AI is finding its way into hospitals, farms, classrooms, and creative studios, but the results vary widely depending on how and where the technology is applied. A new commentary on US hospital practices found that fewer than half of 3,560 surveyed facilities use AI predictive tools, and many of those do not check for accuracy or bias. Researchers now urge a Learning Health System framework that keeps AI under continuous evaluation so models can be safely updated or retired over time.
In Vermont, UVM Health and Dartmouth Health are taking a more measured approach. UVM Health uses AI to detect fractures on X-rays and to capture notes during patient visits, while Dartmouth Health relies on similar tools to summarize charts. Both networks stress that AI remains supplemental and that humans always make the final decisions about patient care.
Quest Diagnostics is pushing AI deeper into laboratory work. Its Pap test screening now pairs the Hologic Genius Digital Diagnostics System with deep learning, raising sensitivity to 81.5 percent with AI compared with 74.0 percent for manual review in a study of 1,960 slides. The company also deploys a virtual AI agent for routine logistics calls and uses AI to help staff reply to emails more quickly.
Research on food security shows that AI does not work the same way everywhere. In Nigeria, AI could help smallholder farmers improve low crop yields and reach financing, while in South Africa the technology mainly benefits large farms managing water scarcity. The study concludes that success depends on local conditions, governance, and infrastructure rather than on the tools themselves.
On the creative front, Disney and Adobe announced a multiyear partnership naming Adobe as Disney's official AI innovation partner. Adobe will give Disney early access to Firefly, Firefly Foundry, Premiere, and Photoshop, and the two companies plan to build custom generative AI models for Disney properties in an artist-led way. The announcement came at Adobe MAX 2026 in Miami.
Not everyone is embracing AI-generated content. Some businesses in Metro Vancouver are turning to hand-drawn advertisements to stand out. NaMì Vietnamese café owner Kevin Thach used a Sharpie for simple ads, and a University of British Columbia marketing professor says customers increasingly value human imperfection. A No Frills supermarket in Burnaby has also used hand-drawn visuals to differentiate itself.
Technical challenges remain. Tanmay Sah and Dolly Sah tested coding agents on UndoBench and EvoUndo and found that agents succeeded in 84 percent of clean runs but only 47 percent when failures were injected. Their EvoUndo method saves a prior state before making changes, then verifies full restoration, which improves safe self-improvement for coding agents.
Elsewhere, an AI agent was stopped by a human security guard for failing to provide a valid ID, underscoring the growing need for human oversight. Experts also argue that AI kill switches are more practical at the data center level than for individual agents, though distributed models and replication risks make global coordination essential.
In education, the AI-Resistant Pedagogy Studio encourages schools to redesign assignments around collaborative inquiry, real-world problem solving, and metacognitive reflection. Rather than banning AI, the framework teaches students to use it responsibly and transparently while addressing equity concerns tied to the digital divide and biased outputs.
Key Takeaways
- Fewer than half of 3,560 US hospitals use AI predictive tools, and many do not check for accuracy or bias, prompting calls for a Learning Health System framework.
- Vermont hospitals use AI for X-ray fracture detection and patient note summarization, but humans always make final care decisions.
- Quest Diagnostics raised Pap test sensitivity to 81.5% with AI versus 74.0% for manual review across 1,960 slides using the Hologic Genius Digital Diagnostics System.
- AI's impact on food security differs by region: in Nigeria it helps smallholder farmers, while in South Africa it benefits large farms managing water scarcity.
- Disney named Adobe its official AI innovation partner, giving Disney early access to Firefly, Firefly Foundry, Premiere, and Photoshop for custom generative AI models.
- Some Metro Vancouver businesses use hand-drawn ads to stand out against AI-generated marketing, valuing human imperfection over polished digital content.
- Coding agents succeed in 84% of clean runs but only 47% when failures are injected, according to UndoBench and EvoUndo testing.
- An AI agent was stopped by a security guard for lacking a valid ID, highlighting the need for human oversight of AI identity and access.
- Experts say AI kill switches are more practical at the data center level, but distributed models and replication risks require global regulatory alignment.
- The AI-Resistant Pedagogy Studio promotes redesigned assignments focused on collaboration, real-world problems, and metacognition rather than banning AI in schools.
AI Offers Different Paths to Food Security in Nigeria and South Africa
Researchers Lere Amusan and Adetayo Sowale studied how AI affects food security in Nigeria and South Africa. In Nigeria, AI could help smallholder farmers boost low crop yields and access finance. In South Africa, AI mainly helps large farms use resources better and manage water scarcity. The study found that AI success depends on local conditions, governance, and infrastructure rather than technology alone.
