Today’s inane image of the day:
I feel like everyone’s talking about AI taking jobs. It seems like the tech world has been impacted by the AI boom with reductions in force. I’ve heard that those that managed to keep their jobs are using AI-generated code in place of an entry-level coder. They are AI-augmented workers. It’s crazy to think that when I was growing up, computer programmers/software engineers had the most stable job prospects. After all, we rely so heavily on computers and the software that runs on them!
Growing up in an immigrant household with a scarcity mindset, it was always on my mind that I wanted to select a stable career. One that would always have stable job prospects. The typical teaching was that doctors, lawyers, and engineers would always have work. I think that even in this day and age of AI, this continues to ring somewhat true.
I was having a conversation with someone about AI and it came up that we will still always need judges. We will always need a human to render decisions based on the rules that we’ve set forth. That’s a very human job. So there are avenues for lawyers.
And we will always have physical products that need engineers to be part of the design/creation/maintenance.
But anyway, I’m a doctor, so I’m going to speak on the field I practice in.
Specifically, I am an anesthesiologist and an ICU physician. When I was deciding whether or not to do a fellowship after my anesthesiology residency, I actually opted for critical care medicine because I thought it was an insurance policy for if anesthesiologists were to become obsolete. One thing you should know about the field of anesthesia is that we’ve seen a lot of ups and downs over the last few decades. There was a time where anesthesiologists struggled to find work which then set the stage for a huge shortage of anesthesiologists so the pendulum swung dramatically the other way and people were being paid crazy salaries.
Currently, I believe we are in a bubble in the anesthesia market. The job market is so hot. There are some really nice salaries being offered. People have been jumping ship from groups because groups down the street are poaching them with more money for less work. Why would you pass up that offer?! But history tends to repeat itself so I am skeptical that the demand for anesthesia services will remain this high.
Either way, you came here to read a rant about how my job [the anesthesiologist part; maybe I’ll tackle the ICU part at a later date] is AI-proof. Apparently this Forbes list agrees with me. (The author should have investigated the difference between a nurse anesthetist and an anesthesiologist since CRNAs were #1 and we were #11 on this list. I’ll interpret this as because the training pathway to become an anesthesiologist is longer and more expensive, it fell lower on the list.) (Also of note, 7 out of the 20 on that list are medically-related so I guess those of us in medicine picked wisely!)
So… let’s get into it:
These are the reasons that anesthesiologists and those of us that are also critical care trained are protected, for now:
- Real-time decisions with life or death consequences
- Procedures
- Liability/regulatory limitations
- Interdisciplinary coordination
- Human touch
- Cost
The areas where I want AI/technology to help us:
- Preoperative assessments
- Early alert warning systems
- Drug dosing and delivery systems
Real-time decisions with life or death consequences
Everyone loves to compare the practice of anesthesia with being a pilot. There are critical times in a flight that a pilot’s decision making could profoundly impact the trajectory of the airplane… the same thing goes for anesthesia providers. These critical portions of an anesthetic are typically induction [like takeoff – the process of anesthetizing/sending someone to “sleep”] and emergence [landing the plane – turning off the anesthetic and waking them up]. Procedural portions [intubation – placement of a breathing tube or device, arterial lines, central lines, epidurals/spinals, regional blocks, etc] of our practice can also have complications/consequences and thus are critical.
Sometimes unexpected reactions/complications happen related to the anesthetic. Patients can have anaphylaxis to our medications – having personally cared for a patient who had anaphylaxis before, it has major hemodynamic consequence and high risk of mortality [death] if not immediately managed. Luckily, the operating room is literally one of the safest places to experience this medication complication because we are well-equipped and well-versed with the treatment of this entity.
The other critical times are related to surgical factors. Sometimes we can anticipate major blood loss AND hemodynamic swings [like an open aortic repair… we know a patient will bleed and experience swings in blood pressure with clamping and unclamping]. Other times, a surgical complication can occur and it’s a surprise.
While the surgeons I work with typically are excellent communicators, they’re also PERFORMING SURGERY. So if something isn’t going smoothly, they often get hyper-focused on fixing the problem. One skill I learned in training and during my practice is being attuned to the signs [suction, tense voices, request for more surgical clips/ties in rapid succession, etc] that I need to be on high alert.
Either way, the operating room is an extraordinary dynamic environment. AI can work well with non-time sensitive medicine. Outpatient clinics. As scribes. When learning about a specific diagnosis or trying to generate a differential diagnosis. Radiology images. But at this point in AI technology, there isn’t a robot that can “read the room” and respond.
