Seems ChatGPT is starting a cult.
I think Bryan Johnson may have already gotten that same message
Alex Kantrowitz podcast interview with Microsoft’s “AI CEO” Mustafa Suleyman on their recent breakthrough medical diagnosis system:
Their new system can diagnose better than doctors, where the test involves not just look at medical charts, but also interviewing the patient so that it can drill-down, if need be. He says that in some of their tests human doctors were only 20% accurate in reaching a diagnosis while their AI was over 85% accurate.
He mentions how they test their system on fresh mystery cases to diagnose that appear in The New England Journal of Medicine – like an “ultimate Crossword for doctors”, as Suleyman says – that couldn’t have been in the training data. So, good performance on those shows that the system is able to generalize.
As with many AI results like this, one needs to take it with a grain of salt and ask, where do humans do better (if at all)? But overall it looks to me to be a really promising advance!
I’d say Bryan is correct-directionally correct. I like his ideas and what he’s popularizing.
Robot surgery on humans could be trialled within decade after success on pig organs
AI-trained robot carries out procedures on dead pig organs to remove gall bladders without any human help
The surgical robots were slightly slower than human doctors but they were less jerky and plotted shorter trajectories between tasks. The robots were also able to repeatedly correct mistakes as they went along, asked for different tools and adapted to anatomical variation, according to a peer-reviewed paper published in the journal Science Robotics.
The authors from Johns Hopkins, Stanford and Columbia universities called it “a milestone toward clinical deployment of autonomous surgical systems”.
…
“We were able to perform a surgical procedure with a really high level of autonomy,” said Axel Krieger, assistant professor of mechanical engineering at Johns Hopkins. “In prior work, we were able to do some surgical tasks like suturing. What we’ve done here is really a full procedure. We have done this on eight gallbladders, where the robot was able to perform precisely the clipping and cutting step of gallbladder removal without any human intervention.
“So I think it’s a really big landmark study that such a difficult soft tissue surgery is possible to do autonomously.”
I would guess if it starts to look like there is serious money to be made in this in the not-too-distant future, AI companies will swoop down with large amounts of money and try to do it all themselves. (Something to watch to see what this looks like going forward is OpenAI’s effort in robotics – e.g. there are rumors they have made a breakthrough in this, and in an interview Sam Altman offhandedly said, “I think we have some new technology that could just do self driving for standard cars way better than any current approach has worked.”)
As a robotic surgeon, we are a long ways from AI doing robotic surgery on humans. Tech and AI progression is exponential but I can’t imagine AI robotic surgery on humans within 10 years.
The authors and researchers who put together that article title are an optimistic bunch. I think regulations will greatly slow down tech like that. Anything that can potentially kill someone with the slightest error will have to pass a lot of tests until it gets approved in human use without direct supervision.
However… it would not surprise me if the technology can actually do what they claim within just a few years. Researchers are starting to crack general robotics in the real world in unseen environments:
Previous robot demos like this (for example from Boston Dynamics, Honda’s ASIMO, or Moley Robotics) were highly scripted, and did not have the ability to complete general tasks in unstructured environments like in the above video. In fact, even just a year or two ago robots were nowhere near as impressive as in that video for general tasks. It’s a bit frightening to imagine where it will be in even just one more year.
I agree. regulation will be huge. Also liability. Who’s responsible when the robot makes a mistake and the patient is harmed?
In regards to the gallbladder surgery, it was a dead pig so the robot didn’t have to deal with bleeding, movement from respiration, smoke from cautery. Bleeding will be huge because often in surgery you accept mild bleeding knowing it won’t cause problems and you are still able to dissect a somewhat bloody field. This technique can even lead to superior outcomes when you are operating on vital structures that shouldn’t have cautery used adjacent to them. Knowing what bleeding is problematic and should be stopped vs what is better left alone is knowledge gained only through experience and moment to moment decision making. I’m probably underestimating AI, but I just don’t see it making that decision correctly and as accurately as an experienced surgeon. At least not anytime in the next 10 years. Also, a gallbladder for an experienced surgeon is often a <30 minute case. Plenty of our surgeries can be hours long with infinitely more surgical moves. Again, a huge task for a non-human to complete.
I’m using a new AI enabled web browser which I’m finding interesting… you can easily have any video analyzed while you are on that video, and sumarized (without having to open a new window and go to ChatGPT or other LLM…
You can try the Beta version here: