Ruogu Fang’s Post

We could be over fear of AI with science, that’s the way of humans improve their life experience about the new challenges. This is an interesting Scooping Review where Ryan Han and colleagues explore the potential of AI to improve care management, clinical decision-making efficiency, and more, and it may be a step on get this goal. They do this by examining randomised controlled trials on AI in clinical practice: “Future AI trials should prioritise patient-relevant outcomes to fully understand AI’s true effects and limitations in health care”

Sharing Step 4 of my journey into AI. This work was part of the SRI program hosted by the Harvard Undergraduate OpenBio Laboratory. I’m deeply grateful to the SRI organizers for making this opportunity possible and to my incredible mentor, Milos Vukadinovic, for his guidance and support. From my article: "This project wouldn’t have been possible without my mentor, who did more than just guide me — he listened, gauged my capabilities and challenged me! His support helped me navigate tough decisions and pushed me to grow every step of the way." Check out the full story of my journey into AI and healthcare here: https://lnkd.in/dDXqvkGH

This article examines the implementation of generative artificial intelligence (AI), specifically large language models (LLMs), into medical practice and the implications of embedding human values into these models. It highlights recent advancements in AI, such as LLMs' ability to perform complex tasks such as writing essays, passing exams, and generating empathetic messages. Nonetheless, despite these advancements, significant concerns prevail about the reliability of AI in healthcare due to issues like factual inaccuracies and confabulation. Addressing these concerns involves understanding the human values embedded in AI models. The authors emphasize the importance of human value judgements in clinical decision-making, tracing its concept back to the 1950s. They illustrate how AI can reflect as well as mitigate societal biases, using examples from medical imaging. Clinical scenarios demonstrate the complexity of aligning AI recommendations with diverse human values. Values enter AI models through training dat, model design, and usage. This raises questions about whose values are encoded into these models and the need for periodic retraining to address dataset shifts. The article stresses the ongoing importance of clinicians in guiding AI-driven decisions. In summary, AI in healthcare must reflect patient values and societal goals accurately. Clinicians play a crucial role in ensuring that AI recommendations align with ethical considerations and patient needs.

The use of AI in health care has seen remarkable growth in the past 5 years. But there is uncertainty, including about meaningful risk to patients and clinicians. In a scoping Review, Ryan Han and colleagues explore the potential of AI to improve care management, clinical decision-making efficiency, and more. They do this by examining randomised controlled trials on AI in clinical practice. 🗨️ “Future AI trials should prioritise patient-relevant outcomes to fully understand AI’s true effects and limitations in health care”, say authors. Find out more ➡️ Figure 2: Randomised controlled trials of artificial intelligence in clinical practice across countries and specialties https://hubs.li/Q030PZgD0

1. The University of Texas has established the Center for Computational Medicine to integrate AI into healthcare. 2. Charles Taylor, inventor of HeartFlow, has been appointed to lead the center. 3. HeartFlow is the first widely used AI technology approved by the FDA for diagnosing heart disease non-invasively. 4. Taylor will create tools to model disease progression and personalize care, while also helping researchers and startups. 5. He aims to train future doctors in the use of AI in medicine. 6. Dr. Claudia Lucchinetti highlighted the center as a significant advancement for health care in Texas and beyond. 7. The university has existing resources, including the fastest academic supercomputer and other specialized centers. 8. Taylor will hold a joint position at both the Dell Medical School and the Oden Institute, a first for the Oden Institute. 9. Taylor expressed enthusiasm about Austin's innovative environment, likening it to the spirit of Silicon Valley. 10. He anticipates faculty collaboration on AI applications in clinical settings and aims to be a mentor for implementing ideas. 11. Taylor envisions the emergence of new technology companies affiliated with the University of Texas in the next five years.

