GPT-4 outperforms doctors in the accuracy of diagnosing vision problems

GPT-4, proved to be significantly more effective than non-specialists in identifying vision problems and providing advice. A study conducted by the University of Cambridge showed that the model’s clinical knowledge and logical thinking skills were at the level of ophthalmologists.

The GPT-4 was tested in a setting where 87 patients with certain eye diseases were presented with a series of scenarios. The model had to choose one of four options to make a diagnosis or recommend a treatment; GPT-4 scores were significantly better than those of non-specialists and comparable to those of interns and ophthalmologists. The questions covered a wide range of vision problems, including light sensitivity and visual impairment. Note that the content of these scripts is not available online, making it unlikely that GPT-4 will be able to learn from this data.

Artificial intelligence technology has long been used in healthcare, and this case study demonstrates its potential in triaging patients with vision problems. It is useful in determining which cases require urgent ophthalmological examination, which can be referred to a general practitioner, and which do not require treatment at all. Dr. Arun Thirunavukarasu, lead author of the study, said that GPT-4 demonstrates the same ability as an expert clinician in diagnosing and recommending symptoms of eye problems.

The results of this study suggest that GPT-4 can be used in healthcare settings, especially in areas where access to specialist ophthalmologists is limited. However, much work remains to improve and optimize these models. Developers should also consider that patients may use AI as an alternative to seeking medical advice online.