Procedures
While we are not surgeons… as I mentioned above, a big part of our work still involves procedures. What kind of procedures?
- Procedures anesthesia providers routinely perform:
- Peripheral IV (PIV) access
- Intubations
- Arterial lines
- Epidurals
- Spinals
- Peripheral nerve blocks
- Central lines
- Pulmonary artery catheters
- (With additional training) Transesophageal echocardiograms (TEE)
While there are certainly robots that can be created to perform some of these tasks, there’s also a reason why you have yet to see one employed in the hospital. Patients are unique. They have unique anatomy. Robots/algorithms do better when they are given a set of instructions. Literal instructions.
A great example of how technology may augment, but does not replace humans… the “vein finder.” I once was on a pediatric anesthesia rotation and we had a cute baby with no veins (for those of you with cute babies, the more “prosperous” – in Chinese culture the chubby babies – your child is, the harder it is for us to find a vein!). We used a vein finder device and placed a catheter… in an artery. Since we’re human, we did not inject medications through this vessel (which is dangerous)… we identified that it drew back too easily and tested our hypothesis that it was an artery rather than a vein (hooked up the catheter to a pressure bag and confirmed a systemic pressure). So… it’s a good thing that a robot didn’t just cannulate whatever vessel there was and use it as this could have caused harm to the patient (e.g. embolic events, limb ischemia, etc).
Many procedural specialties are somewhat protected for now because of the unique characteristics of most human bodies. While we mostly have similar organ configurations, the patient that arrives with situs inversus (organs are on the opposite side of the body) may confuse the algorithm. It probably could be trained to figure out what to do next, but one thing that is uniquely human is our creativity and ability to create a new solution or apply a unique approach to problems that arise in real time. While AI is “smarter” than a human, I would argue that it is not more creative than a human. It is only able to generate what already exists in its training model.
Also, procedures taken over completely by AI would require the use of robotics to actually do the physical action. As we’ve seen time and time again, it is extraordinarily difficult to get robots to do complex tasks. At this point, automation of warehouses and supply chain things are where there’s growth in the robotics arena, but medical procedures probably will take a bit longer to see actual autonomous devices.
With an increasing use of minimally invasive techniques, there certainly exists an opportunity to train AI to augment human decision making for the best. We will need to conduct studies to really dive deeper into how AI-augmented care impacts decision making and whether it will actually improve outcomes or lead clinicians astray (i.e. if AI is hallucinating, will the human be able to detect the error and correct it or will it lead a human to doubt oneself?).
Liability/regulatory limitations
Let’s say in the future we have an AI anesthesia robot administering medications and managing an anesthetic. If the robot makes a mistake/hallucinates and a patient is harmed, who is responsible? The company? The hospital that decided to get AI robots to replace human anesthesiologists? Who do we sue?
This kind of ties into some of the regulatory limitations. It’s hard to predict in this current political climate how regulations may change. AI companies are looking to make money, so it makes the most sense to tap into the fields where there is minimal red tape, and potential for revenue generation and huge growth exists. I’d argue that with how much regulatory red tape there is in medical licensing and the litigious landscape of American medicine, most companies are not incentivized to tackle replacing physicians performing technical skills or fields that require quick action in response to a changing patient situation.
The specialities that are ripe for AI right now are those with static findings/datasets that a model can be trained on like radiology and pathology.
The other challenge that arises is that studies need to be done to demonstrate that the AI anesthesiologist is better/cheaper than a human. There are a lot of components to anesthesia that make it hard to break down into
Interdisciplinary coordination
The operating room often feels like controlled chaos. There are a lot of people that need to be part of an operation to make it happen. There continues to be work done to ensure that we have systems in place maintain patient safety, but the reality is, when there are unexpected complications or challenges that arise with equipment, there is a lot of coordination and teamwork necessary to manage these types of situations. AI requires a dataset to train from so that it can make a “judgment” and “decision” on what to do next. The issue is that while there are certainly common things that can go awry when working in the operating room, there are also a lot of things that are unique that we face in the operating room that are hard to train a computer with. One example is that in order for certain operations to occur, it requires specialized equipment that also needs to be sterilized and checked for sterility. Often there are new devices that our device representatives know well enough to guide a physician on use. While one could say that all of these “jobs” can be replaced by a robot, I’d argue that robots aren’t anywhere near as cost effective (see below) as our current system of humans. One potential benefit of a robot in the operating room might be that it’s feelings won’t get hurt (Maybe? Do our AI chatbots have feelings and sentience? Probably not, but with how good chatbots have gotten at mimicking emotion, it certainly can seem that way.)
