Surgery as a case study in imitation learning

While artificial intelligence advances rapidly across numerous domains, certain breakthroughs stand out for their pace and potential. Among the most striking developments is imitation learning, AI systems that acquire skills by observing and replicating expert demonstrations rather than through explicit programming. Surgical care has become a key proving ground for this technology, where robots master intricate procedures by watching surgeons, adapt to anatomical variations in real time, and expand access to expertise. Yet realising this potential at global scale demands equally sophisticated technology across the entire stack. The challenge ahead becomes building infrastructure capable of coordinating these AI systems across borders, institutions, and regulatory frameworks while maintaining the same pace of advancement.

A turning point in surgical care

Robotics has been part of operating theatres for more than two decades. Early systems magnified the surgeon’s view and translated hand movements into steadier motions. Every action, though, still came from a person. The present moment is different. Artificial intelligence is beginning to take on steps of a procedure itself.

The years 2024 and 2025 mark a turning point. Research groups are showing that robots can learn by watching expert demonstrations, while hospitals are deploying new platforms that give clinicians sharper tools. Together, these developments highlight the potential of imitation learning, where robots acquire surgical skills through observation and apply them to deliver safer, more accessible care. The significance lies in outcomes patients and staff can already feel: safer movements, shorter stays, and the possibility of bringing skilled care to more people.

Learning by watching

One of the most striking demonstrations comes from Johns Hopkins University¹. Researchers trained a system called SRT-H by showing it recordings of gall bladder removals, annotated with simple task descriptions. The robot learned to grasp tissue, place clips, and cut safely, then carried out the procedure on pig organs. Across eight trials, it completed every step, adjusting to anatomical differences and responding to spoken instructions in real time².

While the novelty is impressive, the real breakthrough lies in the method itself. Instead of hard-coded routines, the robot learned by imitation, much like a trainee surgeon refining their skill. That ability to adapt mid-procedure shows a path for transferring reasoning, not only tasks. It is still early as these experiments were outside the human body, but the principle has been proven: AI can learn surgical reasoning through observation.

What makes imitation learning powerful is scale. Training a robot on hundreds of recorded surgeries creates a library of skill far beyond what any single human can access. Systems can combine video, audio and sensor data, learning not just what to do but why certain decisions matter. This is where vision-language models and reinforcement loops become critical. A robot can link what it sees with verbal instruction, then refine its approach with feedback. It is a shift from scripted assistance to adaptive collaboration.

Research breakthroughs to patient care

Research advances are matched by developments in hospitals. The da Vinci 5, cleared in the United States in 2024³ and CE-marked in Europe in July 2025⁴, has entered use at centres including UF Health⁵. The system builds on earlier generations with three-dimensional vision, higher resolution imaging, and instruments that move with steadiness beyond human hands.

For patients, this translates into smaller incisions, less trauma, and quicker recovery. Surgeons report clearer visualisation and more precise control, while hospitals see reduced operating times, shorter stays and fewer complications. These gains are measured at the bedside and in recovery times, not just in technical specifications.

Global demand now

The context for these changes is a rapidly expanding global market. Analysts forecast that surgical robotics could grow from about eight billion dollars in 2022 to more than eighteen billion by 2029⁶. Indeed, growth is not uniform: North America leads adoption today, Europe is catching up rapidly following CE approvals, and Asia-Pacific is predicted to see the fastest expansion thanks to investment in teaching hospitals and rising healthcare spend.

Several forces are driving this global adoption. Ageing populations bring higher surgical volumes and pandemic backlogs have pushed hospitals to increase throughput. At the same time, patients increasingly expect minimally invasive procedures as standard, and payers see reduced complications as a way to manage costs. Hospitals are not adopting new systems for prestige alone, rather they are responding to structural pressure.

Global access when?

The deeper story is access. In India, rural community health centres face an 80% shortage of surgeons, gynaecologists and other specialists⁷. Similar shortages are seen across sub-Saharan Africa, where urban hospitals concentrate scarce expertise. AI-assisted platforms could reduce this divide. A regional hospital equipped with robotic systems can perform more routine procedures locally, or could even be remotely operated and supervised, sparing patients the cost and risk of travelling hundreds of kilometres.

Training offers another pathway to improving access. If systems can learn from expert video, they can also help humans learn. AI systems trained through imitation learning can extend expertise across geographies, giving junior clinicians realistic guidance until muscle memory and judgement align. Best practice can move faster from top centres into everyday care. This is how advanced medicine spreads. Not by waiting decades for more specialists, but by accelerating how skills are shared.

Ethics

Critically, trust is decisive for mainstream adoption. Surveys show that people are not rejecting innovation, they are just asking for clarity. A UK poll found more than half the public support AI in healthcare tasks, though few are ready for full autonomy. NHS staff were more supportive but stressed oversight⁸.

In the United States, the most recent Philips Future Health Index 2025 survey of more than 16,000 patients and 1,900 healthcare professionals shows a nuanced picture. While 70 percent of U.S. patients report worsening health because of delays in seeing a doctor, around 60 percent believe AI can help by enabling departments to serve more people more effectively, easing bottlenecks in access. Yet trust remains conditional. Only 29 percent fully trust AI for basic health advice, and 70 percent raise concerns about privacy and reduced human interaction, with women showing higher apprehension in areas such as cancer diagnosis⁹.

These surveys point toward a consistent pattern. Patients recognise that AI could shorten waits and improve equity of care, but they also want safeguards: transparent use, clear accountability, and systems that preserve human oversight. Support for adoption is high, but confidence will only grow if implementation tackles these concerns directly.

Data privacy

Privacy concerns sit at the heart of this trust gap. Behind every capable system is a dataset, and in healthcare that means patient information. Training AI for surgery requires scale and diversity, but privacy cannot be treated as secondary. The European Union’s AI Act, in force since August 2024, now makes this legally binding. It classifies most healthcare applications from diagnostic tools to surgical robots as high risk, with strict requirements for accuracy, robustness, and human oversight¹⁰.

For developers, this means transparent risk assessments, audited data governance, and clear explanations for patients and clinicians about how systems work and when they are in use. Hospitals must be able to intervene at any point, and conformity assessments ensure these safeguards are in place before deployment. Non-compliance carries significant penalties, signalling how seriously regulators view healthcare AI.

The AI Act also highlights a sovereignty challenge. Many surgical platforms today are designed with American hardware, trained on U.S. patient data, and then deployed in Europe, where they inevitably interact with European patient information. That data may later be anonymised and folded back into global models. This flow raises real questions: whose standards apply, whose data rights prevail, and who is accountable across borders?

Maturity signals

Evidence of uptake continues to grow. Intuitive, the company behind the da Vinci systems, reports 14 million procedures completed worldwide. New models bring more than one hundred design improvements, and regulatory approvals keep pace, with CE marking now in place across Europe. With surgical AI now in practice, the focus turns to responsible scaling.

What matters most is not a single robot or platform but the infrastructure around it. For surgical AI to travel across institutions and countries, systems must coordinate intelligence in ways that respect sovereignty, law, and culture. Responsibility has to be part of the architecture, not an addition later.

This is where Stelia’s focus lies: creating the conditions that let intelligence flow while remaining accountable.

Better outcomes for more people

Artificial intelligence is already reshaping the operating theatre in measurable ways. Research demonstrates that robots can learn by observation, while hospitals track real improvements and patients are beginning to experience the benefits firsthand.

The promise is not automation for its own sake. It is safer operations, faster recoveries, and wider access to skilled care. The real measure is lives improved through responsible intelligence. That is the horizon worth working toward.