[ad_1]
This is because health data such as medical imaging, vital signs, and data from clothing materials may vary for non-specific reasons, such as life or background noise. The machine learning algorithms that are popular with professional companies are the best since the possible patterns find shortcuts to finding “correct” answers. it will not work in the real world. Smaller data makes algorithms more easily steal in this way and creates blind spots that lead to negative consequences for the hospital. “The villagers are stupid [itself] to think that we are creating models that work better than they do, “says Berisha.” That enhances AI hype. “
Berisha says the problem has led to an interesting approach in some areas of AI clinical research. In studies using algorithms to detect Alzheimer’s symptoms or cognitive impairment in word processing, Berisha and colleagues found that higher studies also reported to be more accurate than smaller ones — as opposed to what big data has to offer. A comments of studies that attempt to detect brain disorders from medical scans and another for studies that try to diagnose autism and machine learning also mentioned a similar approach.
The risk of algorithms that work well in early studies but acting differently on actual patient information is not just speculation. A 2019 study found that a multidisciplinary approach to patients’ care led to increased access to additional care for people with health problems. putting white patients ahead of Black patients.
Avoiding such biased systems requires big, relevant data and careful testing, but distorted data is a common occurrence in health AI research, due to historical and ongoing health differences. A The 2020 study by Stanford researchers found that 71 percent of the data used in the studies they conducted in-depth study Most U.S. medical facilities originated in California, Massachusetts, or New York, with fewer or fewer representations from 47 other countries. Low-income countries are not even represented at all in AI health education. Comments published last year in a survey of more than 150 studies using predictors for the diagnosis of disease or illness, it was found that many “show a faulty approach and are at greater risk for bias.”
Two investigators involved in these errors recently formed a nonprofit organization called Nightingale Open Science measure and control the amount and quantity of data available to researchers. It works with health systems to collect clinical pictures and data related to patient records, inform them, and make them available for non-invasive research.
Ziad Obermeyer, a Nightingale cofounder and assistant professor at the University of California, Berkeley, hopes that providing access to data will enhance competition that results in better results, such as larger, larger image openness. it helps to improve movement in machine learning. “The most important thing is that the researcher can do and say whatever he or she wants in terms of health because no one can monitor the results,” he says. “Data [is] closed. ”
Nightingale has joined other projects that are trying to improve health care AI by improving data availability and quality. The Lacuna Fund contributes to the development of machine learning systems that represent low- and middle-income countries and are engaged in health care; a new project at University Hospitals Birmingham in the UK with the support of the National Health Service and MIT is developing standards to assess whether AI systems are based on non-discriminatory information.
Mateen, editor of the UK’s report on epidemic algorithms, is fond of special AI projects like these but says AI’s prospects for health also depend on changing health systems. it is usually an IT standard. “You have to invest in the root of the problem to see the benefits,” says Mateen.
Some of the Best WIRED Stories
[ad_2]
Source link
