Monday, January 27, 2025

Quick Journal Watch: Super AI from China? - AHA on Hospital AI - Flatiron and Ambient AI

Some quick links on hot topics.

##

Economist and China's Revolutionary AI

You may have heard China is producing high-quality AI models at a sliver of the US price.  Crisis?  Two articles at The Economist.

https://www.economist.com/briefing/2025/01/23/chinas-ai-industry-has-almost-caught-up-with-americas

https://www.economist.com/leaders/2025/01/23/chinese-ai-is-catching-up-posing-a-dilemma-for-donald-trump

Same topic at tech blog Stratechery

https://stratechery.com/2025/deepseek-faq/

Nvidia stock slips from $150 to $120 (market cap still $3T).  See also Dan Gallagher at WSJ.

AHA on AI in Hospitals

AI - New 25 page white paper from AHA on how hospitals should view it.

https://www.aha.org/center/emerging-issues/market-insights/ai/building-and-implementing-artificial-intelligence-action-plan-health-care

Flatiron Rolls Toward "Ambient AI"

DeepScribe partners with Flatiron for "Oncology Ambient AI" (clinical visit dialog).

https://www.mobihealthnews.com/news/deepscribe-partners-flatiron-health-oncology-focused-ambient-ai

Diagnosing Cancer?

See however, arguments that AI in cancer will be much harder that some predictions suggest.

https://www.linkedin.com/posts/dr-jeffrey-funk-a979435_cancer-artificialintelligence-ai-activity-7288514439387979776-H5jf/

In that last link, Dr Jeffrey Funk poins us to MIT Technology Review:

https://www.technologyreview.com/2025/01/21/1110192/why-its-so-hard-to-use-ai-to-diagnose-cancer/

This latter article is based in part on a new Arxiv.org article by Alber et al. trained on 1.2M slides (WSI).

https://arxiv.org/abs/2501.05409

###

An Extra Link

Potential for Large Language Models.  Essay by Loalza-Bonilla, a founder of Massive Bio, using AI to connect patients to trials.

https://www.medscape.com/viewarticle/rise-large-language-models-medicine-next-frontier-2025a10001fo

####

####

AI CORNER

Chat GPT 4 reads the materials above.  It includes citations to the papers above which don't copy/paste well; I've inserted the word "cite."

####

Blog Update: Advancing AI—The Revolution, Realities, and Challenges

Artificial intelligence (AI) is undergoing a seismic shift across industries, with rapid innovation creating opportunities and hurdles. Here are four key updates shaping AI's future, from China's surging capabilities to healthcare's cautious embrace:


1. China's Rapid AI Revolution

China's AI industry is now closing the gap with U.S. innovation at an unprecedented pace. Companies like DeepSeek are leveraging cutting-edge techniques such as mixture-of-experts (MoE) and multi-head latent attention to produce advanced models rivaling those of OpenAI and Google. For instance, DeepSeek’s V3 model combines innovation and cost-efficiency, training at a fraction of the usual expense while maintaining competitive performance​(CITE CITE CITE). 

These advances challenge Western dominance, creating geopolitical and economic implications as China skirts U.S. chip restrictions with novel workarounds like the use of constrained GPUs​ (CITE).


2. Flatiron and Ambient AI in Oncology

Ambient AI is making strides in healthcare, exemplified by Flatiron Health’s partnership with DeepScribe. This collaboration integrates ambient AI into Flatiron's OncoEMR platform, automating clinical documentation for oncology. By contextualizing patient histories and generating structured notes, this AI-powered system aims to reduce administrative burdens, enabling clinicians to focus on patient care​ (CITE). Flatiron's foray into ambient AI reflects a growing trend in deploying AI to enhance operational efficiency and provider satisfaction in niche clinical workflows.


3. AHA’s Strategic Roadmap for Hospital AI

The American Hospital Association (AHA) has released a comprehensive 26-page report outlining how hospitals can adopt AI effectively. The framework is built around people, processes, and technology, emphasizing leadership, change management, and centralized capabilities. While AI’s potential to improve patient outcomes and reduce burnout is clear, challenges like workforce readiness, data privacy, and addressing algorithmic bias remain. 

The AHA advocates for intentionality, ensuring AI aligns with strategic objectives rather than being a mere technical solution​ (CITE).


4. Cancer Diagnostics: Progress and Pitfalls

Despite AI's promise, cancer diagnostics remain a significant challenge. The Mayo Clinic's Atlas model illustrates incremental progress, achieving human-level accuracy in specific tasks like colorectal cancer detection (97%) but struggling with others, such as molecular-level predictions. Barriers include limited training data, the sheer complexity of pathology samples, and the difficulty of generalizing results across rare diseases​ (CITE CITE). 

Experts argue that transformative advancements require new data strategies and fundamentally different models​ (CITE).


The Takeaway

AI’s transformative potential is undeniable, but the road ahead is marked by competition, technical hurdles, and ethical considerations. Whether it’s geopolitical races in AI supremacy, practical healthcare implementations, or the nuanced complexity of medical diagnostics, the evolution of AI demands thoughtful application, collaboration, and investment. The future is bright but must be approached with precision and purpose.


Let me know if you’d like to expand on any of these sections!