HeartGenesis and Compilence have agreed to form Compilence Care and establish an experimental program at an existing hospital in Japan. The aim is to combine robotic support and AI assistants with clinical expertise, preserving rare human skills and extending their reach.

In 2018, I predicted that robots would learn to perform surgical tasks by drawing on human expertise. At the time, I was exploring the intersection of machines, tissue regeneration, AI, and cybersecurity at the Basic Research Meeting.[9] The prospect raised a question that has stayed with me: when a machine takes on a physician’s task, how do we preserve the knowledge and judgment that make the task safe?

Eight years later, that prospect is beginning to take practical shape. HeartGenesis and Compilence have agreed to create a company, Compilence Care, and develop an experimental robot-augmented, AI-assisted setting within an existing hospital in Japan.[7] The planned site would bring the question into a clinical environment: which parts of a physician’s expertise can be transferred to machines, under what conditions, and with what benefit to patients?

HeartGenesis and the participating physicians would contribute surgical methods and clinical experience. Compilence develops the underlying AI infrastructure for controlling and tracing machine behavior. The proposed company would bring those contributions together for patient care and surgical development. The agreement sets a direction; company formation and deployment of the hospital program remain ahead. Any clinical use must be supported by appropriate evidence and the required institutional and regulatory approvals.

The da Vinci surgical system, developed by Intuitive, provides the historical starting point. It is a surgeon-controlled robotic platform used for minimally invasive operations: the surgeon works at a console, guiding instruments inside the patient’s body. The system received FDA clearance through the 510(k) process in July 2000.[1] Robotically assisted surgery was already established when I spoke in 2018. The frontier I was looking toward was the transfer of learned skills into increasingly autonomous execution.

That transition now poses a practical challenge: how can an exceptional physician’s expertise become available beyond their own operating room, with its conditions, limits, and accountability intact?

Research outside this collaboration offers one example of what is becoming possible. In a 2025 report on Surgical Robot Transformer-Hierarchy, or SRT-H, developers described training a robotic system through imitation learning on demonstrations of surgical tasks. The system then autonomously performed the clipping and cutting steps of a gallbladder procedure on eight previously unseen, excised porcine gallbladders. Those steps comprised 17 tasks, and the researchers reported completion without human intervention in all eight tests.[14] These were experiments on tissue outside the body, covering part of an operation.

For Compilence Care, the primary goal is to preserve rare human clinical skills and extend their reach: retaining expertise built over decades and making its validated, transferable elements available to more teams and patients. Robot-augmented hospitals and healthcare centers could then use that expertise more widely while allowing specialists to focus on complex decisions. Reducing repetitive work and the cost of care would be a further benefit, whose scale must be demonstrated alongside safety and clinical outcomes.

When a Physician Leaves, What Does the Clinic Retain?

A clinic retains medical records, instructions, and equipment. Yet much of its capability resides in individual people: how a surgeon recognizes a difficult case, chooses a sequence of actions, notices a deviation, and decides when to depart from the original plan.

After a specialist leaves, the team may lose the ability to reproduce that work at the same level, even with all the documentation available. Transferring expertise that a physician has accumulated over twenty years requires training, shared practice, and the review of difficult cases.

The work of cardiac surgeon Pradeep Nambiar, developer of the Nambiar Technique, illustrates where a solution can begin. Nambiar previously performed minimally invasive coronary bypass operations at Cleveland Clinic. In a 2013 publication, Nambiar and Chandermohan Mittal described a minimally invasive coronary bypass technique through a left mini-thoracotomy, without robotic assistance. The authors characterize the technique as reproducible and connect it with the potential for training.[2]

In his manuscript on the future of robotic surgery, Nambiar writes:

The value lies in a consistent, high-stakes workflow that can be broken into phases, annotated for anatomy and instrument paths, and linked to outcomes.

For future robotic support, both the phases themselves and the conditions for moving between them matter. To transfer a skill to a machine, each action must be linked to the clinical grounds for performing it.

Video shows movement. Skill includes understanding when that movement is permissible, when it is dangerous, and when it must not be performed at all.

The Cost of Waiting

Another experienced specialist cannot be trained quickly. Meanwhile, demand for care is growing faster than many health systems can expand their teams. The WHO projects a global shortage of 11 million health workers by 2030, primarily in low- and lower-middle-income countries.[3]

Waiting can have clinical consequences. A systematic review and meta-analysis published in The BMJ in 2020 found an association between treatment delays for several cancers and increased mortality. For the surgical indications examined, every four weeks of delay corresponded to a mortality hazard ratio of 1.06-1.08.[4] These estimates apply to the diseases and treatments studied and help quantify the cost of time in specific clinical situations.

For a patient, delay can mean a worsening condition and fewer treatment options; in some cases, the patient does not survive the wait. The causes vary: shortages of specialists, equipment, funding, or donor material. Expanding care requires addressing each of these constraints.