Human touch
Admittedly, it is heart-warming to read about the ElliQ robot companion that was designed for older adults. There’s a real loneliness crises among older adults (and probably younger adults post-pandemic). Loneliness is associated with cognitive decline, so if this device provides companionship for our older adults, why not use it as a way to also combat cognitive decline/dementia?
That being said, most people don’t get surgery/procedures everyday. For most people, this is one of the most vulnerable moments in their lives. If society gets to a point where humans have AI robot companions, I could see that helping in the perioperative setting with calming anxiety and providing company. Currently, I do think most patients would feel more reassured by knowing there is a human caring for them and that human connection/trust is vital to the surgical journey.
It’ll be interesting to see how this evolves over time. More people are falling in love with their AI chatbot so…maybe the time is nearer than I thought for humans to accept robots caring for them.
Cost
As I type this up in 2026, I’d say that human labor is probably still cheaper than an AI-powered anesthesiologist or ICU doctor. There are certainly things that AI will be “cheaper” in being able to do (administrative things, coding, summarizing, etc) but the reality is that the physical labor, critical thinking, and variety of skills required in order to provide anesthesia and attend to an ICU are not replaceable by a single robot. There are many pieces of what I do in my work that could be augmented or even done by AI (like my pre-op job – more on that below), but the combination of all of the things would cost a lot more money than I cost at this time. Maybe someday the cost of the technology will be less than a human, but for now, it seems like we’re safe.
Areas of anesthesia that AI can/will influence/impact
The areas of anesthesia where I think AI makes sense:
- Preoperative assessments
- Early alert warning systems
- Drug dosing and delivery systems
My leadership role is in the realm of pre-operative evaluation – reviewing a patient that is scheduled for surgery and assessing whether they are optimized or not to proceed. This area is absolutely an area where I WANT AI to come and help summarize the incredibly bloated medical record. Now, the technology is only as good as the data that’s fed into it, so I anticipate that humans will still need to validate the summaries that are generated and review the citations (links to the original note that it extracted information from) that it includes. One recent example of where summarizing a patient’s chart could be erroneous without a human: I saw in a single note that reported that the patient had a specialized device implanted, but then noticed that there was no evidence of the device in the patient’s imaging. So I called the patient and it turns out that the patient’s partner, not the patient, had this device implanted. We will never know whether this was a scribing error or a human error with note entry… but this is an example of how human judgment, instinct, validation, and oversight will still be necessary in this realm.
Another area in the anesthesia realm that I think AI could really make a positive impact on patient care is with early alert warning systems. The anesthesia record is ripe with data that, if analyzed, could really create prediction models for early signals for adverse events. The most common reason for hypotension in our world is unanticipated bleeding or under-resuscitation during a case. I believe that AI should be applied to analyzing the dataset of anesthetics to be able to provide a warning indicator for when a patient may be at risk (e.g., analysis of arterial line pulse pressure as an early indicator for hemorrhage). Now… this requires a high index of suspicion from the anesthesia team to place invasive monitoring like an arterial line, so we (humans) still need to stick around for a bit. And we still need a human to respond to the warning system.
One of my colleagues is working on a device to provide suggested dosing and to also optimal evidence-based recommendations for maximum dosing of local anesthetics (which can lead to a dangerous complication called LAST if over-dosed). Using AI technology for decision support is another area that I’d welcome a computer that’s programmed with good evidence/data to make suggestions for optimal dosing. In our world of anesthesia, we administer a lot of medications… some medications should be dosed by actual body weight others ideal body weight and sometimes we should use adjusted body weight. Training helps us learn which one is which, but with new medications that get added to the market and new studies on optimal dosing, it’d be amazing to leverage AI to incorporate this information in a way that is easy for the user to apply to a patient. Truly using evidence-based, precision medicine would be an ideal end goal!
Target-controlled infusion anesthesia delivery systems already exist in the market. Although these devices are more widely used in Europe, there are limitations to the technology and there are guidelines surrounding their use cautioning around its application to patients outside of the studied in the patient model (i.e., healthy, non-obese patients). This is a great example of a technology that attempted to automate an anesthetic… but looking at the error margin that exists (the devices’ mean estimated concentration of the anesthetic could be 25% from the actual) and the fact that the vast majority of my patients do not fall within the validated model (I don’t take care of healthy, non-obese patients frequently), make me rather skeptical of its applicability to my personal practice. That being said, these devices certainly are a ripe area for exploration and application of AI models. However, these systems would make more sense in an ICU setting first rather than in the operating room where there will be more dynamic changes to the patient’s status depending on the surgery and potential complications of the surgery.