FDA commissioner Dr. Marty Makary announced a significant milestone: “We’ve achieved our goal earlier than expected and within budget,” he stated. “Tasks that used to take a scientific reviewer two to three days [previously] now take only six minutes.”

What ELSA Offers… And What It Doesn’t

The FDA employs thousands of reviewers, analysts, and inspectors who handle vast amounts of unstructured data like clinical trial documents, safety reports, and inspection records. Automating even a fraction of this workload yields substantial benefits.

ELSA assists FDA teams in accelerating various critical processes. Staff are currently utilizing it to summarize adverse event data for safety evaluations, compare drug labels, create basic code for nonclinical database setup, and pinpoint high-priority sites for inspections, among other duties.

This final function—using data to determine where inspectors should focus their efforts—could influence how the FDA monitors the pharmaceutical and food supply chains and enhances the delivery of its services.

Crucially, though, the tool doesn’t operate independently without human oversight. The system organizes information so that experts can make decisions more swiftly. It streamlines routine tasks rather than handling complex judgments.

Safety Focus: No Industry Data, No External Training

One of the major concerns surrounding AI systems in the public sector involves the utilization of data and third-party AI systems. Makary tackled this issue head-on by stating that “All information remains within the agency. The AI models aren’t being trained on data provided by the industry.”

This stands in stark contrast to AI methods employed in the private sector, where numerous large language models have faced criticism for being trained on proprietary or user-generated content. In corporate environments, this has fueled a growing need for "air-gapped" AI solutions that keep data confined within the organization.

Thus, the FDA’s approach differs from many commercial tools, which frequently depend on open or external data sources. The agency isn’t developing a publicly accessible product; instead, it’s creating a tightly controlled internal system designed to improve its operations.

Other Federal Agencies Are Gradually Advancing with AI

Federal departments have been sluggish in moving beyond AI experimentation. The Department of Veterans Affairs has begun testing predictive tools to manage appointments. The SEC has explored market surveillance AI for an extended period. Yet, few have advanced to full-scale implementation.

The federal government employs thousands of workers processing immense quantities of information, much of it unstructured and stored in documents, files, and even paper. Consequently, AI is primarily being applied to operational and process-related activities. It appears poised to become a crucial component of how agencies handle data, provide recommendations, and execute actions.

Nevertheless, amidst these AI advancements, U.S. federal agencies, particularly the FDA, have experienced reductions in personnel and government contracts. The FDA has encountered notable decreases in staffing levels and programmatic cuts, affecting various aspects of its operations. These AI improvements may help mitigate some of these cuts or potentially amplify their effects on staffing and programs at the FDA.

Makary succinctly expressed that ELSA marks the start of AI integration within the FDA.

“Today’s introduction of ELSA will be the first of many initiatives to follow,” he noted. “This is how we’ll enhance our service to the American people.”

The above is the detailed content of FDA's New AI Tool Cuts Review Time From 3 Days To 6 Minutes. For more information, please follow other related articles on the PHP Chinese website!