I’ve been reflecting on ideas from Yuval Noah Harari's Nexus and a recent NEJM review by Raj Manrai and Zak Kohane on human values in AI. Harari emphasizes the power of self-correcting mechanisms within large information networks—institutions like scientific societies that thrive by identifying and correcting errors using peer review. These mechanisms, though imperfect, are vital for ensuring the adaptability and resilience of broader systems like religion or government. This principle resonates deeply in the context of AI. As highlighted in the NEJM paper linked to below, AI systems can generate vastly different outputs depending on the values embedded in their design, training, or prompts. For instance, an AI tasked with assessing a child with potential growth hormone deficiency might prioritize different outcomes if prompted as a clinician versus an insurer. This underscores the need for a "self-correcting value" in AI applications. Beyond embedding specific values like patient safety or clinical accuracy, we should design AI systems with a meta-layer of self-correction. For example, an AI triaging patients should incorporate not only values like patient safety but also a mechanism—whether internal or external—to identify and rectify errors. Harari's insights remind us that no system, AI or otherwise, can achieve true robustness without the humility to embrace and address its flaws. In AI, this translates to designing systems that don’t just execute tasks but actively learn, adapt, and self-correct to uphold human values. What are your thoughts on integrating self-correcting mechanisms in AI? How can we ensure they remain effective as these systems evolve? Isaac Kohane https://lnkd.in/ec7ZstiH https://lnkd.in/exNfa5wJ

On Thursday, January 9th, 2025, Dr. Carsten Marr, Founding Director of the Institute of AI for Health at Helmholtz Munich, shared valuable insights during his guest lecture as part of the Medical Information Sciences (#MIS) series at the Universität Augsburg. +++ 𝗚𝗘𝗥𝗠𝗔𝗡 𝗩𝗘𝗥𝗦𝗜𝗢𝗡 𝗜𝗡 𝗧𝗛𝗘 𝗖𝗢𝗠𝗠𝗘𝗡𝗧𝗦 +++ His lecture, titled "AI in Biomedical Research and the Diagnostics of Blood Diseases", focused on the role of #artificialintelligence (AI) in biomedical research and diagnostics. 𝙏𝙝𝙚 𝙡𝙚𝙘𝙩𝙪𝙧𝙚 𝙘𝙤𝙫𝙚𝙧𝙚𝙙: 🔷 The fundamentals of machine learning and AI. 🔷 Applications enabling highly precise characterization of individual cells. 🔷 How these advancements can contribute to personalized medicine. A big thank you to Dr. Marr and all participants for an engaging session and the discussions! 🤝 🔗 Stay informed about future events by subscribing to our newsletter: https://bit.ly/3NsmKqJ (find the form at the bottom of the page) #machinelearning #research #personalizedmedicine #digitalhealth

Complex systems create inherently interdisciplinary frames. A discussion with some of my MCiM classmates on genAI adoption and implementation in healthcare. Sharing our 34min podcast final project of Sara Singer's Organizational Behavior class "Leading and Managing Health Care Organizations: Innovation and Collaboration in High Stakes Settings".

It’s already Spring Quarter at Stanford and we are underway again with Med216: Generative AI and Medicine. Last year we meant to launch a small research seminar to explore generative AI topics in medicine, and we were overrun with student interest to explore these topics with us. This turned into one of the most popular electives at the medical school where we were joined by students from medicine, business, engineering, and more. We have an incredible speaker line-up this quarter and will cover current applications of Gen AI in medicine, areas of active research, regulation of Gen AI, legal concerns of medical AI, the VC landscape, and more. Big thanks to Bryant Lin, Longsha Liu, my co-conspirators in making the course happen this year. Appreciate Ambar Bhattacharyya for the continued push to share our learnings widely. We’ll share what we can here if you want to follow along. Let us know what your burning questions are for this growing area in medicine. #generativeai #digitalhealth #aiinhealthcare #aimed

Recent progress in generative artificial intelligence (AI) has given rise to large language models (LLMs) that can be prompted to craft persuasive essays,1 pass professional competency examinations,2-4 and write patient-friendly empathetic messages.5 Amid growing recognition of the capabilities of LLMs, many people have expressed concerns about their use in medicine and health care, citing known risks of confabulation, fragility, and factual inaccuracy.6 As these risks are measured and mitigated, some of the unresolved questions that are coming into focus concern the “human values” that will remain embedded in AI models, both in their creation and in their use, and how the “values of an LLM” may not align with human values even if LLMs no longer confabulate and have been scrubbed of obvious toxic output. Such human values pertain broadly to the principles, standards, and preferences that reflect human goals and guide human behaviors (see the Glossary). As we review here, LLMs and new foundation models, as technically impressive as they are, are only the latest incarnation in a long line of probabilistic models that have been integrated into medical decision making, which have all required that their creators and implementers make value judgments. https://lnkd.in/ejKcVEZH