Shulou(Shulou.com)05/31 Report--

Benshaw found that many friends are interested in information about how racial prejudice affects the effectiveness of artificial intelligence algorithms. Benshaw collated and shared some information about how racial prejudice affects the effectiveness of artificial intelligence algorithms.

According to a new study published in the journal Science, researchers have found that commonly used medical systems can lead to racial bias in patient care decision-making algorithms.

The study's authors explain that the algorithm has been deployed everywhere to assess the needs of patients.

"large health systems and payers rely on this algorithm to position patients as' high-risk care management 'programs," Ziad Obermeyer, MD, of the School of Public Health at the University of California, Berkeley, and his colleagues wrote. These projects aim to improve care for patients with complex health needs by providing more resources (including more attention from trained medical staff), thereby helping to ensure good coordination of care. Most health systems see these plans as the cornerstone of population health management, which is widely believed to effectively improve results and satisfaction while reducing costs. "

When studying the algorithm (Obermeyer and others did not specifically track the game when studying the algorithm). It is found that the forecast focuses on medical costs (such as insurance claims) rather than medical needs. Black patients have "lower medical costs, and health is the condition even if specific complications are taken into account", which means that accurate projections of medical costs automatically include a certain degree of racial prejudice.

Correcting this unintentional problem may increase the proportion of black patients who get more help through the algorithm from 17.7% to 46.5%, the authors point out. So they tried to find a solution. By retraining the algorithm to focus on the combination of health and cost, rather than just future costs, the researchers reduced the bias by 84%. They extended the work into the future to "build continuous (unpaid) collaboration" to make the algorithm more efficient.

"these results show that label deviations can be resolved," the authors write. "there is no need to change our programs that are suitable for this algorithm (for example, by using new statistical techniques to de-relate predictive variables to race or other similar solutions). Instead, we must change the data we provide to the algorithm, especially the tags we provide for it."

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.