether from economists like Paul Krugman, journalists like John Markoff, roboticists like Rod Brooks or futurists like Ray Kurzweil.

Disagreement about robots and our future is flying fast and furious. Are we on the edge of a robotics revolution, analogous to the PC revolution of the 70s? If so, will we see similar patterns of economic activity – automation of a great deal of work, death of some occupations, birth of others? Are current-day robots a harbinger of times to come, just as canaries were for coal miners until the early 20^th century? Or are we experiencing a “robot revolution” in name only? Are cutting-edge robots like Baxter a red herring – do they misdirect our hopes and fears about our future?

If you’re looking for the punch line of this piece, here it is: we have evidence that robots are both red herrings and canaries in a coalmine. We are arguing about this issue as if it’s either/or when both views can be usefully seen as true.

Robots as Red Herrings
Imagine a robot that costs almost nothing to produce (as in less than a cent), costs almost nothing to ship, that runs on almost no power, that could be customized for a wide variety of applications and that could execute billions of tasks every second. Obviously this robot would be software. We’d be excused for assuming it would sell pretty well, and for assuming software might play some role in the economy.

Because it already has. According to the BES, there was an installed base of 1.85 trillion dollars of software and related hardware in the US in 2011, which has grown an average of about 3.3%/year since 2004. Though there is ongoing disagreement about the role of IT in our economy, recent analysis lends strong support for the view that IT is mostly being used as a substitute for human labor on routine, non-physical tasks. Many economists point to IT as a key contributor to the “hollowing out” of the global economy – otherwise known as the “death of the middle class”. As Erik Brynjolfsson of MIT is fond of saying, it’s hard for a tax preparer to compete with TurboTax – especially when this software is actually free these days to new users, and integrates seamlessly with Mint.com, a free, online tool for personal finance.

From this angle, software has had a truly massive impact on the kinds of tasks humans can do and get paid for. On a global scale, we’re talking hundreds of millions of jobs and livelihoods, broad changes to the structure of work and occupations, and dramatic shifts in the way we encounter each other and the working world.

Yet when we talk about change associated with technology in the economy these days, we talk mostly about robots.

This seems a little strange, given that the IFR (a global authority on the robotics industry) estimates that there is an installed base of just 9 billion dollars of robotic systems in US and Canadian industry put together. That’s less than half a percent of the installed IT base. While recent growth numbers are large (38% in 2011), the IFR projects the global annual growth rate at about 6% through 2015. At quadruple this rate, it would take robotics 23 years to catch up with IT, assuming IT stood still. So if we want to investigate the technologically-related automation of jobs, work and the economy, maybe we should be focusing more (maybe a lot more) on Watson, Google and Robo-Traders on Wall Street. Maybe robotics is a red herring.

Robots as Canaries
But then again, maybe not. Many roboticists, economists, and journalists are claiming that we are on the cusp of a period of discontinuous progress in robotics technology. If this is true, then maybe we’ll soon see exponential – not linear – growth rates in robotics. Then again, maybe we already have.

How could we tell? One promising first step is to judge robotics based on performance per unit of time. While Moore’s law predicts the number of transistors on a chip as a function of the passage of time, its progress is almost perfectly correlated with performance-oriented metrics in other domains, such as processing speed, memory capacity and sensor quality. From this perspective, it is reasonable to assume that robots’ performance of basic tasks would correlate well with progress in underlying technical domains. If it hasn’t, then a revolution seems less likely.

Given that we all seem concerned with robots that may replace non-routine physical labor, Rod Brooks has suggested in a few talks that we should look at performance in autonomous robotic navigation as an indicator of progress. This does seem like a reasonable proxy: autonomous navigation requires a system to sense its environment, plan next steps and enact those steps through actuated movement, more or less without human input. The greater the performance of such a system, the more sophisticated each of the component technologies (sensors, AI and actuation). These requirements and related technological progress seem likely to generalize to other physical tasks in a dynamic environment.

So where do we stand on this front? Hans Moravec’s (Stanford) lab robot could navigate 20 meters in six hours in 1979. Ian Horswell’s (MIT) robot could navigate 2000 meters in six hours in 1992. The best car in the second (2005) Darpa Grand Challenge navigated 212 kilometers in 6.52 hours (about 194,500 meters in six hours). The winning car in the 2007 Darpa Urban Challenge navigated 96 kilometers of dynamic urban environment in 4 hours and 10 minutes (about 23,000 meters in six hours). By 2010, the Google Car was capable of autonomous navigation for six (or more) hours straight in normal traffic. If we presume it obeyed the speed limit and did this driving on a highway (at 65 mph), the car could travel 390 miles during that time (or roughly 627,000 meters). Here is a plot of these achievements (on a logarithmic scale) over time:

If anything, this approach may underestimate the rate of change; the task has gotten more difficult. The 2005 Darpa Grand Challenge cars would have made much better time if they had to traverse 212km in a straight line, indoors, with no obstacles. Incidentally, the cost of just the camera on Moravec’s 1979 robot was $159,000 (in 2012 dollars); a far superior camera can be purchased today for less than $1.

Put together, this indicates the cost of robotic systems is dropping rapidly just as performance is increasing exponentially. While it is all the rage to predict the future based on exponential technological trends, this “back of the envelope” exploration should put us on notice: stories of linear progress in robotics should be justified, not taken for granted.

Robots as Red Herrings and Canaries in a Coal Mine
Work is scarce and wealth is increasingly in the hands of a few. This leads many of us to be scared, angry and curious about the future. Robots are a lightning rod for these emotions and questions – they move with apparent intention, they do tasks that sit at the core of what most of us define as “good, honest work”, and they’re rapidly getting cheaper and more capable. If we want to point a finger at technologies that are “responsible” for the inequities and challenges in our economy, however, it seems pretty clear that we should be pointing to IT. The vast majority of tasks that are being handed off to technology are still going to software and computers, and the evidence indicates that this will be true in the short and medium term. Robots are a red herring from this perspective.

But from another perspective, they are also canaries. We have seen apparently exponential progress in robotic capacity over the last 30 years. How many of us think that, in a hundred years, we won’t have highly autonomous, mobile, dexterous robots that can adapt to a wide range of tasks in a wide range of environments – for less than the cost of human labor? Are we willing to bet our national education and job training investments on the assumption that this is a hundred years away? Fifty years? Thirty? If this transition is going to happen on a mass scale, at best we probably have seventy years to prepare for it. That’s only two generations away, and that means we should be planning for it now.

See all the posts in Robohub’s Jobs Focus.