DailyPost 2671
ARTICULATE MEDICAL INTELLIGENCE EXPLORER (AMIE)
Without getting into the recent progresses made in the field of AI, suffice to say that we have hearing of breakthroughs at regular intervals, after the conversational AI tsunami hit mankind, the ChatGPT. All this cumulatively are set to transform our existence in a manner, that it would be unrecognizable from todays, in a very short period of time. From Nvidia recent offerings to Gemini, to AI wearables to the recent Google product in the making; Articulate Medical Intelligence Explorer, in short AMIE, have the promise of a different world. If it fans out in the manner visualized and mostly it will, it is bound to immensely change the ease of healthcare.
AMIE is tool which Google claims based on the recent video that is really good at diagnosing the patient and is arguably better than the doctors. How good is the system is the first question; that can be answered better, if we know what they have set out to make. For the uninitiated this is a research AI system based on LLM, as most of AI tools are in some way or the other. Inspired by the challenge to changing health care with precision, AMIE, LLM enabled, is optimized for diagnostic reasoning and conversations. The model has been trained and evaluated on many dimensions that reflect quality in real world clinical consultations. As per the ground level need, it is being created both from the perspective of clinicians and patients.
Given the complexity of the field to be handle and complex tasks ahead, the need to scale AMIE across varieties of disease conditions, specialties and scenarios was a necessity. This was done through a mechanism of novel self-play, which has been the creation of Google. Self-play based “simulated diagnostic dialogue environment with automated feedback mechanisms to enrich and accelerate its learning process.” This abridged the learning process and could be done fully in the lab itself. You can call some variant of the creating synthetic data for exponential learning which us one of the base requirements of these systems.
There was need to keep improving the diagnostic skills of the model, which could be parameterized for accuracy and articulate being its first name, the conversational quality needs to be at least of the human level. This was done with the introduction of an “inference time chain-of-reasoning strategy.” Without validations at every iterative stage, the learning can turn out of self-defeating at the end. For this purpose, AMIE was tested “prospectively in real examples of multi-turn dialogue by simulating consultations with trained doctors.” Though still in the nascent stage, but with the promise of moving in the direction proven, the pathbreaking healthcare AI tool seems to be on the horizon.
ONLY AI CAN TRANSFORM HUMAN HEALTH WITH ALL IT COMPLEXITIES.
Sanjay Sahay
Have a nice evening.