AI Tools Require Unique Evaluation Measures for Journalists' Usage
The world of journalism is witnessing a significant shift with the advent of Generative AI in the Newsroom. This project aims to develop benchmarks tailored to journalism, focusing on six core use cases: information extraction, semantic search, summarization, content transformation, background research, and fact-checking.
However, a growing concern is that the evaluation methods used by major AI companies may not accurately reflect the capabilities of these models. A study of ChatBot Arena, a widely used benchmark platform, revealed that companies like OpenAI, Meta, and Google test numerous variants of their models privately and release only the best-performing scores, potentially misrepresenting the actual capabilities of the models.
This process, according to the study authors, encourages overconfidence, and there's a recognition that popular benchmark tests used to evaluate AI models' performance fail to capture their real-world capabilities. Performance tests often reward models for guessing rather than declining to answer if they aren't certain.
The BBC conducted its own tests and found that AI tools often distort the content of articles. Building open datasets for newsroom benchmarks raises questions about confidentiality and resources. A recent Muck Rack study found that while AI models cite journalistic sources, little is known about whether they accurately represent reporting, cite sources correctly, or provide sufficient context when repackaging articles.
In an effort to address these concerns, researchers are pushing for a fundamental rethinking of how large language models are evaluated. They advocate smaller, task-based evaluations grounded in social-science methods, prioritizing adaptability, transparency, and practicality, and focusing on "highest-risk deployment contexts" such as medicine, law, education, and finance.
Individual newsrooms should try to evaluate AI tools directly on the tasks they care about most, designing "fail tests" that reflect their editorial priorities. AI researcher Nick McGreivy likened this situation to letting pharmaceutical companies decide whether their drug should go to market.
The field of medicine and law is already developing domain-specific benchmarks for AI models, and efforts are underway for journalism as well. Establishing clear standards for the third-party evaluation of AI models in journalism is crucial for ensuring responsible and trustworthy uses of AI. As AI continues to play a growing role in the newsroom, it's essential that we prioritise accuracy, transparency, and ethical considerations in the development and evaluation of these powerful tools.
Read also:
- User Data Analysis on Epic Games Store
- AI Inspection Company, Zeitview, Secures $60 Million Funding for Expansion
- Ongoing trade friction as the American administration levies fresh import taxes on goods arriving from China
- Tech Titan Google Surpasses $3 Trillion Market Cap as Gemini Ousts ChatGPT, Boosting Google's Chat Platform
