Overview
A new discipline, termed Evidence-based AI, has been introduced in a perspective published in Frontiers in Artificial Intelligence. This framework aims to apply the rigorous standards observed in medicine and toxicology to agentic software systems. The primary intent is to standardize the use of high-stakes AI, with specific mention of its application in toxicology.
Research Context
The perspective, titled "Evidence-based AI: from trailblazer to trustblazer?", focuses on establishing a formal discipline for artificial intelligence. The motivation behind this framework is to integrate established rigorous standards, typically associated with fields like medicine and toxicology, into the development and application of AI systems. This suggests an effort to enhance reliability and trustworthiness in AI, particularly when deployed in sensitive or critical contexts. The paper suggests an evolution from AI as a 'trailblazer' to a 'trustblazer' through this evidence-based approach.
Approach
The research approach involves introducing a formal discipline known as Evidence-based AI. This discipline is conceptualized to integrate rigorous standards derived from existing fields, specifically medicine and toxicology, into the development and evaluation of agentic software systems. The paper outlines the application of these standards to AI, rather than proposing new standards for the AI itself. The explicit goal is to standardize high-stakes AI applications through this framework. The lead author, Dr. Thomas Luechtefeld, collaborated with Dr. Thomas Hartung of the Johns Hopkins Bloomberg School of Public Health on this work.
Findings
The introduction of the Evidence-based AI framework provides a mechanism to apply established rigorous standards from medicine and toxicology to agentic software systems. This formal discipline is presented as a means to standardize high-stakes AI, with toxicology specifically identified as an area for its application. The concept posits a transition for AI from a 'trailblazer' role to a 'trustblazer' role through the adoption of this evidence-based methodology.
Why This Matters
This framework is designed to standardize high-stakes AI, particularly in fields such as toxicology. By applying stringent, established standards from medicine and toxicology, it aims to enhance the reliability and trustworthiness of agentic software systems in critical applications.