Google Health, Google Deepmind and Google AI have unveiled Med-PaLM M, a large multimodal generative model that flexibly encodes and interprets biomedical data. It can handle various types of medical data, including clinical language, medical images, and genomics, and performs well on a wide range of tasks, all using the same set of model weights.
Medicine is inherently multimodal.
Thrilled to share Med-PaLM M, the first demonstration of a generalist multimodal biomedical AI system with a stellar team @GoogleAI @GoogleDeepMind @GoogleHealth
Paper: https://t.co/oEMXZSW2bK pic.twitter.com/ZgEtG0gXEs— Vivek Natarajan (@vivnat) July 27, 2023
It was built by fine tuning and aligning PaLM-E, a language model from Google AI, to the medical field using a specially curated open-source benchmark called MultiMedBench. MultiMedBench consists of 7 biomedical data types and 14 diverse tasks, such as medical question-answering, generating radiology reports, and identifying genomic variations. With over 1 million samples, this benchmark encourages the development of generalist biomedical AI systems.
Med-PaLM M excels in all tasks on MultiMedBench, often outperforming specialist models by a significant margin and even surpassing PaLM-E, proving the importance of adapting the model to biomedical data.
The key idea behind building a large-scale biomedical AI is to use language as a common framework for different tasks. This allows the AI to combine knowledge from various sources and transfer skills across tasks more effectively.
Excitingly, preliminary evidence suggests that Med-PaLM M can generalise to new medical tasks and concepts and perform multimodal reasoning without specific training. It can accurately identify and describe medical conditions in images using only language-based instructions and prompts, even if it has never seen such cases before.
To assess the practical use of Med-PaLM M in clinical settings, radiologists evaluated AI-generated reports at different scales. The AI’s error rate was found to be comparable to that of radiologists from previous studies, indicating its potential clinical usefulness. The big tech launched the first version of MedPaLM in December, 2022.
Read more: Responsible AI Takes Center Stage at Google I/O Connect
Google’s Unwavering Commitment to AI in Healthcare
Google’s Med-PaLM 2, a medical chatbot that answers medical questions, has been a fan favourite since its launch.
Med-PaLM 2 is built upon Google’s language model, PaLM 2, and uses LLMs tailored to the medical domain. The AI has demonstrated impressive performance on medical question-answering datasets, achieving high accuracy on the US Medical Licensing Examination (USMLE)-style questions and the Indian AIIMS and NEET medical examination questions.
Google acknowledges the complexity of personalised medical care and recognises that Med-PaLM 2’s results may not be generalisable to every medical question-answering setting and audience. The AI is trained on medical Q/A datasets but excludes patients’ personal data to adhere to ethical norms.
While having access to patients’ personal data could enhance Med-PaLM 2’s efficiency, privacy concerns are likely to prevent many patients from sharing such information. Google ensures that customers testing Med-PaLM 2 will retain control of their data in encrypted settings, inaccessible to the tech company, and the AI program will not ingest any of that data.
Read more: Google Takes AI Healthcare in Its ‘Med PaLM’
The post Google Unveils Multimodal Generative AI Model Med-PaLM M for Healthcare appeared first on Analytics India Magazine.
