A statistician says machine learning is hurting science

Feb 18, 2019, 7:05 AM EST
(Source: Marco Hazard/flickr)
(Source: Marco Hazard/flickr)

Dr. Genevera Allen, a statistician from Rice University in Houston, has red flagged the wrong approach of using machine learning techniques in analysis of scientific data sets, saying that the algorithms pick patterns that exist only in data and not in the real world.

Dr. Allen said that flawed machine learning techniques are creating a “reproducibility crisis” in science studies across all areas, resulting in complete wastage of time, money and efforts used in collecting huge data sets, notes the BBC.

Dr. Allen cited the inaccurate results of a number of studies, which were derived with machine learning analysis but failed to stand the test of time when other researchers repeated the same experiment with different datasets, writes Digital Trends.

She pointed out that the researchers have been overly keen on the use of predictive models without caring at all for their accuracy. The trend is hurting science, and this could be corrected by developing next generation machine learning and statistical techniques that can self-report the uncertainty in their findings.