AMIE can serve as a complimentary tool for ophthalmology. It not only matches specialist performance, but also augments ophthalmologist clinical reasoning, leading to improved accuracy, consensus and planning. [Preprint] https://t.co/T7x7W2cNx1

We are excited to share our latest pre-print enabling effective human oversight for AMIE, our research diagnostic dialogue AI. We introduce a new asynchronous oversight paradigm, decoupling history-taking by AMIE from sharing a human-approved diagnosis – a thread 🧵:

I didn’t think to say this on the podcast but the true biggest lesson @AndrewLBeam taught me is that I’m no good at super smash bros!

It was super fun chatting about AMIE with @arjunmanrai @AndrewLBeam @alan_karthi and I’m incredibly grateful for the opportunity!

Great fun for @AndrewLBeam and me to hear about the latest with the AMIE project and other work at Google with @alan_karthi and @apalepu13. Coming soon to @NEJM_AI Grand Rounds

Gemini powers our multimodal health research! 💙 In our new paper on multimodal AMIE, we're pushing conversational diagnostic AI beyond text to handle images such as skin photos, ECGs, and clinical docs, which provide crucial context in healthcare. Blog: https://t.co/cdFPbufisP

✨New study from our team @GoogleDeepMind @GoogleAI - AMIE goes Multimodal✨ Our research conversational diagnostic AI now fluently considers visual photos/tests. In randomized OSCE study AMIE outperformed PCPs in simulated consultations in which patients uploaded photos of skin

"@dwarkesh_sp should read our papers" @vivnat & @apalepu13 explain that ... yes, AI systems can generate biomedical hypotheses that human experts find insightful & significant "AI doctors" exist today, and "data centers full of AI geniuses" are pretty clearly coming soon! 🧵

Excited to share that two of our @GoogleAI @GoogleDeepMind studies on AMIE have been published in @Nature! Papers: 1. https://t.co/NeSf9E7dBq 2. https://t.co/TdCOIF7wAe Blog: https://t.co/4w4eAOFQbb

Huge thanks to @labenz for hosting @vivnat and me on the @CogRev_Podcast. Really enjoyed discussing @GoogleAI @GoogleDeepMind's latest research on AMIE (A conversational diagnostic AI) & AI co-scientist (AI aiding scientific discovery).

2 important papers @Nature today indicate the potential for #AI to improve diagnostic accuracy and overall clinical performance for patient management and interactions @GoogleDeepMind @vivnat @danmcduff @taotu831 @alan_karthi https://t.co/TzuXgG05Jl

🚨 New paper 🚨 Excited to share new capabilities for AMIE our conversational medical AI @GoogleAI @GoogleDeepMind We advance AMIE beyond diagnosis 🩺 towards treatment 💊 over time w/ guidelines . Dive into 🔗 https://t.co/w5aEXX2OgU with wonderful co-last author @alan_karthi !

We are excited to share our latest results on AMIE, @GoogleAI @GoogleDeepMind conversational medical AI. TL;DR – in a blinded OSCE, AMIE was equivalent or better at not only diagnosis – but also at making management decisions. This is a big deal! 🧵⬇️ https://t.co/CTgrYQjQ2Z

AMIE, our research AI doctor from @GoogleAI & @GoogleDeepMind, just got a nice upgrade! Not just diagnoses anymore, AMIE can converse, consult and provide treatment recommendations, prescriptions, multi-visit care, all guideline-compliant. Blog - https://t.co/nsR3CFGR04

@NICEComms @BMJBestPractice @taotu831 [7/N] Incredible thanks to @hardyshakerman, @valentinlievin, @alan_karthi, @AdamRodmanMD, @taotu831, @vivnat, @ckbjimmy, @KhaledSaab11, @RyutaroTanno, @davidstutz92, @yongcheng16, Kavita Kulkarni, @SaraM66905, @joelle_barral, @dr2w, @AvinatanH48021, @ymatias, James Manyika and

@NICEComms @BMJBestPractice @taotu831 [6/N] Management reasoning is a highly challenging task, and far more research is needed before safe real-world implementation is possible. Nevertheless, our study demonstrates the strong potential of AI systems in this domain.

@NICEComms @BMJBestPractice @taotu831 [5/N] To assess AMIE’s medication reasoning, we leverage medication labels from 2 national drug formularies (US, UK), to develop a challenging pharmacist-verified MCQ benchmark. AMIE outperforms PCPs on the harder questions, benefitting significantly from access to external

@NICEComms @BMJBestPractice @taotu831 [4/N] Ultimately, AMIE’s management plans were rated as equivalent or better than those of PCPs, particularly in their preciseness and alignment with guidelines. Specialist physicians and patient actors largely preferred AMIE’s consultations and management reasoning.

@NICEComms @BMJBestPractice [3/N] Meanwhile, the Dialogue Agent (~system 1) , trained on simulated multi-visit dialogues, engages in fast and intuitive conversation while maintaining a persistent agent state. See this thread from @taotu831:

[2/N] We developed a two-agent system inspired by Kahneman’s dual process theory of cognition in “Thinking Fast and Slow”. The Mx Agent (~system 2) leverages authoritative guidelines (@NICEComms, @BMJBestPractice) and inference-time reasoning to carefully plan patient care. See