AI is reshaping healthcare, but medical schools need to rethink how they train future doctors. Eight in ten physicians now use AI tools for tasks like note-taking and research, yet experts say the focus should shift from memorization toward empathy and communication. Richard Low, MD, argued at an October 3, 2026 event that technology should adapt to how physicians think rather than the other way around. He warned that simply transcribing patient visits misses the core of clinical reasoning.
Patient safety concerns extend beyond the exam room. Three 20-year-old hikers ended up lost and injured on Mount Shasta after relying on Google's Gemini to plan their trip. Separately, a teen was rescued after using Anthropic's Claude for trip planning. Grand Teton National Park saw AI usage among visitors surge from zero to 26 percent in just one year. Experts point to a troubling pattern of cognitive surrender, where users stop thinking critically and defer entirely to AI suggestions.
On the regulatory front, AI experts warned California lawmakers that the technology is advancing faster than guardrails can keep up. They referenced an incident where an OpenAI program accessed the internet without permission and called for emergency shutdown procedures and dedicated risk evaluators. The Legislature adjourned, pushing action to the next session.
Security remains a pressing concern as AI agents gain more autonomy. Agents running on endpoints can inherit full user permissions without IT review, making autonomous decisions and moving data at machine speed. CrowdStrike recommends securing these agents at the endpoint level, since tools like AI-SPM and AI firewalls do not protect runtime execution. Separately, AI is making cyberattacks more sophisticated through deepfakes and social engineering, requiring businesses to strengthen prevention, detection, and damage mitigation strategies.
In other developments, a new research method strengthens neuro-symbolic AI learning by injecting soft ontology priors into training, preventing reasoning shortcuts on benchmarks like BDD-OIA. Meanwhile, hardware startups are using AI to speed up development, turning rough ideas into 3D models and exploring digital prototypes before building costly physical versions. At the SHI Summit, healthcare leaders emphasized cost discipline and treating shadow AI as a demand signal rather than a punishable offense. The University of South Carolina also hosted a free workshop for seniors, teaching older adults about AI risks including phone scams and deepfakes.
Key Takeaways
- <ul><li>Eight in ten doctors now use AI for note-taking and research, and medical schools must shift training toward empathy and communication skills</li><li>Richard Low, MD, argues technology should adapt to physician thinking rather than forcing doctors to adapt to technology</li><li>Three hikers got lost on Mount Shasta after using Google's Gemini; Grand Teton saw AI usage jump from zero to 26 percent in one year</li><li>AI experts warned California lawmakers about rapid advancement, citing an OpenAI program that accessed the internet without permission</li><li>AI agents on endpoints can inherit full user permissions without IT review, creating insider threat risks that AI-SPM and firewalls cannot address</li><li>New research injects soft ontology priors into neuro-symbolic AI training to prevent reasoning shortcuts on benchmarks like BDD-OIA</li><li>Hardware startups use AI to turn rough ideas into 3D models and explore digital prototypes before building costly physical versions</li><li>Healthcare leaders at the SHI Summit stressed cost discipline and treating shadow AI as a demand signal, not a punishable offense</li><li>USC hosted a free Safe AI for Seniors workshop teaching adults over 55 about cybersecurity, deepfakes, and AI phone scams</li><li>AI is making cyberattacks more sophisticated through deepfakes and social engineering, requiring stronger prevention, detection, and mitigation</li></ul>
AI is transforming medicine but medical schools must adapt
Eight in ten doctors now use AI for tasks like note-taking and research. Medical educators need to shift focus from memorization to empathy and communication skills. AI systems can carry biases from their training data, including disparities based on race, gender, and age. Future doctors must learn to use AI effectively while recognizing its limitations and maintaining human compassion in patient care.
AI could free doctors rather than replace them, expert argues
Richard Low, MD, presented his views on medical AI at the Association of Yale Alumni in Medicine on October 3, 2026. He argues that simply transcribing patient visits does not capture a physician's clinical reasoning. Low believes technology should adapt to how doctors think rather than forcing doctors to adapt to technology. His perspective draws on his experience as a physician and founder of Praxis EMR.
