In May 2013, the FDA approved a system called Sedasys, a computer-assisted personalized sedation device built by a division of Johnson & Johnson, to administer propofol sedation for routine colonoscopies without an anesthesiologist in the room. It was, on paper, exactly the scenario this profession has worried about for a decade: a regulator-approved machine doing part of an anesthesiologist's job, cheaper and at scale.
By March 2016, Sedasys was gone. Not banned. Not linked to a safety failure. Ethicon, the J&J division that made it, withdrew it from the market citing business priorities, after years of weak sales in a reimbursement environment that never rewarded hospitals for using it. The technology worked. The market did not want it, in large part because the anesthesia profession, hospital administrators, and insurers all found reasons not to adopt a machine that automated a procedure most of them still wanted a physician standing next to.
That history matters more than any capability benchmark for understanding where this profession sits today. The displacement risk is classified as Moderate, with a horizon toward 2048, one of the longer horizons on any Moderate-risk profession AI Doomsday tracks. AI is doing real, measurable work inside the operating room right now. It has not gotten meaningfully closer to replacing the person responsible for keeping a sedated patient alive, and the one time it tried commercially, the obstacle was not capability. It was everyone else in the room.
Key Points
- Anesthesiologists are rated Moderate risk with a displacement horizon toward 2048, one of the longest horizons among Moderate-risk professions, reflecting a technology curve that has repeatedly outpaced clinical and market adoption.
- Sedasys, an FDA-approved automated sedation system, was withdrawn from the market in 2016 for commercial reasons, not safety failures, three years after approval, illustrating that regulatory clearance does not guarantee adoption in anesthesia.
- AI monitoring tools already reduce real clinical risk: Edwards Lifesciences' Hypotension Prediction Index helped 41% of monitored patients avoid intraoperative hypotension versus 12% in unmonitored controls in published studies.
- The U.S. anesthesiology workforce faces a projected shortfall of roughly 6,300 physicians by 2036, with 56.9% of practicing anesthesiologists over age 55 and record numbers of unmatched residency applicants, a demand problem AI does not solve.
- The American Society of Anesthesiologists frames AI explicitly as a co-pilot for risk stratification, monitoring, and documentation, not a substitute for the physician who holds legal and clinical responsibility during sedation.
What an Anesthesiologist Actually Does
The job is frequently reduced, in public imagination, to administering a dose and watching a monitor. In practice, it is continuous risk management under changing physiological conditions: assessing a patient's fitness for anesthesia beforehand, titrating drugs in real time against a body that is actively responding to surgery, managing airway and breathing, anticipating and reversing complications, cardiac events, allergic reactions, sudden blood loss, before they become emergencies, and coordinating that entire process with a surgical team working on a timeline that does not pause for uncertainty.
It is a job defined by the cost of a rare failure rather than the difficulty of the routine case, which is precisely why automation in this field has always been evaluated against a different bar than automation in, say, bookkeeping or customer support.
The Machine That Already Tried
Sedasys is worth dwelling on because it inverts the usual AI Doomsday pattern. Most professions on this list face a technology that is improving and a market eager to deploy it as soon as it clears a capability bar. Anesthesia sedation cleared that bar in 2013, with FDA sign-off, for a narrow but real use case, healthy patients undergoing routine colonoscopy or endoscopy. And the market said no anyway.
THE PRECEDENT
Industry retrospectives on the Sedasys withdrawal are explicit that no safety concern drove the decision. The system was pulled amid weak sales, a reimbursement structure that did not favor its use in bundled GI procedure payments, and resistance from the anesthesia community that would have been displaced from cases it covered. A technology can clear every regulatory hurdle and still lose to the economics and politics of the room it was built to enter.
That does not mean the underlying technology stalled. Closed-loop anesthesia delivery predates Sedasys by five years: McGill University's McSleepy system, demonstrated in 2008, was described by its own developers as a "world first" for fully automated anesthesia delivery, monitoring depth of hypnosis via EEG, pain response, and muscle relaxation simultaneously, explicitly framed by its creators as assisting anesthesiologists rather than replacing them. Target-controlled infusion, a less autonomous but widely adopted precursor, has automated propofol dosing against pharmacokinetic models in clinical use since 1996.