Another medical example is cardiac surgeon Federico Benetti’s work with HeartGenesis. A company release from May 2024 describes long-term follow-up after cell therapy in a small series of patients with severe heart failure. Benetti, a co-founder of the company, presented these findings at a medical congress in Tokyo.[5]

Applying such an approach requires patient selection, trained specialists, and long-term follow-up. Its role in treating heart failure should be determined by clinical studies and independent assessment of the results. This creates a specific task for robotic support: expanding a team’s capacity to perform validated procedures and provide ongoing patient care.

Benetti also has a direct connection to the development of minimally invasive surgery: in 1996, he coauthored a multicenter study of video-assisted coronary operations without cardiopulmonary bypass.[11] His 2010 paper on MINI-off-pump coronary artery bypass reported long-term results in 40 patients and identified technological advances as a way to overcome anatomical difficulties and make the technique easier to reproduce.[12]

In a direct interview for this article, Benetti recalled his experience with an early experimental robot at Intuitive in California:

I was at Intuitive working with the first experimental robot in California, but I did not feel it was ready.

Benetti explains that he worked with his own engineer and an engineer from Vista to develop and improve the system at his foundation in Argentina, where he also performed the initial cases. He describes the system’s subsequent use at Intuitive and links that work to his invitation to work with its early robot. His contribution was to the development of the system; he explicitly rejects the suggestion that he invented 3D technology. He also recalls how his xiphoid approach evolved into MINI-OPCAB because many patients’ anatomy required opening the lower part of the sternum.[13] His account brings the discussion back to the practical demands of surgery: anatomy, access, visualization, and the readiness of the tools.

That experience provides a foundation for discussing the next stage: transferring surgical skills to a robotic system.

Federico Benetti describes that prospect this way:

I helped develop methods of minimally invasive coronary surgery. Now we have a path and the tools to move toward performing these operations autonomously.

Dr. Federico Benetti and Dr. Mehmet Oz

How Does a Physician’s Expertise Become a Machine’s Skill?

Transferring a physician’s expertise to a machine is one of the hardest problems in this field. An experienced surgeon continually interprets anatomy, tissue response, and changes in the patient’s condition. Much of that judgment is implicit, and a recording of the surgeon’s movements captures only part of it.

The proposed path connects four things: the physician’s experience, a defined task and its limits, the robot’s actions, and human oversight. Building those connections requires two complementary kinds of work. Demonstrations provide examples from which a robot can learn how to act. Clinical experts must also make explicit why an action is appropriate, what measurements support it, and which changes require stopping or calling for help.

For a particular step, the surgeon and engineering team would need to identify the relevant anatomy and sensor readings, annotate decisions and exceptions, and compare the robot’s responses across cases it has not encountered before. Correctly reproducing a movement in one case is only the beginning. The system must also demonstrate that it recognizes when the conditions for that movement are absent. Validation must progress from controlled testing toward appropriately authorized clinical studies.

Yurii Chudinov, co-Founder and CTO of Compilence, distinguishes the protocol from the skill: E.L.I.A. independently describes tasks and limits, while the physician’s experience becomes Knowledge Tissue, whose fragments can be improved, transferred, or deprecated and removed from active use.[6]

In his account, Delegates supply sensor and telemetry data and ensure accurate processing of new inputs. ARC translates protocols into machine instructions for traceable, real-time execution. Its admissibility levels allow behavior to be corrected even when a specific action was not explicitly programmed, within the protocol’s boundaries.[6]

The fidelity of the digital skill representation must be checked with clinical experts. Perception, dexterous movement, and the ability to handle unfamiliar anatomy also require their own development and validation. Describing a protocol and digitizing experience are parts of that work; neither alone establishes clinical readiness.

Yurii summarizes the connection:

This links human-defined boundaries, through feedback, to the machine’s actual state.

For example, a task may require a measurement to remain within a clinician-defined range. A missing or out-of-range reading should trigger the prescribed stop or request for human intervention. Demonstrating that response reliably is part of the validation required before clinical use.

A Hospital Setting for Testing the Transfer of Expertise

The planned Japanese hospital program would provide a setting for testing how this chain fits into daily work: assigning tasks, preparing staff and machines, monitoring patients, and responding to changes. Robots would support selected physical tasks, while AI assistants would help staff work with clinical information and protocols. The model keeps clinicians responsible for decisions and makes the conditions for machine action explicit.

The first development area, Elderly Care Robotics and Hospital Care Robotics, would address routine tasks, measurements, logistics, and monitoring for hospital patients and older adults. AI assistants could help staff interpret incoming information and identify when attention is needed. Each function would require validation, clear limits, and a defined point for human involvement.

The second area, Robotics for AI-augmented / assisted heart surgery, would focus on selected stages under an experienced surgeon’s control. The long-term goal remains fully robotic surgery, approached through validation of individual capabilities. Locating development within a hospital does not itself authorize autonomous operations on patients.

In both areas, clinical experts define the methods and indications. Clinical decisions remain with the physician; responsibilities for equipment, software, and the organization of care must also be clearly assigned. The practical test is whether a task can be delegated safely enough to give specialists more time for patients.

