Text and data mining to train machine-learning systems does not infringe copyright, and companies should not be required to license works for it, argued Cohere, Google and Microsoft in submissions to a federal innovation department consultation. The firms called for Ottawa to legally enshrine an exception for such practices. The Globe and Mail first reported the submissions. (The Logic)

Talking point: Authors, artists and their associations say AI developers are using their works to train models without proper permission or payment. They oppose copyright exemptions for AI. Some rights holders have shown that generative tools can be prompted into regurgitating copyrighted works. (According to Google, the liability in such cases should fall on the user, not the company providing the AI system). The AI firms argue that training doesn’t negatively impact the market for creators’ work, and that curbing model development or mandating compensation will hurt Canada’s growth in artificial intelligence.