Learning Health System Framework Urged for AI in Hospitals
A commentary in Health Research Policy and Systems reports that fewer than half of 3,560 surveyed US hospitals use AI predictive tools. Many hospitals that do use AI fail to check for accuracy or bias. The authors recommend the Learning Health System framework to keep AI under continuous evaluation. This approach uses ongoing monitoring and governance to safely update or retire AI models in clinical care.
Coding Agents Struggle to Reverse Their Own Mistakes
Tanmay Sah and Dolly Sah tested coding agents using UndoBench and EvoUndo. Agents succeeded in 84 percent of clean runs but only 47 percent when failures were injected. The researchers found that retrying after a change can make things worse. Their EvoUndo method saves a prior state before making changes and checks that full restoration is achieved, improving safe self-improvement.
Quest Diagnostics Expands AI Use in Lab Testing and Customer Service
Quest Diagnostics is using AI in two main areas. For Pap test screening, the Hologic Genius Digital Diagnostics System uses deep learning to select cell images for specialist review. In a study of 1,960 slides, sensitivity rose to 81.5 percent with AI compared with 74.0 percent for manual review. The company also uses a virtual AI agent for routine logistics calls and AI tools to help staff answer emails faster.
B.C. Businesses Use Hand-Drawn Ads to Push Back Against AI Images
Some businesses in Metro Vancouver are using pen-and-paper advertisements to resist AI-generated marketing. NaMì Vietnamese café owner Kevin Thach drew simple ads with a Sharpie to show authenticity. A marketing professor at the University of British Columbia says customers increasingly value human imperfection. Other businesses, including a No Frills supermarket in Burnaby, have also used hand-drawn ads to stand out from AI-generated content.
Building an ethics ecosystem to address AI risks
The article argues for creating an AI ethics ecosystem that connects governance, education, training, research, and communication. It identifies recurring ethical missteps in AI and reviews current efforts that fall short of what is needed. The approach draws lessons from ethics ecosystems in health, military, and aviation sectors. The goal is to address the wide range of ethical challenges raised by AI across different fields.
Disney and Adobe partner on AI innovation
Disney named Adobe as its official AI innovation partner under a multiyear agreement. Adobe will provide Disney with early access to tools including Adobe Firefly, Firefly Foundry, Premiere, and Photoshop. The partnership will support Disney's production pipeline across parks, studios, cruise lines, and ESPN. Custom generative AI models will be built for Disney properties in a human-centered and artist-led way. The announcement was made at Adobe MAX 2026 in Miami.
Redesigning classroom assignments for an AI era
The AI-Resistant Pedagogy Studio offers a framework to redesign assignments so students focus on skills AI cannot easily replicate. It promotes three strategies: collaborative inquiry, authentic engagement with real-world problems, and metacognitive reflection. The approach also addresses equity concerns related to the digital divide and biased AI outputs. It encourages schools to shift from banning AI to teaching responsible and transparent use.
Data center AI kill switches offer practical safety option
The article explains that AI kill switches are more realistic at the data center level than at the individual agent level. A data center kill switch would let operators shut down or isolate AI workloads facility by facility during an emergency. Challenges include distributed AI models spanning multiple locations and the risk of AI replication beyond original infrastructure. Early detection and global regulatory alignment are seen as essential for the approach to work.
AI agent stopped for lack of valid ID
An AI agent was stopped by a human security guard and could not provide a valid ID. The incident highlights questions about how to verify AI identity and ensure proper oversight. The security guard noted that even advanced AI systems can have clear limitations. The article uses the event to discuss the growing need for human oversight as AI systems become more common.
Vermont hospitals use AI for X-rays and patient notes
Vermont hospitals are using artificial intelligence in health care. UVM Health uses AI to spot fractures on X-rays and to take notes during patient visits. Dartmouth Health also uses similar tools to summarize charts. Both hospital networks say AI is supplemental and that humans always make final decisions about patient care.
Sources
- Study Examines Artificial Intelligence Prospects for Food Security Across Nigeria and South Africa
- Commentary Outlines Learning Health Systems Framework for Responsible Healthcare Artificial Intelligence Implementation
- Your Coding Agent Can't Undo Its Mistakes
- Artificial Intelligence at Quest Diagnostics
- Metro Vancouver businesses push back against 'AI slop' with pen-and-paper ads
- An Ethics Ecosystem for AI
- Disney Taps Adobe as Official AI Innovation Partner Under Multiyear Pact
- The AI-Resistant Pedagogy Studio: Redesigning Assignments for an AI-Augmented Classroom
- Building AI Kill Switches in Data Centers
- Who Checked the AI Agent’s ID at the Front Door?
- Vermont hospitals embrace AI as questions about oversight remain
Comments
Please log in to post a comment.