Another Thing AI Can Do Now? Diagnose Cancer Better Than Doctors
BioMind still has to prove itself in everyday practice, but the results are encouraging.
PHOTO CREDIT: Getty Images
If I say the words "artificial intelligence" to you, you probably think of Telsa's self-driving cars or Netflix's recommendations. But close your eyes. Picture a world where AI conducts tumor analysis, diagnosing cancer with greater accuracy than highly qualified doctors.
Ok, now slap yourself awake. The world you dreamed and the one in front of your eyes should match, because as Xinhua reports, the AI system BioMind already has defeated a 15-member team of Chinese doctors in cancer diagnosis.
Say hello to BioMind
Developed at Beijing's Intelligence Research Centre for Neurological Disorders, BioMind is the product of thousands of hours of careful image training. Doctors used images from the Beijing Tiantan Hospital archives to help the AI learn what neurological diseases looked like.
The results are encouraging. BioMind can diagnose brain tumors with an accuracy of 87 percent, compared to a 66 percent accuracy from doctors. It was correct in predicting brain hematoma expansion at 83 percent, compared to 66 percent from doctors.
The high rate of accuracy is important for a few key reasons.
- It can improve how soon professionals can get you started on appropriate treatments. In many cases, catching cancer earlier makes a significant difference in the odds of surviving the disease. It can also mean fewer, less complicated procedures and, therefore, higher quality of life.
- Greater accuracy can reduce unnecessary suffering, ensuring your doctors don't give you unnecessary treatments that won't help.
- Doctors can put more time into research that finally will cure cancer once and for all.
On top of these benefits, BioMind also can offer up a diagnosis faster than your qualified doctor. A human professional typically takes about half an hour to do the job, but BioMind can give you an answer in half that time. The less time doctors spend in diagnostics, the more they can spend on face-to-face patient care. That interaction fosters the trust that has been shown to boost positive patient outcomes, and it means the doctors can get additional, patient-provided data that yields a more holistic picture of what is happening.
It's not just medicine where AI could let people off the hook
AI isn't going to take all human jobs. (Relax, OK?) But BioMind is just the latest example of areas where AI already outshines us. For example, AI does better than we do in
Language--Professionals now can use AI that transcribes audio as well or better than human transcribers. Lip reading is also better through services such as LipNet. These systems have strong potential applications not only in general business and security, but also for those with disabilities. Notedly, translation is still very much an AI work in progress, although there are some good AIs that are helping people acquire new languages quickly.
Playing games--Whether it's video games, Jeopardy, poker or Go, various AI systems such as Google's AlphaGo have come out as the winner in games that require strategy, risk and the application or retrieval of information.
Completing tasks in dangerous environments--Engineers have created AIs that can unload equipment, search for specific objects or even defend territory. Those advancements mean that science can advance even as we minimize loss of life.
Hiring--Companies like Unilever are introducing AI systems that dramatically reduce the number of candidates hiring managers need to consider. There are still legitimate concerns about how to make these systems free from bias, of course, but in terms of simply reducing the piles of resumes executives have to read through, AI is highly efficient.
A few grains of salt on that huge plate of awesome
BioMind and the other examples above show we need to be respectful of the power AI already has claimed. But there's also evidence that AI doesn't always live up to the hype it initially flies in on. For instance, consider the IBM Watson system. Just a few short years ago, like BioMind, experts shoved it in the faces of oncologists. They told physicians that the system would be wildly helpful in recommending treatments, particularly for lung cancers. The results, however, have been underwhelming, particularly since the system cannot be changed quickly to accommodate a shift in human-based standards of care. And more recently, Stanford researchers developed an AI meant to help diagnose skin cancer. That AI, which relies on image training similar to BioMind's, still has to prove itself, too.
BioMind might follow Watson's path--or it might not. The comparison, however, demonstrates that we cannot necessarily stick a "results guaranteed" label on these types of technologies. We can dream that they surpass us and allow us to do good, and we can be grateful when they actually deliver. But predictions can be inaccurate. So think with your head even as you hope. Stay grounded. Be patient and test the hades out of the ideas. And if it turns out that Human Betty in Cubicle C can do the work better, just cut her a check and get it done until the tech finally catches up.
BY Jeremy BerkePeter Kotecki and