
Research Fellow - Generative AI @AIHI_MQ . GP (and PhD). Primary Care, Research Methodology, Applied NLP with a pinch of Philosophy. Views my own
Sydney, Australia
Joined February 2012
- Tweets1.9K
- Following469
- Followers320
- Likes7.8K
Grateful to have received a @cosmosinstitute grant to build Quorum: a deliberation tool for clinical decision-making
Instead of 1 AI voice Quorum brings several archetypes with a voice of its own
Learn & choose what to take & what to dismiss
We are building now. We have quorum
Congratulations to our 80 new Cosmos grantees!
Our largest cohort to-date will be building prototypes for AI that supports human autonomy or truth-seeking (in partnership with @TheFIREorg.)
The projects range from wearable AI for memory and cryptographic tools for scientific collaboration to a benchmark based on pre-1940 thought on machines and autonomy.
Our grantees come from sixteen countries, including the US, UK, Nigeria, Ethiopia, South Korea, Costa Rica, India, and Czechia.
They join a community of Cosmos grantees and fellows whose projects have grown into companies, non-profit research organizations, and papers published at leading venues like ICML, COLM, and IEEE.
Read what all eighty are working on: blog.cosmos-institute.org/p/…
David Fraile Navarro retweeted
📢 New paper out in @BMJ_HCI, one of the largest UK survey yet of GPs using ambient AI scribes (n=1003, August 2025).
The headline numbers:
📈 14% of GPs are already using ambient AI scribes. Another 39% plan to soon. Only 46% have no plans at all. That's fast diffusion for a technology barely on the radar two years ago.
🏆 Heidi Health dominates — used by 86% of adopters.
⏱️ 80% say it cuts documentation time. 70% say it cuts cognitive load. 55% rate the AI notes as better than standard documentation.
⚠️ But it's not clean: 32% report errors often or always, and 14% call those errors significant-to-critical. Errors spike in multiparty consultations (38%), complex histories (35%), and non-English encounters (31%). Yet, GPs believe the notes are still better than with no AI.
🖊️ Consent is patchy: only 63% routinely ask patients before using it. Among those who do ask, ~10% or fewer decline — suggesting consent is workable in practice, it's just not consistently happening.
Our read: GPs aren't treating AI output as finished documentation — they're treating it as a draft that still needs checking. That's likely why time is saved even as errors get caught and corrected. But the uneven accuracy across patient groups is a real equity flag, and regulation/consent practice needs to catch up fast to adoption.
Full paper (open access): informatics.bmj.com/content/…
With @AnnaKharko Carolina Garcia, @dafraile , Brian McMillan, Jens Gaab, Cosima Locher, and @EnricoCoiera
#DigitalHealth #AI #GeneralPractice #HealthTech #PrimaryCare
Today is a very sad day @RichardLehman1 a close friend, mentor & extraordinary man who embodied the best of what General Practice is has passed away. He was a light I followed since training & later came to know & work alongside. He's deeply missed but his legacy will outlive him
David Fraile Navarro retweeted
The era of AI-informed patients is upon us.
Given the choice and quality available, it seems inevitable that patients and these tools will now become the main drivers of medical practice.
@crblease @dafraile
Brief summary of several of the patient-facing AI health chatbots
jmir.org/2026/1/e99230
We have Quorum.
Most AI decision assistants are often single voice and this can lead you astray: sycophancy, monoculture, no dissent.
You can't completely prompt your way out of it. I'm building a different architecture. (link to repo below)🧵
Pattern generalises: legal review, peer review, editorial work, due diligence. Use responsibly. But let me know if it helps!
Demo: dafraile.com/quorum-demo
Code + design docs: github.com/dafraile/quorum
1
1
46
Built at the OpenAI Codex Hackathon Sydney last week (and then refined over the weekend). Thanks to @gabrielchua and team for the unlimited Codex access, used it heavily for the runtime scaffolding.
Really fun session!
55
David Fraile Navarro retweeted
"I invite the conversation upfront now. Tell me: what did Dr. ChatGPT and Dr. Google say to you?"
Listen to the Medicine and Science Podcast - bit.ly/4t1BjVR
You probably have seen the headlines on the latest paper bashing AI.
Our response to the Nature Medicine paper claiming ChatGPT Health under-triages 51% of emergencies is now on arXiv.
Short version: the evaluation format drove the failure, not the models. Link below 👇
All code, prompts, raw model outputs, and adjudication data are public (still working on expanding tests and results): github.com/dafraile/nature_e…
📄 arXiv: arxiv.org/abs/2603.11413
With @FarahMagrabi and @EnricoCoiera
1
1
202












