FLARE-AI: A New Era in AI Flaw Reporting
By writing Every week AI Lab means I occasionally encounter AI models that behave badly and strangely. Usually there is nothing to do except share these stories with you. But that could soon change.
Introducing FLARE-AI: A Crowdsourced Solution
A group of AI researchers created a crowdsourced website, Flaw Reporting for AI (FLARE-AI), to report and track harm caused by AI. If, for example, a chatbot generates malware or a recipe for making bombs, discloses personal information, or triggers delusional thoughts in users, FLARE-AI could be used to sound the alarm. The open source code behind the system allows others to verify a problem and forward reports to modelers, as well as to organizations like MITRE, a nonprofit that tracks problems with technical systems. It’s a bit like Downdetector, which compiles real-time user reports on global service outages affecting things like apps and websites.
A Step Toward Accountability in AI
The website is another step in the group’s ongoing work in AI reporting, which I first wrote about last year. Group members also consulted on a congressional bill announced in June, which would see the US government take a central role in tracking this type of AI misbehavior.
“Currently, there is no centralized, accountable way to report flaws in AI systems,” says Avijit Ghosh, an artificial intelligence policy researcher at HuggingFace who co-led the development of FLARE-AI with computer scientists Elaine Zhu and Shayne Longpre.
Collaborative Development and Its Importance
The alarm system was developed in collaboration with 49 AI experts from 32 different organizations. In a paper describing their work, the researchers say their initiative could prove crucial as AI is adopted more widely and agent systems become more powerful. They say the lack of a consistent way to report AI flaws is a significant problem.
“I think it’s a very good initiative,” says Jessica Ji, a researcher at the Center for Security and Emerging Technology think tank. Ji says the researchers are right to note that existing reporting mechanisms are fragmented and AI models are black boxes. “I support anything that makes AI more transparent,” she says.
Beyond Technical Flaws: Addressing Broader Issues
Although bugs and cybersecurity issues are getting a lot of attention, especially lately, Ghosh tells me that issues with AI systems cover topics like psychological harm, discrimination or bias, and misinformation. He adds that different companies have different standards on these issues, meaning some issues go unrecognized. “In the absence of a coordinated disclosure system, there is no external mechanism to enforce transparency,” says Ghosh.
Real-world Implications and Challenges
A series of recent incidents involving popular AI tools shows how easily the technology can deteriorate.
This week, a company called LayerX revealed a way to trick AI-infused web browsers, including OpenAI’s Atlas and Perplexity’s Comet, into strengthening their guardrails. Convincing the AI model behind the browser that it’s playing a game, for example, could cause the browser to turn malicious and attempt to hijack a website. (The companies responsible for the affected browsers have fixed the problem, LayerX claims.) And last April, Johann Rehberger, a security researcher, discovered a way to trick Claude into disclosing personal data using images generated by ChatGTP.
AI also introduces bizarre new types of problems. Last year, OpenAI was forced to update its models after discovering they were too sycophantic, which sometimes seemed to encourage delusional thinking.
Potential and Limitations of FLARE-AI
Rumman Chowdhury, CEO and founder of Humane Intelligence PBC, says FLARE-AI could be a useful way for many AI developers to implement ways to report problems with their tools. But she adds that such initiatives often come with serious challenges.
For more details, visit the original article Here.
“`
