Inventor: "We programmed the best surgeon’s techniques, based on consensus and physics, into the machine"
Third Party: “While in a technical sense, semi-autonomous suturing is a ‘grand challenge’ problem of surgical robotics, clinically much suturing and bowel anastomosis is done by staplers which can do the whole thing in seconds,” he wrote in an email. “Clearly the task they chose does not justify the elaborate equipment they used.”
So, they used machine learning to perform a limited part of a surgical procedure, but with freedom of motion and adapting to tissue changes.
At the end of the article they talk about self-driving cars and how we have lane-assist and the next level is taking over. The inventor likens this technology to current lane-assist technology, but it seems more involved than that.
Personally I do not want any machine learning consensus algorithm with control over a surgical procedure in my body. I think the DaVinci type systems (tele-operation, but without autonomy) are awesome -- they allow superhuman precision in delicate surgical tasks.
However, until AGI is a solved problem I do not want an automated algorithm making decisions with a knife to my organs.
Let the automation algorithm highlight portions of the screen or anticipate the next viewpoint, but keep it away from the controls.
I'm with you. The sheer amount of human error in the medical field that causes death and injury is astounding [1], and I would gladly put my life in the hands of a capable machine instead of a fallible human if given the choice.
I would already be happy if medication for a patients wouldn't be handled by passing paper notes from one nurse to the next. There is no fancy machine learning needed to make sure that you get what the doctor prescribed in the correct dose.
Where do they still do that? I live in a developing country (some people call it third world) and the private hospitals here have been using all electronic medication ordering and fulfilment for quite a while, at least since 2009[1].
Doctor enters the prescription into a networked computer. It goes to a robotic pill picker/packer that affixes the label with instrucitons. Outpatients pick up the medication at the desk on the way out. For inpatients it is sent to the nurses' station on the floor of their hospital room.
It's been a while since I've been to a hospital in the US. I just assumed they were already doing this.
It is a big change from the traditional shift change in many hospital units, where nurses going off duty typically confer in a hallway or at the nursing station with the nurse coming on for the next shift, giving a rundown of their patients’ status and needs. In some cases nurses may simply write up a report in the medical record for the next shift to read.
It also 'helps' that EHR software has been forced onto hospitals at such an early point. My father constantly checks what orders are given to a patient now because one time, an order he placed at a patient's previous visit to the hospital was placed at the current patient's visit instead, almost two months later. Luckily the screw up was just with food assigned to the patient (high fiber diet vs liquids), but imagine if it was something like a drug being prescribed.
I work with EHRs. Patient context bugs (patient number, visit number, chart number, encounter number, or any combination thereof) don't come up THAT often but it seems like everyone has them. It's actually surprising the occurrence is low since most of them are giant piles of WinForms spaghetti by now.
But then again the software in those machines will be written by humans and will most likely have bugs - which might influence the outcome of the operation. So the questions is whether you trust more the surgeon or the engineers of that machine.
As an earlier poster said, it is nothing to do with "trust". Software has bugs, surgeons make mistakes. What matters is the relative degree of statistical risk.
If the risk of falling victim to a bug/mistake by a software surgeon is lower than that of a human surgeon, then simple desire for self-preservation you should choose the software. To choose the human over the software, even when it's provably more dangerous is allowing a baseless prejudice to put you at higher risk. It's equivalent to being given the choice between two human surgeons of differing skill levels, and choosing the lesser skilled surgeon on the basis of their ethnicity.
How about a surgeon that's been up for 36 hours, hungry, with a splitting headachy, and some family problems on the mind. A human surgeon might be awesome and can still make a ton of mistakes for reasons unrelated to their medical abilities. Meanwhile, while a programmer could easily make a mistake, other people and tests of the code could spot a lot of mistakes that could be corrected before the system reaches its first patient.
I'm going to take a different tack: How about we stop hazing medical professionals at the beginning of their career and only ask them to work 8 hour shifts with an appropriate amount of time off until their next shift?
Along with lmm's comment, where would all the extra doctors come from? You can't cut hours worked in half (or close to it) without doubling your staff to meet the demand. US hospitals (especially ERs) are already running over patient capacity in many US cities
Simple, make the hell of residency go away, and allow more foreign doctors to come in to US and practice after going through a reasonable residency program. I have known very smart and competent Russian and Ukrainian doctors who work as lab employees because they are not allowed to practice medicine here.
