In the first GPT-4 language model study, ophthalmologists were given an evaluation eye exam. Even though it lacks human eye doctors, GPT-4’s output was so far behind them, that it raised eyebrows and choices about the role of AI in healthcare.
The study, now published in PLOS Digital Health, compared GPTP-4 with doctors of different experience levels (ranging from the least to the most experienced doctors). The AI answered a total of 87 multiple-choice questions on ophthalmic topics ranging from photophobia to wounds. The training data for GPT-4 is a closely guarded secret, which is probably the result of internal progress in improving training for AI at OpenAI.
The results were impressive. Overtaking the juniors for GPT-4 was easily accomplished, and the average score of well-trained experts is now only better than the latter, at the same time, one expert was impressed with his expertise, but GPT- The average score of 4, being in a tie with the rest of the qualified doctors, did not differentiate it from the others. It is worth noting that these AI models like PaLM 2 and LLaMA were not up to standard in terms of efficiency compared to this AI system.
Such research shows the huge role AI can play in enhancing medical applications. Media like ChatGPT, enabled by GPT-4 or better technological advancements, can certainly improve healthcare by helping in screening, early diagnosis, and information retrieval. Experts also recommend that the essential cognitive skills required for good medical practice should not be ignored. AI, although powerful in many applications, has not yet demonstrated the level of complexity (nuance) required for accurate diagnosis. Thus it requires the co-participation of human expertise.
This invention underpins many innovations as improved versions become hits. Think about AI providing initial assistance to clients at the beginning of a consultation, allowing doctors to focus on the most complex cases. The demonstration of GPT-4 has opened up the field of AI and allowed doctors to see how artificial intelligence can play a positive role in medical diagnosis and perhaps many other clinical areas.
👁️ #AI is much better than non-specialist doctors at assessing eye problems and providing advice, Cambridge researchers have found.
GPT-4 could triage patients and decide which #ophthalmology issues are emergencies that need immediate attention👇 https://t.co/nX9OYQb1XR
— Cambridge University (@Cambridge_Uni) April 18, 2024