What AI Is Doing Now
The clearest current success is predictive monitoring. Edwards Lifesciences' Hypotension Prediction Index analyzes arterial waveform data to flag impending intraoperative hypotension five to fifteen minutes before it occurs. In published clinical evidence, patients monitored with HPI spent roughly two minutes in a hypotensive state during surgery, compared with 28 minutes for unmonitored patients, and 41% avoided intraoperative hypotension entirely against 12% in controls. A 2025 review found EEG-based depth-of-anesthesia monitoring reaching roughly 89% accuracy using deep learning, and computer-vision airway assessment tools screening for difficult intubation with sensitivity above 80%.
Documentation is the other clear win. A 2025 study found large language models extracting structured registry variables from operative notes in about 1.2 minutes, down from 15.5 minutes manually, at over 97% accuracy, and improving prediction of 30-day postoperative mortality by as much as 38.3% over traditional methods when applied to clinical notes at scale. None of this replaces the anesthesiologist's presence in the room. It compresses the administrative and predictive layers around that presence, similar to the pattern AI Doomsday has documented across academic research and nursing: the paperwork and pattern-recognition shrink, the moment-to-moment judgment does not.
Why the Job Still Requires a Physician
The American Society of Anesthesiologists has been consistent and public about where it draws the line. Its research council frames AI explicitly as a tool for risk stratification, monitoring augmentation, and workflow support, with the anesthesiologist retaining final clinical decision authority. The FDA's regulatory posture reinforces the same boundary: autonomous AI systems for high-risk procedures like anesthesia continue to require direct medical supervision, a human-in-the-loop standard that Sedasys itself was approved under and that no successor system has been cleared to operate without.
Liability compounds the technical and regulatory caution. When a sedated patient has a hemodynamic crisis mid-procedure, someone with clinical authority and legal responsibility has to decide, in seconds, how to respond. No jurisdiction currently assigns that responsibility to a closed- loop delivery system, and the Sedasys episode suggests that hospitals and insurers were unwilling to test that boundary even for a narrow, low-acuity procedure with regulatory sign-off already in hand.
The Real Pressure Is a Shortage, Not Automation
The more consequential story for anyone entering this field is workforce, not automation. Industry workforce analyses point to a projected shortfall of roughly 6,300 anesthesiologists in the United States by 2036. Nearly 57% of the current workforce is over age 55, and 2025 residency matching data shows more than 1,200 applicants going unmatched against roughly 1,800 available first-year positions, a pipeline bottleneck layered on top of an aging supply. The ASA issued a public statement in 2024 warning that the workforce shortage itself poses a threat to surgical access, a very different framing from the displacement narrative that usually accompanies AI coverage of a medical specialty.
That context reframes the AI tools already in deployment. A profession running short of physicians relative to surgical demand has a direct incentive to use predictive monitoring and automated documentation to make each anesthesiologist more efficient per case, not to reduce headcount. The 2048 horizon reflects that dynamic: displacement pressure exists, but it is outrun by a labor shortage that automation is currently being recruited to soften rather than cause.
How to Use AI as an Anesthesiologist Now
For hemodynamic management: predictive tools like HPI are a genuine safety upgrade when integrated into practice, giving a fifteen-minute warning window that manual monitoring cannot match. Treat the alert as a prompt to investigate, not a diagnosis.
For documentation and risk scoring: LLM-based extraction and risk prediction tools cut administrative time meaningfully. Review their output the way you would a resident's note, useful, fast, and still your signature on the chart.
For depth-of-anesthesia and airway assessment: EEG-based and computer- vision tools are strong enough to be a second opinion worth having, particularly in ambiguous cases, but the published accuracy rates, even at their best, still leave a meaningful error margin that clinical judgment has to cover.
What I Think
The 2048 horizon looks right to me, and the Sedasys history is the reason why. This is one of the few professions on this list where a company already built the thing people worry about, got it approved by the actual regulator, and still could not get anyone to use it at scale. That is a stronger signal than any capability benchmark, because it shows the resistance to full automation in this field runs through economics, liability, and professional structure, not just unfinished technology.
What I would watch instead of capability is the workforce math. A field with an aging, shrinking supply of physicians and rising surgical demand has every incentive to lean harder into AI-assisted monitoring and documentation, and that pressure is a more realistic driver of adoption over the next decade than any argument about what a closed-loop system can technically do. The tools are already good enough for parts of this job. What determines whether that matters is whether the same forces that killed Sedasys in 2016 have actually changed, or whether they are just waiting for the next version.
"A machine has already been cleared by the FDA to do part of this job, and the profession is still here. The lesson isn't that the technology failed. It's that clearance was never the hard part."