Source: Bloomberg Law

The U.S. Copyright Office has rejected a request for a hacking exemption that would have allowed researchers to circumvent copyright protections to investigate bias in artificial intelligence (AI) systems. The request sought to enable researchers to hack into AI models to study potential biases embedded in their training data, which often consists of copyrighted material.

The Copyright Office ruled that existing copyright exceptions, such as those for fair use, were sufficient to allow researchers to conduct bias investigations without violating copyright laws. However, critics argue that fair use provisions are too narrow and do not provide enough flexibility for researchers who need to access and analyze large datasets to uncover hidden biases in AI systems.

The ruling is seen as a setback for researchers studying AI bias, as it limits their ability to probe the underlying data sets that train AI models. The article suggests that the decision could hinder efforts to address algorithmic biases and ensure that AI systems operate fairly and without discrimination.