AI Models Outperform Doctors in Radiation Safety Tests
Seven AI chatbots scored significantly higher than humans on standardized radiation safety assessments in a recent study.
Updated on Oct. 26, 2026 in Artificial Intelligence
Live Poll
Do you trust information provided by AI chatbots more than information provided by human experts?
A recent study comparing radiation safety knowledge shows that seven AI chatbot systems significantly outperformed human participants. Researchers found that while human accuracy ranged from 32% to 60%, AI models consistently achieved accuracy rates between 80% and 90.67% on a standardized questionnaire.
Why it matters
This research highlights a widening gap between human professional knowledge and machine-based performance in specialized safety domains. The results demonstrate that LLMs now process and recall complex safety protocols with higher accuracy than trained medical professionals.
The study utilized a 25-item multiple-choice questionnaire to benchmark seven distinct AI models against 344 human participants. Microsoft Copilot achieved the highest accuracy, while Perplexity AI ranked lowest among the tested systems.
The players
Microsoft Copilot
An AI assistant developed by Microsoft using OpenAI's large language models to integrate with their suite of software products.
ChatGPT-5
An advanced large language model developed by OpenAI that showed performance variances based on time of day in the study.
Claude
A family of large language models developed by Anthropic that focuses on constitutional AI and safety alignment.
Gemini
A multimodal AI model suite developed by Google for natural language, image, and code processing tasks.
Perplexity AI
A search-focused AI conversational engine designed to synthesize web content into direct, cited answers.
The details
Researchers administered the same 25-item assessment to human subjects, including dentists and medical doctors, and to seven AI chatbot systems: Claude, DeepSeek, Microsoft Copilot, Gemini, Perplexity AI, ChatGPT-5, and ChatGPT-4o. The AI systems process large-scale datasets to identify correct procedural or technical responses based on encoded patterns. While humans rely on training and recall, these chatbots leveraged their internal architectures to achieve superior scoring in the context of radiation safety protocols.
Timeline
October 26, 2026: Article publication regarding the study results.
The Tech Race
This study adds to the growing body of evidence measuring LLM capabilities against human expertise in standardized testing environments. It confirms a trend where AI models consistently clear benchmarks previously dominated by licensed medical practitioners.
This research suggests that AI tools could serve as highly effective supplementary references for medical practitioners in safety-critical situations. Professionals should monitor further testing as these models are increasingly evaluated for clinical decision-support roles.
The takeaway
AI models are now demonstrating higher precision than human experts in specialized radiation safety tests. Watch for future studies comparing model performance against board-certified clinical examinations to determine if these results translate into improved patient-facing outcomes.
Further reading
For more on the performance of generative models, see the latest in Artificial Intelligence.
More information
Read the complete peer-reviewed research article.
Source note: This article includes information reported by Nature.
Live Poll
Do you trust information provided by AI chatbots more than information provided by human experts?