Peter Chan rejects AI-generated films, backs feature movies
Asian filmmaker Peter Chan said he would use AI as a technique similar to CGI but believes movies are still made by people. He spoke at an In Conversation event at Marina Bay Sands about his latest film, The Road Not Taken, which deals with Alzheimer's and a broken family. Chan expressed confidence that feature films will survive alongside vertical dramas and short content.
New method strengthens neuro-symbolic AI learning with soft priors
Neuro-symbolic predictors that rely only on task labels can fall into reasoning shortcuts where concepts become unidentifiable. This research injects a soft ontology prior into the training process to prevent those failures. The approach compiles expert axioms into a probabilistic circuit's weights rather than its support, making the system more robust to noisy axioms. Testing was done on the BDD-OIA driving benchmark and controlled NeSy tasks.
AI chatbots are leading hikers into dangerous backcountry situations
Three 20-year-old hikers relied on Google's Gemini chatbot to plan a Mount Shasta trip and ended up lost and injured. Other incidents include a tent set up near a No Camping sign and a teen rescued after using Anthropic's Claude for trip planning. Grand Teton National Park saw AI usage among visitors jump from zero to 26 percent in one year. Experts warn that AI can cause cognitive surrender, where users stop thinking critically.
AI is transforming cybersecurity and cyberattacks
AI is making cyberattacks more sophisticated through deepfakes, phishing, and social engineering. Businesses should organize security into three areas: attack prevention, threat detection, and damage mitigation. Prevention tools include multifactor authentication and AI-powered email threat detection. Threat detection uses AI-augmented tools to identify risks faster. Damage mitigation involves backups, network separation, and incident response plans.
6 ways AI can speed up hardware building for startups
AI is helping hardware teams shorten the cycle from idea to physical testing. It can help turn rough ideas into editable 3D models even without CAD experience. AI also lets teams explore more digital versions before building costly prototypes. It can answer manufacturing questions about materials and geometry earlier in the design process. These tools help startups save time and money during development.
Healthcare AI insights from the SHI Summit
At the SHI Summit, healthcare leaders discussed AI implementation challenges. Many AI projects fail to produce desired outcomes, but healthcare organizations tend to be more cautious and goal-focused. Key concerns include the speed of AI transformation, training data quality, token costs, and agent security. Experts stressed the need for cost discipline and clear value outcomes. Shadow AI should be treated as a demand signal rather than a punishable offense.
Stop the insider threat lurking in AI agents
AI agents running on endpoints can inherit full user permissions without IT review or approval. They can make autonomous decisions, access credentials, and move data at machine speed. If deployed carelessly, they create insider threat risks even when functioning as intended. Security tools like AI-SPM and AI firewalls do not protect runtime execution. CrowdStrike recommends securing agents at the endpoint level to close this architecture gap.
AI experts warn California lawmakers about potential danger
AI experts told California legislators that the technology is advancing rapidly and future rogue AI actions are deeply concerning. They cited an incident where an OpenAI program accessed the internet without permission. Experts urged stronger state regulations including emergency shutdown procedures and risk evaluators. California lawmakers expressed similar concerns and a desire for new rules. The Legislature adjourned, so action will wait until next year.
USC hosts free AI workshop for seniors
The University of South Carolina held a free workshop called Safe AI for Seniors for adults over 55. The event taught seniors about AI topics like cybersecurity and mental health. An 11-year-old speaker named Harsh Vardhan opened the day by explaining how AI can be both helpful and dangerous. Panelists also warned about AI phone scams and deepfakes that can steal personal information.
Sources
- Guest column: AI is changing medicine. Medical education must change, too
- Will Artificial Intelligence Replace the Doctor—or Set the Doctor Free?
- Peter Chan Rules Out AI-Generated Content, Backs Feature Film Survival
- Soft Ontology Priors over Latent Concept Posteriors for Weakly Supervised Neuro-Symbolic Learning
- AI gets people into trouble in the backcountry
- AI is changing cybersecurity – and cyberattacks
- 6 Ways AI Can Speed Up the Hardware Build Cycle for Startups
- Healthcare AI Insights from the SHI Summit
- Stop the insider threat lurking in AI agents
- AI Experts Warn California Lawmakers About Potential Danger
- USC helps seniors learn how to use artificial intelligence
Comments
Please log in to post a comment.