Surgery and the limits of the robotic ideal

In the 1940s and 1950s, Isaac Asimov, ‘the father of science fiction’, wrote I, Robot, a collection of short stories initially published in American science-fiction magazines. The 2004 I, Robot film, which portrays a frightening dystopian rise of robotic technology and overthrow of humanity, is a far cry from Asimov’s original vision for robots’ development.

Asimov was a professor of biochemistry during an era of pioneering discoveries, including Franklin, Wilkins, Watson and Crick’s elucidation of the structure of DNA (1953). He was a technological optimist. In his contemporary context of wartime industrialisation, early computing and concerns surrounding automation, he imagined the genesis of robots within the framework of his ‘Three Laws of Robotics’. The first law is at the centre of these: ‘a robot may not injure a human being or, through inaction, allow a human being to come to harm’; the laws as a complete set represent ethical safeguards to ensure robots serve humanity safely.

Through the I, Robot series and later works such as The Caves of Steel and The Naked Sun, Asimov frequently depicts robots as rational, ethical and reliable agents for good, that work to stabilise and benefit human society. In his speculative fiction, Asimov explored many of the themes confronting contemporary society with the expansion of artificial intelligence. The international community has moved to establish safeguards for AI, which eerily mirror aspects of Asimov’s Three Laws. The Council of Europe Framework Convention on Artificial Intelligence and Human Rights, Democracy and the Rule of Law (with 21 signatory states) insists on a human-centric approach, mandating that AI activities respect human dignity and individual autonomy.

The human workforce first started to be displaced en masse by machines during the Industrial Revolution. In the 21st century, this historical development has entered a new phase, with the advent of robotic sorting and packaging plants, robotic waiters, driverless cars and AI personal assistants. The adoption of these technologies is pervasive and quasi-universal. Perhaps Asimov would have marvelled that today, robotic technology has developed so quickly and is so accessible that autonomous vacuum cleaners and lawnmowers are relatively banal aspects of reality.

Clinical medicine is perhaps the most literally ‘human’ profession. Surgeons aim first to understand, and then manipulate human biology. Will medicine and surgery be the final frontier for full robotic automation?

Robotic surgery, or more accurately robot-assisted surgery, has been used for around 20 years and is routinely practised worldwide, including in the UK National Health Service (NHS). In the 1990s, robotics scientists developed machines with the ability to translate continuous input from an operator into movement in real-time. This opened the door for the development of surgical robots, which replicate with high fidelity the movements of a surgeon’s hand during operations. The US-developed da Vinci surgical robotic system is one of the pioneers in the field (FDA-approved in 2000), although equivalent Chinese and EU systems have been released to market. These robots are not autonomous, and the operating surgeon remains in full control of the three or four robotic arms, which they manipulate through a console.

Surgical robots’ role in routine clinical practice remains restricted to very specific operations. Throughout the latter half of the 20th century, all surgical specialities have strived to make operating as minimally invasive as possible, reducing risk of surgery and shortening post-operative recovery times. Cue video-assisted ‘keyhole’ operating. This has been a game changer and, in many cases, has superseded the original ‘open’ techniques. Many specialities have embraced keyhole techniques, virtually a revolution in clinical practice, partly because of patient preference, partly because of faster recovery. Internal camera video ‘stacks’ are commonplace in the theatres of abdominal, thoracic and gynaecological surgeons; cardiologists and vascular surgeons similarly deploy devices such as stents inside the heart and blood vessels through incisions barely a centimetre across under X-ray guidance.

The first steps of a traditional video-assisted operation are the same as for its robot-assisted cousin. The surgeon makes a small cut to introduce a camera through the skin; further keyhole cuts are made to insert specialised instruments, which are long and thin to allow access to the surgical target, and are hand-actioned, often with loop handles, like a pair of paper scissors. This method is routinely used to perform operations on a daily basis, such as appendicectomies, cholecystectomies (gall-bladder removal), hernia repairs and hysterectomies. In a robot-assisted operation, the instruments are introduced in the same way but are docked into a robotic arm rather than being hand-held. The surgeon at the console controls the movements of the robotic arms and thus moves, opens and closes instruments.

