Free AI Safety Tool Lets Anyone Audit Algorithms

Imagine having a say in whether the AI deciding your loan application or healthcare treatment is actually fair. That's now possible thanks to a free tool that lets anyone audit artificial intelligence systems, no tech degree required.

Researchers at the University of Glasgow just released the PHAWM workbench, a groundbreaking tool that opens up AI auditing to regular people. Until now, only technical experts examined AI systems for problems, often missing how these technologies actually affect real lives.

The tool addresses an urgent need as AI rapidly spreads across critical sectors. These systems already influence decisions about housing, employment, finance, policing, education, and healthcare. But they can harbor serious flaws like bias and inaccuracies that enforce unfair outcomes.

Professor Simone Stumpf leads the project bringing together over 30 researchers from seven UK universities and 28 partner organizations. She explains that AI applications hold valuable potential but must be carefully monitored by humans to avoid building systems that create unfair results.

The workbench guides users through a four-stage process anyone can follow. First, organizations describe their AI system in plain language without technical jargon. Then they invite stakeholders like patients, customers, or community members to participate in the audit.

Next, participants align the audit with their real concerns and lived experiences of how the AI affects their daily lives. The tool helps them identify potential positive and negative impacts, create ways to measure them, and assess whether the system meets their criteria.

Each participant gives the AI application a pass or fail grade based on their standards. Organizations then collect these diverse perspectives to make informed decisions about developing or purchasing AI applications.

The Ripple Effect

The tool's flexibility means organizations can audit homegrown AI systems or investigate off-the-shelf applications before buying them. This community-centered approach ensures the people most affected by AI decisions finally have a voice in shaping these powerful technologies.

The PHAWM project launched in May 2024 to support regulations like the European Union's AI Act, which balances innovation with protections against unintended harm. The tool and its framework are now available for free download from the project website.

By examining AI from multiple angles, organizations can make properly informed decisions that weigh risks against rewards. The team continues refining the tool through partnerships with health and cultural heritage sectors, with plans to expand into media and collaborative content generation.

This breakthrough puts AI accountability where it belongs: in the hands of communities who live with the consequences.