""The AIs' failure rates exceeded 80 percent when provided with given ambiguous symptoms that could match more than one condition, and for more straightforward cases... they still failed 40 percent of the time.""
""Despite continued improvements, off-the-shelf large language models are not ready for unsupervised clinical-grade deployment... Differential diagnoses are central to clinical reasoning and underlie the 'art of medicine' that AI cannot currently replicate.""
""An AI that leaps to conclusions when not represented with the full picture could have devastating consequences... they may be presented with misleading information and potentially dangerous advice.""
A study published in JAMA Network Open evaluated 21 large language models (LLMs) for their ability to provide medical advice. The results showed that these AIs failed over 80% of the time with ambiguous symptoms and 40% with straightforward cases. Unlike human clinicians, LLMs often jumped to single answers, leading to poor performance. Experts warn that AI's inability to perform differential diagnoses poses significant risks, as misleading information could result in dangerous health advice for users seeking help.
Read at Futurism
Unable to calculate read time
Collection
[
|
...
]