Great idea. But how do we get to there from here? Particularly when doctors' professional bodies are large, powerful, and full of people who are convinced that that never did them any harm?
As a software developer who has been stitched back together by an MD who has been on shift for 36 hours, I'm with the machine also. No relevant paternal affiliation.
My guess is that the software will be better at the well know tasks. And similarly to automatic cars, they will struggle with the unusual scenarios. (Saying that I don't know a great deal about surgery).
I'd argue that a surgeon's time is wasted doing grunt engineering work, they are better as project managers and test engineers because they can come up with tests to prove it works better than the engineers. It's rare for someone to excel at two scientific fields, and you'd need excellent engineers for that kind of project.
I don't think you'll see fully automated surgery soon because most ML algorithms can too easily be fooled by unusual sensor data or unexpected states -- and these are very much AGI problems as well. Those algorithms are no where near the reliability levels required for approval as medical devices. We will have advanced assist-mechanisms (like davinci robots) long before the decision making processes are incorporated into a robot surgeon. Essentially, there won't be a "computer surgeon" for a long time and the difference between Human Surgeons using modern assist technology vs a fully computer surgeon mortality will be blurred as assistance technology improves. So there will never be a 1% vs 0.5% mortality. It will be equal mortality at best before AGI allows improvements under as many or more conditions as a human. Just my opinion of course.
I would guess the automatic parts will be like a toolkit of tasks which the surgeon can delegate to an algorithm. You could imagine that an algorithm could sew skin together with a level of accuracy to prevent scarring, rejoin nerve and small blood vessels quicker and more accurately, or remove cancerous tissue tagged with a biomarker, while avoiding nerves or important structures nearby.
The surgeon could put the surgery into the controlled situation that the algorithm understands, and make some over-arching decisions through a UI (for instance, he could say 'join this nerve on this side to this nerve on the other', which might be difficult for the algorithm to work out), and then the algorithm would accomplish that task, and return control back to the surgeon.
More simply, a program could prevent a human surgeon using a machine like a Da Vinci from touching a nerve, or define the limits of bone removal.
The latter example does already exist in clinical use:
That may never come. The best-performing machine learning algorithms work by optimising a measure of error. Experience says that while it's easy enough to take this error to something between 10% - 30%, it's excruciatingly difficult to push it any further down.
see, I'm with you WHEN that's the comparison for the particular surgeon who would be performing my procedure. Not the average of all surgeons vs the computer, since that stat is not necessarily applicable to my situation.
You might hope the be the special snowflake who gets the heroic surgeon, but I wouldn't bet on it. Unless you are medically interesting in some way, you're likely to just get the next schmoe on the rotation.
> However, until AGI is a solved problem I do not want an automated algorithm making decisions with a knife to my organs.
AGI isn't yet discovered yet we allow planes to fly themselves, markets allow automated agents to spend billions and so on. What if the algorithm was proved to be 10x safer than a human expert? Even without AGI they could be better than humans.
> Personally I do not want any machine learning consensus algorithm with control over a surgical procedure in my body
Isn't the surgeon's brain a machine learning algorithm in physical form? Humans are more versatile today and can adapt do different situations, but I don't see this as a argument against machine learning necessarily.
The book Complications, by Atul Gawande is a fantastic account of medicine as an imperfect science that I would recommend to anyone interested in the field.
Things like this make me hopeful about perfecting medicine and surgery to avoid common human mistakes--even if that is still decades away.
in an almost the same vein, i would heartily recommend 'the medical detectives' for bizarre accounts of medical mysteries and once 'debugged' almost 'obviously that's how it would have happened' explanations of those incidents...very cool :)
> In about 40 percent of its trials, the researchers intervened to offer guidance of some sort—as in the GIF above, where a human hand is seen holding the thread. In the other 60 percent of trials, STAR did the job completely on its own.
Did the robot open the patient? Did the robot locate and section the organ? This is a surgeon as a sewing machine is a tailor. It's a tool. To call it a surgeon is very misleading.
The robot did something that previously took a very highly trained person to do. It's a huge accomplishment and not misleading at all.
You may not have read the article because it explicitly mentions the system you linked to:
> The current state-of-the-art robot for soft tissue surgery is the da Vinci system from Intuitive Surgical, but it’s not automated at all. The da Vinci is a teleoperated system, in which the surgeon sits at a console and manipulates controls in dexterous maneuvers that are mimicked by tiny tools inside the patient’s body.
