Will Ai Replace Human Doctors: Can AMIE, Google’s Medical Chatbot Diagnose You Better than Human MDs?
Imagine you’re feeling crummy, but finding a top-notch doctor feels impossible. Enter AMIE, Google’s brand new virtual doc, a medical chatbot ready to tackle whatever you can through at it! ???? are optional.
Will Ai replace human doctors anytime soon? Buckle up, because we’re about to find out.
AMIE is showing that it is already able to outperform Board Certified MD’s in diagnosing both common and complex cases according to the study they just released.
Google AI is Developing a New Kind of AI Doctor called AMIE, A ChatDoc
AMIE is a Articulate Medical intelligence Explorer, an LLM AI that chats with humans like ChatGPT.
A Chatdoc.
Imagine AMIE like a super-smart med student who is being trained to be the best doctor to ever exist…in every speciality.
AMIE is actively learning from patient/doctor conversations, figuring out how diagnoses were made, and how to best built trust and empathy with you so they can best understand what is wrong.
But here’s the twist: AMIE didn’t sit in class and take notes like your doctor.
AMIE is not even a real person.
It is like if you were able to be all around the world, sitting in with different medical specialists at the same time, learning directly from them in their native language!
This amazing possibility has helped AMIE understand diseases and talk to patients in all sorts of new and different ways.
AMIE has been practicing on virtual patients and actors pretending to be sick. The AI has tested its skills at taking medical history, asking questions, and even showing empathy (understanding how the patient feels).
Finally, the big test came! AMIE went head-to-head with real doctors in simulated chat consultations. No one knew who was who – a human doctor or AMIE. They both talked to real patients, in real time, asking questions and trying to figure out what was wrong with them.
Afterward, everyone compared notes: how well did they gather information, connect with the patient, and reach the correct diagnosis?
The results were crazy! It is not a question of will AI Replace Human Doctors, but when will they team up.
Will Ai replace human doctors? Yes and no. Ai is proving to be superior in many ways to humans. Just like with other professions, the Doctors that learn to use AI as the amazing tool it is destined to be will be rewarded with healthier patients and an easier workflow. Training to use AMIE and other Chatdocs will be a necessity to keep up with the times. If Doctors don’t, they will lose their patients to the Chatdocs.
Google has successfully developed an AI that can not only communicate with real patients using a LLM chat interface but perform better than Board Certified MD’s in diagnosing both common and complex cases according to the study they just released (see link below)
Link to Full Study Paper; Towards Conversational Diagnostic AI
The LLM, AMIE, analyzes written word by asking questions and refining their diagnostic plan until a working diagnosis is determined.
Just as in real life, the better you communicate with the patient, the higher the confidence of the diagnosis.
The crazy part is that empathy and clear communication is built into the AMIE LLM to ensure that a relationship is fostered with a wide range of personality types from around the world.
The study used the usual text based ways we are communicating with LLMs today.
Consultations happened with real people acting like patients talking with either real doctors or AMIE.
Then either AMIE or the MD blindly had to figure out, over chat, what was wrong with them.
BE sure to see the randomized, double blind crossover study linked above.
When reviewing the findings between real MDs and AMIE the results were fascinating.
The development of AMIE combineds it’s superpower to self-play and continually learn with fine tuning by human critics, then letting it self-play & learn.
This saw AMIE continuing to improve with each feedback loop.
The feedback loop created resulted in significant improvements in all 6 categories tested to the point where it is now beating MD’s evaluations.
The many fine tuning inputs given to AMIE by critics, that interrupted the self-play learning, looks to be key in AMIEs improvements over MD interactions.
This continuous feedback loop of learning, critiquing, self-play learning, critiquing, self-play learning on so on obviously has a massive advantage over humans.
This advantage would appear to be set to only grow as more and more real world consultations are experienced by AMIE.
MD’s and Patient’s Trust Grew Over Time as AMIE Grew
MD PERSPECTIVE
- Top-3 Diagnostic Accuracy
- Management Plan
- Escalation Recommendation
PATIENT PERSPECTIVE
- Empathy
- Patient’s Confidence in Care
- Perceived Openness & Honesty
One Big Concern; Human’s Education & Logical Reasoning Limit
When looking in the real world data sets they used I do have one real concern that jumped out at me about patient inclusivity from a language and logic perspective;
- The real world data sets included medical reasoning, summarizations and limited real world conversations.
- But it did not seem to train AMIE using messy, real world natural speech that often include slang, jargon, sarcasm, humor, utterances, and obscure references because there simply were not enough, good quality recordings.
- This leaves the model with a blind spot when chatting with people that lack the proper words to describe their symptoms.
When will Ai replace human doctors and will this cause the divide between the educated and uneducated’s access to healthcare to widen? Will this cause people to become aggravated with the technology like they have online chatbots when trying to interact with businesses?
Complicated Cases are Harder to Solve for Both AMIE and MD’s
We know real MDs have a hard time solving complex cases during office visits but now we have data that shows that AMIE does perform better than MD’s when dealing with complex cases.
Be sure to see those results above of the study of diagnostic accuracy in clinically complex cases?
- AMIE alone outperformed unassisted MD’s by nearly 25% at almost 60%
- MD’s assisted by AMIE came in just above 50%
- MD’s assisted by search arrived around the mid 40%s
- MD’s unassisted was below the mid 30%s
This shocked me.
I would have assumed that a Doctor using this AI would have been far superior to an AI working by itself.
Assisted randomized reader study setup to investigate the assistive effect of AMIE to clinicians in solving complex diagnostic case challenges from the New England Journal of Medicine.
Specifics for those who want to get into the deeper details.
Here is the link to the blog where Google discusses all of this in greater detail. AMIE: A research AI system for diagnostic medical reasoning and conversations. *This is also the source for most of the images in this post.
An objective structured clinical examination style (OSCE) was followed with either a board certified MD or the AMIE LLM Model chatting with the real person.
This blindly evaluated how real MD’s and AMIE gathered histories, listened & built rapport, used empathy, and determined medical diagnosis when speaking with real people via chat.
Soon enough, we will be able to access the most current medical AI from our cell phone.
I want to let my idealist side run wild for a minute. Trust me, I will reign her back in.
Imagine a world where anyone, anywhere can expect accurate diagnostic conversations to be as easy as picking up your cell phone.
Access ChatDoc orders for procedures, bloodwork, etc. at a local care center and simply have the testing done for a set price.
Once the ChatDoc receives the results you are contacted and informed what to do next.
This is both scary and exciting to me.
This is what I think will Actually happen.
Doctor’s will likely be expected (by the financial overlords that control the purse strings) to use this technology as additional “hands on deck’ so they can manage evermore patients.
Reality may fall somewhere between dystopia and utopia.
More burnout for real life Doctors while patients receive more accurate diagnosis, faster.
I still have many questions after writing this article.
- Will AMIE be trained to address teaching us how to feel better by preventing chronic illness?
- Will Chatdocs usher in a true “Health” Care Model rather than continuing the failing “Sick” Care Model?
- Will rural areas that do not currently have other choices adopt this model more readily than cities?
- Will this lead to chatdocs that are versed in everything from anti inflammatory diets to exercise routines so we are not so confused about healthy living as we have been for the last century.
- When will Ai replace human doctors and will this cause the divide between the educated and uneducated’s access to healthcare to widen?
