In what is a major development for AI in medicine and healthcare, Google’s dedicated medical language model—Med-PaLM 2—has begun testing in prestigious hospitals such as Mayo Clinic. Building on the existing language model named PaLM-2, Google is expanding the reach of its AI tools and technologies to facilitate better penetration of its natural language processing capabilities. Its training data is specifically tailored to hold medical conversations, with Google’s AI team projecting it to be a product that might find utility in nations and regions that have limited healthcare access. As medical education and training continue to become revamped with newer technologies, the advent of dedicated language models such as Med-PaLM 2 will further revolutionize the discipline to incorporate the latest technology into its core domains. Med-PaLM 2 is designed to provide better and more accurate information when it comes to medical questions in comparison to general and conversational chatbots such as ChatGPT, Bard, or Claude.
While Med-PaLM 2 is not touted to be perfect or a replacement for human doctors, it certainly shows promise for the AI healthcare industry which seems to be slowly taking root in the backdrop of the LLM boom. Apart from the obvious benefits it offers as an initial point of contact for people seeking medical information, it can also be used as a tool to reach out to potential patients for procedures or follow-up consults. While Med-PaLM 2’s technical details are yet to be revealed, Google’s research on the topic suggests that despite some of its pitfalls, Med-PaLM 2 manages to perform fairly well when compared to its human counterparts. Explorations on the language model are entailed in the following sections.
The Formal Beginning of Medical AI: Med-PaLM 2’s Capabilities
Med-PaLM 2 has impressive reasoning, referencing, coding, and mathematical capabilities.
Med-PaLM 2 is a medical language model that is based on the PaLM 2 language model. The latter is a leap when compared to its predecessor, and is capable of generating long-form text. It elicits a fair bit of advanced reasoning capabilities, and is even able to write complex lines of code. Codey—a coding-specific language model and assistant—is also based on PaLM 2 and has shown coding proficiency. Med-PaLM 2 is built on the same PaLM 2 model for its reasoning credentials alongside its ability to deal with complex math problems. Med-PaLM 2 has been built on a specialized data set that was structured to include complex United States Medical Licensing Exam questions and demonstrations from qualified medical practitioners. The data set was broadly categorized into two distinct sections with one entailing the more technical aspects of healthcare. The other contained information on social and environmental determinants of health such as accessibility, bias, and socioeconomic moorings. Med-PaLM 2 has been undergoing testing across several institutions to gauge its performance and utility in a formalized medical establishment. Much like how AI might end up transforming STEM, healthcare, too, might go down a similar path.
Large language models catering specifically to healthcare domains have recently been gaining more prominence in the discussion since medical practice is primarily carried out via human communication. Language and empathy play a key role in the profession, and making patients feel heard and secure forms a core aspect of healthcare solutions. Apart from communicating well, Med-PaLM 2 is also known to be efficient at compilation and referencing. These tasks form an important part of the medical system, where records and reports are an inalienable part of the diagnostic and treatment process. Moreover, introducing a chatbot based on Med-PaLM 2 might also ease the worries in patients who often hesitate while seeking medical advice. Though firms such as Inflection.ai have worked on creating a “friendly chatbot,” advanced language capabilities along with sentiment analysis can aid in building an able medical assistant. Unarguably, intuitive clinical decision-making and critical thought are the key components of clinical practice. Given that the base PaLM 2 model is trained in nearly 100 languages, it also addresses issues surrounding accessibility and understanding.
Will AI Healthcare Become Medicine’s Future: Concerns Surrounding Med-PaLM 2
Medical language models can perform several functions such as communication, data analysis, and record-keeping.
Med-PaLM 2 has been a promising entry into the growing domain that is medical AI. However, that’s not to say it’s perfect. Like other artificial intelligences, it, too, is prone to suffer from inadequacies. Phenomena such as hallucinations will be especially concerning since the stakes in a healthcare AI setup are far higher and come with several ethical complications. Med-PaLM 2 was evaluated against several questions that were posed to it as well-qualified physicians based in the United States, the United Kingdom, and India. The responses to these questions were evaluated by both laypersons and physicians. In many cases, the laypersons found Med-PaLM 2’s responses to be more conclusive than those written by the physicians. Moreover, Med-PaLM 2 scored a stunning 86.5% on questions modeled on the United States Medical Licensing Examination (USMLE.) Though impressive, this still leaves enough room for questions surrounding privacy, security, and confidentiality.
Deploying tools like a medical chatbot in a formal clinical setup will require robust safety protocols. Also, if such tools are used as initial points of contact with patients, the latter must be informed that they’re communicating with artificial intelligence beforehand. The studies carried out so far on Med-PaLM 2’s capabilities are still limited, and technologies of this scale are refined over some time. Developers, doctors, and other healthcare professionals must work together to set up a clear ethical framework for the deployment and potential uses of medical AI platforms such as Med-PaLM 2.
The Potential for AI in Medicine
Despite the efficiency of Med-PaLM 2, it is built to support doctors and not replace them.
Medical science in the last century has been catapulted into an age where successes of treatment outcomes have grown exponentially, essentially extending the human lifespan globally. This has happened due to the proactive adoption of modern technologies to enhance diagnostic, treatment, and prognostic results for patients. As AI enters a new era of development, much like its more basic electronic predecessors, it, too, is bound to become integrated with just about every walk of human life. As language models like Med-PaLM 2 gain traction and are primed to become more suitable to the clinical environment, the focus should remain on maintaining humans at the forefront of medicine while operating AI to the benefit of patients. Emphasizing responsible AI practices becomes all the more important, for medical ethics and practice cannot be substituted by autonomous systems.
FAQ
1. What is Med-PaLM 2 used for?
Med-PaLM 2 has been built primarily to answer health and medicine-related queries with a higher rate of accuracy when compared to general conversation chatbots. It can also organize and analyze data and is capable of referencing, which is key to medical practice. Future Med-PaLM iterations will be multimodal to analyze X-ray radiographs and Mammograms.
2. What is Med-PaLM 2 trained on?
Med-PaLM 2 has been trained on a broad dataset that contains questions from medical licensing exams, demonstrations by qualified physicians and healthcare professionals, as well as generalized information surrounding the social, environmental, and economic influences on healthcare.
3. What is the future of AI in medicine?
AI is growing to gain greater degrees of prominence in healthcare. While Med-PaLM 2 is the language model approach, other AI-influenced medical tools that aid in simulations, research, data analysis, and clinical studies have also been gaining traction in the past decade.
4. Will AI medicine replace doctors?
Medicine has always been an innately human domain that requires streamlined communication between healthcare providers and patients. Though AI medicine can help bridge medical accessibility concerns, index medical records, provide data analytical support, and become initial points of contact, AI medicine cannot replace human physicians. Medicine requires a variety of skills to be utilized in harmony, and an autonomous system will not match up to the high-stakes requirements of the profession.