James Ryan of HeartGenesis describes the goal of the joint engineering effort:

We are working with Compilence to make this mechanism compact, portable, fast, and capable of operating in near real time. Our goal is to integrate it into real-world autonomous systems capable of working safely alongside people.

That requires measuring the complete response from sensing to action, including the time needed to stop or return control to a person.

The immediate organizational step is to form Compilence Care and define the scope, oversight, and deployment plan for the experimental hospital program.[7] Its value should be measured by whether a transferred skill remains reliable across cases, staff can supervise it effectively, and patients benefit. Longer-term plans also include residential care, with premium retirement communities as a possible initial market. The broader ambition is to make exceptional clinical expertise available to hospitals worldwide.

Expertise Must Travel With the Boundaries of Its Use

To scale across hospitals, the digital representation of a skill must retain its source, version, conditions of use, and limits. Knowledge Tissue allows the team to improve and transfer selected fragments, or retire them from use as clinical knowledge changes. A receiving system must apply the relevant E.L.I.A. protocol to govern their use. Compilence is developing compact representations to reduce storage and processing demands; quantitative benefits remain to be demonstrated.[6]

Compact private data centers of 1–3 MW per site could support several healthcare facilities with approved knowledge, updates, and audit records. Their smaller scale could simplify construction and permitting, reduce land disturbance, and place less demand on the local grid than a large centralized facility, particularly when using existing buildings and power connections.[8] Environmental benefits depend on siting, energy sources, and cooling; all required approvals still apply. Responsible organizations must retain control of access and rule changes, while safe execution and emergency stopping remain local even if the network connection fails.

Why This Is Also a Cybersecurity Issue

The cybersecurity question I was considering in 2018 becomes more consequential as robots gain autonomy. Transferring a specialist’s expertise to a machine requires protecting the entire chain, from the provenance of knowledge and the boundaries of its use to the integrity of sensor data and authorization to act.

An undetected alteration of a skill’s digital representation creates the risk of acting in a physician’s name on a different basis. Data sovereignty in such a clinic must therefore mean control over knowledge versions, keys, updates, logs, and the ability to stop execution.

I also include quantum-resistant cryptographic protection for data storage and transmission channels among the requirements for this infrastructure. Stored data requires strong symmetric encryption and secure key management; communication channels require post-quantum mechanisms for key establishment and authentication. Signatures for protocols and updates must also account for the quantum threat. This is especially important for medical information that remains sensitive for decades: encrypted data intercepted today may be retained for future attempts at decryption.

The future I anticipated in 2018 is beginning to take shape through research progress and concrete plans to bring surgical expertise and AI development into the same hospital setting. HeartGenesis and Compilence’s agreement to form Compilence Care is a step toward that work. The planned program in Japan would test whether rare human skills can be preserved, transferred, and applied under clinical supervision. Its success will depend on evidence that each capability performs reliably and stops when the situation exceeds its limits.

The measure of progress is timely, safe care for more people, with reproducible outcomes and clear human accountability. Those are the standards by which Compilence Care, and every initiative in this field, should be judged.

Sources

1. FDA. Computer-Assisted Surgical Systems; 510(k) decision K990144, July 11, 2000.

2. Nambiar P, Mittal C. Minimally invasive coronary bypass using internal thoracic arteries via a left minithoracotomy: “the Nambiar Technique”, 2013; abstract. Workflow quotation: Nambiar’s manuscript on robotic surgery.

3. WHO. Health workforce. Projected global shortage: 11 million health workers by 2030.

4. Hanna et al. Mortality due to cancer treatment delay: systematic review and meta-analysis. BMJ. 2020;371:m4087.

5. HeartGenesis press release, May 22, 2024.

6. Yurii Chudinov. Human interface to ARC…, language documentation; technical explanations and approved quotation provided for this article.

7. Compilence Care project background provided for this article: the HeartGenesis–Compilence agreement and planned hospital program in Japan. James Ryan’s quotation was approved for this article.

8. J-Comp Research Datacenter, Solution 1: modular sites of 1.5–2 MW. The article uses the proposed deployment range of 1–3 MW per site.

9. Basic Research Meeting, 2018, program excerpt. Chuck Brooks: “Merging of modulating machines and tissue regeneration: when novel paths in AI and cybersecurity are needed.” The prediction is recounted from the author’s recollection.

10. Federico Benetti. Statement on the transition to autonomous surgery, approved for this article.

11. Benetti F et al. Video-assisted minimally invasive coronary operations without cardiopulmonary bypass: a multicenter study, 1996. PMID 8975839.

12. Benetti FJ. MINI-off-pump coronary artery bypass graft: long-term results. Future Cardiol. 2010;6(6):791–795. PMID 21142636; abstract.

13. Federico Benetti. Interview for this article and subsequent audio clarification; timecoded automatic transcript. The interview quotation is lightly edited for clarity.

14. SRT-H: A Hierarchical Framework for Autonomous Surgery via Language Conditioned Imitation Learning. Preprint, 2025; full text.

@BC360 TODAY

Message History

Delivery option(s)