Despite significant advantages in what are minimally invasive operations in relative terms, there are challenges to conventional video-assisted surgery, mainly with operator visualisation and dexterity. The surgeon is required to look at three-dimensional objects on a flat television screen and overcome various perceptual difficulties such as ‘the fulcrum effect’. The fulcrum effect describes the inverted, scaled motion of an instrument’s tip, relative to its handle. The instrument’s point of entry into the body turns on a mechanical pivot. In practical terms, for the instrument’s tip to go down, the surgeon will need to raise their hand rather than drop it, or move their hand right for the tip to move left. In addition, dexterity is compromised by reduced sensation and tactile feedback, amplification of tremors and limited degrees of motion. As a result, if something goes wrong in a video-assisted case, it can be difficult to fix, and the surgeon may have to convert the operation to an open technique. Some surgeons will tell you: ‘If you can’t do it open, you shouldn’t be doing it keyhole.’

The most prevalent robotic systems today enable robot-assisted keyhole surgery and can be considered as the next generation of video-assisted operating. Essentially, the only difference is that the surgeon is not manually operating instruments but does so via the robot. Robotic surgery successfully tackles some of the challenges inherent to conventional keyhole surgery. Regarding dexterity, the robotic arms do not shake and have many more degrees of freedom than the human hand. This has greatly improved the ergonomics of infamously difficult areas to operate in, such as deep within the pelvis. In terms of visualisation and perception, robotic systems are equipped with three-dimensional cameras, and the robotic arms correct for the fulcrum effect to exactly replicate the surgeon’s hand movements. These robot-human interactions paradoxically reveal the fundamental visuospatial limits within which a human surgeon approaches an operation. The robot, used much like any other surgical tool, offers an elegant fix for problems of basic physics.

Even so, laparoscopic, video-assisted surgery remains far more prevalent than robot-assisted surgery. The most basic limitation is cost, and consequently the availability of robotic surgical systems. Furthermore, the learning curve for robot-assisted surgery is steep. It requires rigorous governance for training including proctored (supervised) cases early in a surgeon’s experience. Occasionally catastrophic complications can arise due to errors in visualisation and rough handling of tissues by robotic instruments, which lack a sense of touch; but these same issues arise in traditional video-assisted surgery.

The key barriers to widespread adoption of robotic systems will be easily overcome: with the end of da Vinci’s market monopoly, costs will come down and integration of robot-assisted techniques in surgical training will flatten learning curves. We are far enough along that road that even within a ‘cash-strapped’ public healthcare system like Britain’s NHS, robot-assisted surgery is routine, established practice in many leading hospitals.

That is where we are now, with a human surgeon still very much at the centre of the operating theatre, skilfully using robotic systems as an effective surgical instrument. But what about the future? A recent paper released by King’s College London heralds ‘pivotal and disruptive implications for surgery’ with the integration of Artificial Intelligence and robotics. The authors outline a futuristic vision which includes the redefinition of the idea of surgery itself, where ‘integrated robotic systems… undertake roles throughout all perioperative phases, including intraoperative surgical actions, emergency responses, and assistive and logistical functions’. Multiple authors are human surgeons, so it is no surprise that their bold vision includes the key caveat that ‘surgeons will continue as procedural leaders, responsible for supervision, coordination, and high-level decision-making’.

