Technology will replace 80% of what doctors do – Vinod Khosla
Following the climate panel event on April 5th was a panel on another hot topic: The Robot Will See You Now: The Revolution of Artificial Intelligence in Medicine, held in the University of Toronto engineering building.
I forgot to write down the names of the five panelists, but they included a bio-ethicist, an AI researcher, the CEO of a medical AI company and two physicians.
Among the insights they provided were:
– Public consciousness of ‘AI’ (for those who know anything, it’s usually what was gleaned from the movie Terminator or from a viral robot video) is often disconnected from what it looks like in practice (typically boring data-processing and pattern-recognition)
– There are a number of areas in medicine where repeated human error (miscommunication between patient & doctor, misdiagnosis, getting swayed by pharmaceutical advertisement, etc.) is causing thousands of people to die every year (40,000 per year in the U.S. due to misdiagnosis alone). This creates a strong ethical case to bring in more advanced technology, such as AI, to save lives
– Currently, the main barrier to more use of AI in medicine is institutional: in the U.S. the FDA has not approved any of the more avant-garde systems that significantly change current practices
– So what are the main risks with using AI in medicine? There are a number of them, but right now, most relate to liability (who takes the blame if it fails?) and data (who has access? what is it used for?)
An interesting example of this was in Iceland, where a team of researchers got permission from a sizeable percentage of the population to map their genetic code, and then went on to calculate everyone else’s (who hadn’t given permission, and who pressed charges). The Supreme Court of Iceland eventually ordered the team to destroy all their data. A U.S. team researching rare genetic conditions also had to destroy the DNA data they’d painstakingly collected from thousands of patients, because even mapping part of someone’s DNA can lead to their full sequence (and that of their relatives) to be easily (and illegally) traced and used in unauthorised applications
– The biggest uses of AI so far have been in seemingly mundane applications, such as:
-Converting free text from medical reports (which can often contain a dozen different ways of describing the same illness) into structured data that can be easily searched. For those who had to deal with this problem, AI has been a lifesaver
-High volume processing (of medical samples in labs, or filtering cancer reports)
-Speech & image recognition in telemedicine
– The last speaker, a primary care physician, brought in some interesting tidbits from the front lines. First, that technology has helped reduce to massive amounts of memorisation previously required of doctors. Second, that the electronic medical record, while essential, also regiments the doctor/patient relationship into a more formal and contractual one. Third, that technology has also become a nuisance in terms of patients ‘self-diagnosing’ on Google and expecting to know better than the doctor. Finally he had a word of caution about the limits of AI in diagnosis: for it to work, there has to be a single ‘right’ answer, which isn’t always the case in the real world. At the end of the day, the physician will always be responsible for the accuracy and application of the diagnosis.
– On a closing note, one thing that all the panelists agreed on was that despite its current challenges, medical AI is a booming industry likely to play a prominent role moving forward
I left the talk feeling both better informed and bit dissatisfied. A bit like climate change, this is a slow-moving, incremental, tidal wave of change.