Scientists recently realized that AI could lead to medical research going astray.

Home > Explore

Scientists recently realized that AI could lead to medical research going astray.

2019-02-18 18:32:53 227 ℃

According to the Financial Times, a group of top scientists and medical statisticians warned on Friday that the use of artificial intelligence in some biomedical fields would lead to inaccurate conclusions.

< p> "Many of the research conclusions from the analysis of large data using machine learning technology can not be trusted by me." Genevera Allen, an associate professor at Baylor College of Medicine at Rice University, warned at the annual meeting of the American Association for the Advancement of Science. Machine learning has been used to study the relationship between scientific and medical data and certain phenomena, such as the association between genes and diseases. In precision medicine, researchers look for patients with similar DNA, so that treatment programs can target specific pathogenic genes.

"A lot of technology is for forecasting." "But I've never come back to the conclusion that I don't know or that I didn't find anything because they didn't consider it in their design process," Allen said.

She was reluctant to point out specific cases, but said that the research conclusions of machine learning on cancer data were good examples.

"There are many cases that cannot be repeated." "Clusters found in one study are very different from those found in another," Allen said. Why does this happen? Because most machine learning technologies today say,'I found a group. But sometimes it's more helpful to put it another way: I think some of them are really grouped, but I'm not sure about others.

Once machine learning discovers that there is a specific link between the patient's genes and disease characteristics, human researchers may provide a reasonable scientific explanation for the corresponding findings. But that doesn't mean these findings are correct.

Allen said, "You can always find reasons why certain genes are grouped together." Computer scientists have only recently begun to realize this problem, which may lead medical researchers to take the wrong path and waste resources to confirm results that cannot be repeated.

Allen and her colleagues are working to improve statistical and machine learning techniques so that AI can criticize its own data analysis and point out how likely some discoveries are real rather than immediately relevant.

"One idea is to specifically scramble data to see if the results will remain unchanged." She said.