Complex innovation, such as building an autonomous surgical robot, requires research to master a series of technological leaps. For existing robotic surgical systems, these steps included high-definition, fog-resistant intra-corporeal cameras, wireless communication between the console and robotic arms, and, crucially, the machine’s ability to move in real time with the surgeon’s hand-movements. Creating an autonomous surgical robot that makes appropriate, real-time decisions based on what it encounters inside the human body would be vastly more complex. This really pushes the limits of what can be imagined possible with current technology, and it is challenging to even know where to start exploring this concept. Driverless cars offer a useful analogy. We are sufficiently advanced in the field of driverless cars that these are safely implemented in San Francisco and currently being rolled out in London. Turning driverless cars from a pipe dream into reality required major technological leaps: advanced sensors and real-time AI processing to read the immediate surroundings, complex planning algorithms for safe decision-making, and high-definition mapping to track the car’s exact position.

Driverless cars can understand their surroundings, and this is the first key challenge for the robot surgeon. Information from cameras, radar, ultrasonic sensors, laser-based 3D mapping and GPS is fused to create a robust understanding of the car’s environment. Driving safely largely depends on visual cues, and the multimodal sensors are clearly advanced enough that sensing, understanding and deciding courses of action is possible for driverless cars. Precise self-localisation is also a given with GPS and high-definition, high-fidelity maps of city roads that can be readily created, so the cars always know where they are.

Now consider the robot surgeon’s working environment – the human body. A robot surgeon cannot just rely on visual cues. Without probing blindly, a surgeon must incorporate sensory information including depth, thickness and temperature. Tricky, albeit not unimaginable with modern sensor technology – but how will it localise or know where it is, inside a human?

On the operating table, no two patients are the same. There are obvious externally visible differences in height, weight, sex, age, skin colour, limb length, body hair, and we are just as different on the inside with myriad subtle or occasionally significant anatomical variations. Many organs remain hidden, encased in connective tissue. Creating a universal human roadmap for the robot surgeon is a quixotic endeavour, far removed from the rarely changed and predictable layout of a city’s streets. The first technological leap required would be to create a three-dimensional map pre-operatively for each patient, which could then be fed to the robot. We do currently create 3D reconstructions from cross-sectional imaging, such as CT or MRI scans, but not with the definition and detail that a robot surgeon would require to self-localise with millimetre precision. The key challenge is that humans are living, changing, biological systems, with internal structures that move and grow.

The next technological leap would be to concentrate surgical experience into a decision-making algorithm that the robot could then use to operate independently. Intra-operative surgical decision-making is acquired by surgeons over decades of experience. A hybrid approach, where the robot defers the decision to a human surgeon when it is ‘unsure’, is imaginable, but the algorithm would still be required to know the options. How would it come to know them? Just training the algorithm would be an additional subsidiary challenge, conceptually, because surgical technique and decision-making are individualised to the patient and therefore cannot be standardised; practically, that could be achieved by feeding the machine thousands of hours of intra-operative film.

That points to the next set of problems inherent in machine-led surgery, and these are primarily ethical, regulatory and legal. Should the robot or its manufacturer or the hospital be held liable for errors? How many patients today would accept an operation from a robot surgeon over its human counterpart? The robot surgeon’s error or complication rate would have to be proven to be far lower than the human. Similar rates of error would simply not be sufficient. And early errors would be sure to push back progress by decades. In case of error or malfunction, a human surgeon would have to bail the robot out: would they be held liable for an error they did not commit? Would they even be capable of fixing the robot’s mistake if all routine cases were now performed by robots? What if there was a power outage or cyber-attack? Could human surgeons step in seamlessly? Traditional ‘open’ surgical expertise will still be required in case robot-assisted operations do not go to plan or in the event of system-wide failure: would surgeons still be trained properly despite performing a fraction of the cases their predecessors had in their logbooks?

Medical innovation has its own complex history: high ideals and barriers surmounted, certainly, but nothing won without a degree of grubbiness.

There is no better example than Edward Jenner and the first modern vaccine. The now eradicated smallpox virus, associated with a 30 per cent mortality rate, is estimated to have caused 500 million deaths in its final century of existence. Building on the observation that previous infection with the related but less virulent cowpox virus conferred some level of protection, Jenner inoculated the eight-year-old James Phipps with pus scraped from a milkmaid’s cowpox blisters. He went on to inject Phipps with ‘variolous material’, taken from a smallpox patient, a procedure not without risk, leading to death in one to two per cent of individuals. Thankfully, Phipps survived and, although the Royal Society rejected Jenner’s findings, he self-published his research and eventually gained Royal support to petition Parliament to further his lifelong fight to eradicate smallpox.

