There will be no special rules for your industry. What gets decided in the next six months to a year about artificial intelligence (AI) legislation, and regulatory action on copyrights and AI, will apply with equal force to AI uses in healthcare, energy and other industries as it does to a consumer using Midjourney to make a birthday card.
On Aug. 30, the U.S. Copyright Office (USCO) issued a notice of inquiry (NOI) seeking comment and guidance on a vast array of AI and copyright (and copyright-adjacent) issues. As discussed below, this NOI is almost certainly intended to guide Congress in the several legislative efforts currently underway. The deadline to respond is Oct. 18.
Three of the big issues being addressed in draft legislation and the NOI are— 1) whether using copyrighted works to train AI and machine learning (ML) models should be deemed infringement or excused under fair use; 2) whether works using AI should be protected by copyright, 3) whether developers of AI models should be required to catalog the materials that were used to train the models.
If restrictions are placed on the fair use defense and/or a compulsory licensing regime implemented for models that used copyrighted content, such restrictions might apply to computer vision models that enable surgical robotics and autonomous vehicles.
AI authorship rules apply regardless of the nature of the claimant. So if the consumer’s Midjourney birthday card is deemed public domain because an AI tool did the lion’s share of the work, so too will the product design created with that tool. https://theconversation.com/dall-e-2-and-midjourney-can-be-a-boon-for-industrial-designers-199267.
If legislation is passed that requires those developing AI and ML models to identify in detail what copyrighted materials are included in the model, possibly destroying trade secrets, that transparency requirement might extend to ML models in the finance industry, where secrecy is important. “The use of proprietary models that cannot be copied is key for traders to retain any advantage, and may drive intentional lack of transparency, adding to the lack of explainability of ML models.” https://www.oecd.org/finance/financial-markets/Artificial-intelligence-machine-learning-big-data-in-finance.pdf
The vast majority of the attention driving AI policy decisions is on consumer-facing tools like ChatGPT and Midjourney – and on the fact that the AI engines enabling those tools were created using billions of copyrighted textual and visual works. Questions about whether this kind of ingestion of the works of authors and artists is, or ought to be, deemed fair use and whether those artists and authors deserve payment or other recognition are tremendously important. However, AI is ubiquitous across most if not all industries, and at least some of the AI engines that drive those industrial AI applications may have similarly involved the ingestion of copyrighted works on the basis of fair use.
The USCO’s AI NOI is most assuredly being issued to guide Congress in crafting AI legislation. We already have several legislative efforts underway, including the Blumenthal Hawley framework. Some commentators have expressed concern, if not alarm, over the depressive effect this legislation might have on the development of AI technologies. See Adam Thierer, Blumenthal-Hawley AI Regulatory Framework Escalates the War on Computation. Drawing fine distinctions between consumer-facing AI and other, less-visible, forms of AI is likely to escape the notice of policymakers without active participation by all stakeholders.
AI is in many ways a creature of copyright. Copyrighted works are used to train many AI models, and ownership of AI outputs might be crucial to certain industries. Participation in the policy dialogue is therefore important for stakeholders across the U.S. economy. The deadline is to respond to the USCO AI NOI is fast approaching, Oct. 18.