I agree with the GP, calling this a surgeon is misleading and looking for hype. Yes, it's a great accomplishment. No, this is not a surgeon.
You can start calling it a surgeon when it can autonomously perform an appendectomy, including handling common complications (excessive bleeding, situs inversus, etc) during surgery.
As a programmer who used to work with the development of training simulations for surgeons, I find this tremendously exciting.
The company I used to work for focused on using computers to improve the state of surgery, primarily laparoscopy, by using virtual reality to train surgeons. This is an alternative approach, teaching the computer to perform the procedure itself, rather than to evaluate its performance.
As a side note: I actually spent a fair bit of time exercising with modules for suturing, including specifically for the small intestines, and it's definitely quite fiddly work.
>> Its vision system relied on near-infrared fluorescent (NIRF) tags placed in the intestinal tissue; a specialized NIRF camera tracked those markers while a 3D camera recorded images of the entire surgical field. Combining all this data allowed STAR to keep its focus on its target. The robot made its own plan for the suturing job, and it adjusted that plan as tissues moved during the operation.
>> The researchers trained STAR only on how to perform this particular intestinal suturing procedure. “We programmed the best surgeon’s techniques, based on consensus and physics, into the machine,” Kim said.
So it's a combination machine learning - expert system robot. Machine learning (I assume Convolutional Neural Networks) for vision, some planning algorithm for the suturing job and expert knowledge for the surgical techniques.
That's the shit I'm talking about! Let's see more of that! Don't just end-to-end train some deep net and try to have it learn to do everything from scratch. Use background knowledge! Combine techniques! Be smart, dammit!
Edt: btw, this sort of thing, systems with hard-coded expert knowledge besting human experts, that's age-old stuff. It's how it used to be done before the last winter (AI winter). Nice to see it back.
Edt 2: Apologies for the ex!cla!mation! marks! but I'm so! excited!
Ho-hum. After reading the paper, they didn't use anything like machine learning or expert knowledge. They just kludged the whole thing step-by-step into the robot. Its "planning" was to draw a line between the NIRF tags on the surface of the pig gut.
So it's not even properly autonomous- they're just misusing the term to mean that nobody was remote-controlling it.
Yeah. There have been spontaneous demonstrations of joy and gratitude for our happy new way of life. Said no-one who's job just got automated. Automating work out of existence may or may not be a laudable goal. Assuming it is we've yet to see a roadmap presented that doesn't brick the economy somewhere in the middle of the gap between now and full automation.
There is no roadmap. It'll all be organic (unfortunately?). Its up to us to put guardrails on to ensure people who are automated out of a job can still find purpose and meaning in their life (UBI or some other method to ensure no one goes without housing, food, clothing, healthcare, etc).
EDIT: If it goes south, I assure you, I'll be one of the first to devote the rest of my life to getting humanity back on track with the distribution of resource and knowledge wealth (not fiat currency "wealth" mind you, cause that isn't going to be worth much compared to raw resources and automation knowledge).
Automating work will probably be a good thing in a vacuum. Unfortunately it's happening in an old economic system established during a period of scarcity of labor and capital that rewards and encourages things that will not be very important anymore.
We really need to be thinking about how to distribute the benefits of automation but right now our answer is only those people who already have capital will benefit from it.
Do you long to return to the time when more than 50% of the US labor force had to work in food production? Thanks to automation, less than 2% of our population is employed to feed us all, freeing the rest of us to develop software or drive trucks.
Funny that you mention starvation wages. 50 years ago, you would have had to spend twice as much on food and you wouldn't have gotten anything at the level that's available today. I'd encourage you to read some of Megan McArdle's posts on how we look back at historical food preparation with glasses that are extremely rose colored.
50 years ago is only the 1960's. Food was more expensive then in no small part due to the fact that production was mostly by family farms. The economies of scale that have wiped out rural communities around the country had only just begun to develop in earnest. This meant small family farms were still competitive in the market and could enjoy a high quality of living. I welcome an explanation of how driving family farms out of business with vanishing margins and concentrating what little profits remain in the hands of massively capitalized corporate farming outfits is a net win for society. Note: cheap food is ripe for all of the standard criticism normally aimed at trickle-down economic theory so plan accordingly.
