Artificial intelligence algorithms are trained on web data, but medical information is less widely available and more complex, leading AI tools to produce inaccurate results when used in the health field, Wired reported Jan. 16.
The Alan Turing Institute published a report in 2020 that looked at how AI helped during the COVID-19 pandemic. The study's findings showed that AI had little effect, as experts were faced with bias and lack of access to health data. The study also found that AI tools were flawed in detecting COVID-19 symptoms.
These findings showed a greater flaw with AI: that mixing data and algorithms is difficult to do in healthcare because of privacy concerns and outdated IT infrastructure, among other issues, Wired reported.
AI has been good at performing highly accurate tasks, but has become stagnant as quality and quantity of data for AI applications is sparse.
This makes AI risky to use in health systems, as it can be unreliable or biased toward certain demographics because the data reflects historical and ongoing health inequities.