In the 20th century, the medical ethicists Beauchamp and Childress devised four principles of medical ethics that are commonly expanded to seven: autonomy, beneficence (do good), non-maleficence (do no harm), justice, confidentiality, fidelity and veracity. Jenner was one of history’s greatest medical innovators, described even by Napoleon as ‘one of the greatest benefactors of mankind’, whose discovery saved millions of lives in eradicating the deadly pox. How many modern ethical principles might he have transgressed during his experiment on an unknowing child, with no certainty of its consequences? It should come as no surprise that Jenner’s medical colleagues took time to accept his findings and there was significant indignation and resistance from the public in response.

James Gillray’s 1802 cartoon The Cow Pock — or — the Wonderful Effects of the New Inoculation! reflects this anxiety. It features newly vaccinated patients sprouting bovine features, described in the caption as ‘the wonderful effects of the new inoculation’. However, with efficacy proven over time, and benefits counted in the millions of lives saved, vaccination eventually surmounted popular resistance, leading to successive Vaccination Acts enshrined in law. The 1853 Vaccination Act was the turning point and marked the first time the British state had imposed mandatory medical treatment across the general population. Smallpox vaccination for infants within the first three months of life became compulsory and non-compliance was punished.

This major milestone in public health policy was met with organised resistance, including the formation of the Anti-Vaccination League, the inspiration for the tactics of modern COVID anti-vaxxers. The movement culminated in the 1885 Leicester rally, which gathered, by the highest estimate, 100,000 protesters. Thereafter, a conscientious objection clause was included in the 1898 Vaccination Act. Today, newborns are vaccinated against 15 different infectious diseases during the first 18 months of life, a standard of care replicated throughout the developed world. The UK Health Security Agency estimates that 4,960 deaths and at least 228,447 hospitalisations are prevented annually as a result. If such a benefit could be demonstrated for autonomous robotic surgery, it would eventually overcome public and expert resistance, but this process would potentially take decades and need to rely on an extremely rigorous evidence-base.

The challenges are illuminated by the basic medical interventions robots are good for, such as blood taking, which can be performed autonomously by machines with a similar accuracy to humans. This is a simple task which is commensurate with our current technology: state-of-the-art sensors can detect visual cues to direct a needle for puncture, and the decision-making is basic. Growing from this platform, a minor operations autonomous surgical robot, which might remove benign skin lesions, is within reach. Again, its success relies on the basic quality of the decision-making process, although fringe cases such as lesions that turn out to be malignant may have to be deferred to a human being. Although mainly in experimental contexts, there are also robots which can autonomously perform a specific step in an operation. This remains within the realm of robot-assisted surgery, and unless a vast superiority is proven over the human hand, it will be difficult to justify both the cost and the impracticalities of wheeling in a usually bulky robotic system mid-case to perform a single step.

Robotic assistance in surgery will certainly become more prevalent, but the surgical profession should take care to preserve vital traditional operating skills in case of error or malfunction; surgeons are much more than ‘procedural leaders’. The robot surgeon performing operations independently in routine clinical practice, start to finish, ‘skin-to-skin’, is not going to arrive any time soon. In medicine, we are a long way from the robotic ideal in which, as Asimov writes, ‘you just can’t differentiate between a robot and the very best of humans’.

This essay reflects my own views and experiences, based on real-world practice. Surgeons are certainly not the very best of humans, but the practical decision-making that makes medicine and surgery inextricably bound up with human pain as well as human excellence is what fundamentally differentiates clinicians from robots. This is a science and a project that human beings need years of day-to-day practice to understand; a standard alien to the perfection we so often project onto machines.