Also, define "food", because we can certainly devote some time to discussing processed foods.
"Automate things" is an incredibly coy way of saying "destroy other people's jobs" unless or until all of the magical information economy jobs politicians and industry mouthpieces have been promising for the last 30 years start to materialize.
No. There won't be more "information economy jobs". You need to index the work week, wages, and entitlements to productivity. As less labor is required, we ask less of labor and pay it more for what work it does do.
Human surgeons will supervise the semi-automatic robot surgeon, but instead of having maybe 2% great surgeons and 98% mediocre ones, we will have 100% great surgeons. The role of the human will be to integrate information from other AI tools, like, for example, ones that analyze the DNA, blood work, or propose diagnosis and direct the robotic arms to the selected procedures. It will be very high tech to be a doctor, for sure. And people's expectations will immediately inflate and take it for granted in 20 years.
The thing that everyone seems to be missing is that this will lead to parallelism and miniturisation. One day we will get to the point where pinpoint sized cameras and auto suturers will be able to stitch up incisions without scars. And maybe a few decades after that we will get surgery at the cellular level. Imagine a robotic surgical tool that only cuts tumour and can avoid non cancerous cells and membranes.
> a bot stitched up a pig’s small intestines using its own vision, tools, and intelligence to carry out the procedure. What’s more, the Smart Tissue Autonomous Robot (STAR) did a better job on the operation than human surgeons...
I'm curious if they took into account how a person will do a better job when said job is important/risky to themselves. I. E. stitching up a human where a mistake has dire repercussions vs stiching up a bit of desposable flesh.
Professional pride, along with "this is unusual and I am being graded on it", can be a significant motivator. I'd actually be more worried about the surgeons being tested being under too much stress, not too little stress.
Yes the pig was alive - "in vivo" means on a live subject.
I'd expect a robot + human intervention to be strictly better than a human alone. Surgery is a slow static kind of operation so if anything starts to go in the wrong direction, the human can press pause and intervene. Then sit back and let the robot continue after correcting its actions. That's quite different from a driverless car where decisions have to be made quickly and you can't just pause to stop and consider which way to steer in an emergency.
Posed human vs computers, like other topics, but I'm starting to see reputation as more important. A quality company is better than a dumb one, regardless of internal mechanism. Some ai-based orgs will be ome the fast food of their industry. And your choice won't be ai vs human but you will shop reputation/quality vs price. And like our diets accomodated automation, so will all our choices.
This is another thing, where a sort inverse domino effect exists. The more the machine learns, the less training the human gets, the faster the conversion spreads. Same as why you trust a machine to fly you more by now then a human.
Don't worry everyone, we won't be seeing robotic surgeon's anytime soon, until the next time insurance companies look to enact cost cutting measures. Best guess, in the next few years. If you can still afford insurance then...
These are symbiotic. The research and development pays off because somebody else is willing to ruthlessly buy the new, cheaper/better option instead of the safe traditional choice. Mainframes gave way to minicomputers which gave way to microcomputers which gave way to laptops and now to mobiles.
There are many tasks that machines do today that humans will never outcompete again, and they do it at a superior intersection of benefits and costs. Using a heuristic like that is too lazy and brittle here.
Inventor: "We programmed the best surgeon’s techniques, based on consensus and physics, into the machine"
Third Party: “While in a technical sense, semi-autonomous suturing is a ‘grand challenge’ problem of surgical robotics, clinically much suturing and bowel anastomosis is done by staplers which can do the whole thing in seconds,” he wrote in an email. “Clearly the task they chose does not justify the elaborate equipment they used.”
So, they used machine learning to perform a limited part of a surgical procedure, but with freedom of motion and adapting to tissue changes.
At the end of the article they talk about self-driving cars and how we have lane-assist and the next level is taking over. The inventor likens this technology to current lane-assist technology, but it seems more involved than that.
Personally I do not want any machine learning consensus algorithm with control over a surgical procedure in my body. I think the DaVinci type systems (tele-operation, but without autonomy) are awesome -- they allow superhuman precision in delicate surgical tasks.
However, until AGI is a solved problem I do not want an automated algorithm making decisions with a knife to my organs.
Let the automation algorithm highlight portions of the screen or anticipate the next viewpoint, but keep it away from the controls.