Indian AI clears NEET and AIIMS PG Exams; outperforms GPT-4, MedPaLM

August AI, developed by Bengaluru-based start-up Beyond, has cleared the National Eligibility-cum-Entrance Test for Post Graduation (NEET PG) and the All India Institute of Medical Sciences Postgraduate (AIIMS PG) exams in India, scoring 74.5 per cent. It also secured 94.8 per cent on the US Medical Licensing Examination (USMLE), a test that physicians must clear to practice medicine in the US, outperforming established AI systems, including OpenAI’s GPT-4 and Google’s MedPaLM.

“August AI’s remarkable achievement in these tests showcases its competence and deep understanding of medical knowledge,” Anuruddh Mishra, Co-Founder & CEO of Beyond, said.

Rather than conventional methods, August AI utilises a combination of large language models (LLMs) and a custom orchestration layer to create an innovative educational tool. This technology allows for personalised and natural conversations, closely resembling everyday communication on messaging platforms such as WhatsApp.

Also read: DAX Express. GPT-4 to power healthtech, reduce doctors’ burden of note-taking

“We’ve done over 1,500 health consultations over the last year,” Mishra said, “to figure out what are the best ways to actually guide someone around their health, and fine-tuned August around those conversations.”

This made-in-India AI is also designed to understand the Indian health system. “Understanding the health of people from different ethnicities is still something that the overall healthcare ecosystem is not good at,” Mishra said, “Since the start, we’ve focused on health for India and Indians. We’ve actively been collecting India-specific data and have added this to August.”

As AI becomes more prevalent in healthcare, it is essential to prioritise data security. Mishra says that unlike Google, where every search leads to a targeted ad, all conversations with August are private.

The development team, comprised of engineers, data scientists, and medical professionals, conducted experimentation to refine various LLMs, culminating in the core engine behind August’s health AI. Employing a proprietary ensemble refinement technique, August AI provides precise answers, while minimising inaccuracies and hallucinations. Mishra explains, “Unlike Google, where even a simple health issue like a headache leads to a conclusion of cancer, August thoughtfully allows people to understand their symptoms and even suggests the right type of person they should consult for the issue.”

However, that being said, August does not make any diagnosis. “We don’t think health AIs are there yet,” Mishra said.

“AI can provide preliminary information to patients regarding their symptoms, but it should not be used to replace the advice and opinions of qualified healthcare professionals,” said Srinivas Chilukuri, Senior Consultant, Apollo Proton Cancer Centres. He added that AI may help in alerting patients regarding emergency clinical conditions and guide them towards approaching hospitals immediately.

Also, Mishra points out the major challenges such as cost and collecting specific data in developing Generative AI for healthcare. “As we look to reimagine it and scale, we’ll need support, not just from the existing healthcare ecosystem, but the government, technologists and capital allocators,” he adds.

AI in medical entrance exams

Given the rigorous nature of medical entrance exams, AI-driven tools can provide valuable assistance to medical students. Chilukuri said, “AI can be a virtual tutor where access to personalised teaching is not possible or affordable. It can be used to generate realtime feedback and clarify concepts. There are other benefits such as providing performance analytics, which can aid students in their exam preparation. However, AI cannot replace hard work, dedication and smart preparation strategies. AI can potentially complement the guidance from experienced teachers or mentors to help students achieve success in the entrance exams.”

Leave a Reply

Your email address will not be published. Required fields are marked *