Here are our key takeaways from the presentations that were given on day two of AIMed, the conference on artificial intelligence (AI) in Medicine and healthcare hosted lasted December:
- Forget killer robots, bias is the real AI danger. Algorithmic bias is already creeping into some industries. Social media algorithms are shaping users’ perceptions of new events. Through carelessness we can introduce immoral behaviour into automated systems.
- Due to time pressures, on average a physician will interrupt a patient describing their symptoms within eighteen second. In that short time, many doctors decide on the likely diagnosis and best treatment. Performance gets worse with the more information that a patient gives. You need less information, but higher quality of information – here AI could help to sort through an overwhelming amount of info.
- You should assume that every kind of image a human looks at, including CT scans, a computer could look at more accurately and quickly. Medical imaging through AI should not be looked at as something weird or controversial. It has been years since it was definitively proven that an AI system is better at looking at a CT scan than a human is. The big question is: how do you get image recognition more quickly and effectively into the clinic.
- Dr Anthony Chang compared AI in medicine at this point to a GPS: you don’t have to use it, but why would you ignore it if its there? And he repeated his prediction from the last conference: that although today a doctor could be sued for using AI to treat a patient if something goes wrong, in 10 years a doctor could be sued for not using the commonly accepted AI solutions for practitioners.
- Jeremy Howard, the founder and deep learning researcher at fast.ai, said that in his opinion black box systems (AI systems which don’t explain why they make the decisions they make) don’t actually exist. He said you can lay meta algorithms on top of any AI algorithms to reveal what was important, so there is no such thing as a system which you can’t explore and fully understand.
- The speakers gave predicitons on which year artificial general intelligence (AGI), which is AI that has equal or greater intelligence than a human, would arrive. None of the speakers thought that AGI was impossible, and most of them said it would emerge in the next 10 years. Mike Galvin, VP of Next Gen at Tata Communications, said that AGI will not occur in the US, but in a developing country where nobody expects.
Shark Tank event: The day ended with a startup ‘shark tank’ event, where entrepreneurs pitched their startups to six venture capitalists and were grilled on their solutions, business models, and strategy. Here is the advice that the VCs passed on to entrepreneurs:
- Key takeaway is that companies who are creating technology for the healthcare industry need to think more about who is going to want to buy their solutions, spend time understanding their customers’ companies, try to build a full team rather than being a ‘jack of all trades’.
- The big opportunity for entrepreneurs is to work on very hard problems rather than just obvious problems and think about shifting, rather than going faster or harder on existing problems, to new problems that are not currently being tackled.
- Finally, don’t underestimate competition, even if its pen and pencil vs AI, there is always competition and there is always another way to do things.