Teaching AI to Think Like a Doctor

FEATURED RESEARCH

Large Language Models Are Starting to Show Signs of Clinical Reasoning

For years, passing medical licensing exams meant studying late nights and stacks of textbooks. Now, large language models (LLMs) might be getting closer to thinking like doctors, or at least reasoning like them.

Models making the grade: In a recent study at the University at Buffalo, researchers added structured medical knowledge to a standard AI language model with clearly defined facts like “penicillin treats pneumonia.”

With this extra knowledge, the AI was tested against real medical licensing exam questions (USMLE) and scored at or above the level of many medical graduates. The best model scored 95% on the toughest section.

But is it really reasoning? The researchers think their approach goes beyond memorization. Instead, these models use connections between facts, like how a doctor updates their thinking after each test result.

They call this semantic reasoning and argue that if the AI processes information logically and applies learned connections, then it’s really reasoning.

Maybe these AIs aren’t exactly “thinking” like humans, but they’re definitely moving past simple pattern recognition.

They’re handling medical knowledge in ways that are surprisingly like clinicians. Whether or not we call it reasoning, it’s happening sooner rather than later.

For more details: Full Article 

Brain Booster

Select the right answer! (See explanation below)

Opinion and Perspectives

HIDDEN AI

Most Home Care Workers Don’t Know AI Is Already Part of Their Job

AI is quietly reshaping home health care, but the workers most affected by it often have no idea it’s there.

A Cornell study found that home care workers, like nursing aides and personal care assistants, don’t realize their schedules are being managed by AI algorithms.

These systems match workers’ availability, qualifications and location with patient needs. Most workers, however, had little understanding of how AI works or even that it’s being used in their daily tasks.

Risks of ignorance: Home care agencies trust AI because it’s efficient but the frontline workers don’t know the potential downsides.

AI-driven scheduling can unintentionally reinforce biases and discriminate against women, immigrants, people of color and other marginalized groups.

This is a blind spot because these biases affect workers who already face systemic inequalities.

Toward inclusive AI governance: The researchers argue for more transparency and oversight.

They recommend “stakeholder-first” AI education, teaching workers about AI’s impact specifically in their jobs rather than technical details.

Frontline workers need a seat at the table to shape AI governance. After all, they’re the ones dealing firsthand with its impacts every day.

For more details: Full Article

Top Funded Startups

For more startup funding, read our latest Monthly Report.

Product Pipeline

ORTHODONTICS

AI Comes to Orthodontics as Grin Introduces CoPilot for Smarter Virtual Care

At AAO 2025, Grin unveiled CoPilot, an AI-powered assistant that helps orthodontic practices analyze remote mouth scans, automate tasks, and deliver real-time clinical insights.

Built into Grin’s next-gen Care, Growth, Thrive ecosystem, CoPilot supports more proactive, personalized care while reducing daily workflow complexity.

Together with new features like Grin Hygiene+ and integration with Clarity™ by Solventum, Grin’s platform gives practices a smarter, more scalable way to manage patients.

This results in less chair time, faster decisions, and more confident care without losing the human touch.

For more details: Full Article

Policy and Ethics

AI CLAIMS DENIAL

Lawmakers Demand Physician Review in AI Health Claim Denials

As AI gets more integrated into healthcare, more states are pushing back on its role in one of the most sensitive areas: insurance claim denials.

At least 11 states including Arizona, California and Texas are proposing laws that require a licensed doctor to review any claim or prior authorization denial that involves medical judgment, not just AI.

Proponents say these protections are needed to protect patient care from opaque algorithms that can’t account for individual needs.

While AI may be great at data analysis and workflow automation, doctors and advocates say coverage and treatment decisions belong in human hands.

For more details: Full Article

Byte-Sized Break

📢 Three Things AI Did This Week

• Hugging Face’s Julien Delavande built a tool to estimate chatbot energy use, showing a typical Llama 3.3 email consumes 0.1841 Wh to raise awareness of AI’s environmental impact. [Link]

• South Korea's data watchdog found that Chinese AI startup DeepSeek transferred user data and prompts abroad without consent, prompting a corrective order to delete the data and establish legal data transfer protocols. [Link]

• California’s bar exam sparked backlash after officials revealed AI helped write 23 of 171 scored questions, prompting criticism over transparency and test quality. [Link]

Have a Great Weekend!

Trivia Answer: C) Nose

Your inner ear, big toe, and cerebellum all play key roles in balance. The inner ear detects head movement, your big toe helps anchor you when walking or standing, and the cerebellum coordinates all of it. Your nose? Super useful for breathing and smelling cookies, but not so much for staying upright!