There has been talk for some time about the application ofArtificial intelligence to the medical field, with various implementations with concretely useful implications for patients.
To date, for example, talking about AI involved in theophthalmology with notable successes. As far as the diagnosis of eye diseases is concerned, for years these systems have been equal to human work, often offering even more accurate analyzes than most specialists.
Despite this, a new study has put it to the test GPT-4demonstrated how the model proposed by OpenAI is comparable to human ophthalmologists when it comes to evaluating eye infections and conditions.
The research in question, published a couple of days ago in the magazine PLOS Digital Healthresearchers put GPT-4 to the test, GPT-3.5 e LLamaasking the various AIs 87 questions multiple choice on an ophthalmological theme. The same questions were then asked to five professionals in the sector, three interns in ophthalmology and two young ophthalmologists. The test in question gave surprising results, which demonstrate the full potential of GPT-4.
GPT-4 proves to be an “average” ophthalmologist compared to human professionals
The OpenAI model answered correctly 60 out of 87 questionsexceeding the average score (59,7 correct answers) of the three trainees. The two professionals, on average, had 66,4 correct answers.
Although this indicates that humans generally won on GPT-4, it should be taken into account that one of the two stopped at 56, proving to be less accurate than AI. In fact, GPT-4 was confirmed to be in the “average” of real experts.
And the other models involved? GPT-3.5 got 42 correct answers, while LLama stopped at 28. Although these tests were carried out on a very limited number of samples regarding trainees and active doctors, these are results that reward OpenAI’s work.
Even ophthalmologists outside the experiment, once evaluated, said they were impressed by the AI’s capabilities. For Piers Keanean AI specialist at University College London and an ophthalmologist at a London hospital, the research findings are defined as “Exciting and very intriguing” demonstrating how GPT-4 still has great potential also in this branch of medicine.