Copyrights in the Era of AI

At the heart of Adobe’s nearly 40-year history is our support for creatives—enabling artists and authors with the latest technology that brings big, bold ideas to life. And throughout, we have strongly supported robust copyright protections. We license millions of high-quality images, graphics, videos, 3D objects and templates that are curated and integrated into Adobe products.

Today, advancements in artificial intelligence (AI) are revolutionizing the creative process. The use of AI in Adobe’s products and services helps creative professionals do their work in totally new ways: a graphic designer can now use AI-assisted search to find relevant stock images more easily; a filmmaker reviews footage more efficiently and gets recommendations on possible edits; and creative professionals save time by having the products they use most often, such as Adobe Photoshop, customized to their skill level and areas of focus.

This new world of AI-powered creativity has unlocked opportunities, which are only possible with accessible data to train AI systems. These AI systems and models are “trained” by analyzing massive amounts of data to identify patterns, which automate specific tasks and decision-making. As more data runs through the model, its results become more efficient and refined.

In many cases, the data required for AI systems often includes copyrighted content. This process—usually referred to as “machine learning”— raises a fundamental intellectual property question when the training data includes copyrighted material: does integrating protected material in the process of training an AI model constitute infringement?

The answer to this question lies in the value of the copyrighted material. The fair use doctrine holds that if copyrighted material reduces the market value of the work to its original creator, it is unlikely to be considered fair use. Generally, accessing copyrighted works for use in training algorithms does not reduce the economic value of the work in any measurable way. And, if a tool powered by the algorithm is used to create something totally different, the value of the copyrighted material remains similarly unchanged.

While treatment of copyrighted material seems straightforward, the legal framework for how to treat copyrighted material and AI has been left unaddressed. Several cases – most notably the litigation surrounding Google Books – suggest that using copyrighted works for the non-expressive purpose of training AI models amounts to fair use. But without specific guidance from courts or regulators, researchers and developers in the field of AI and machine learning lack the kind of certainty that is needed for AI to flourish.

This fact begs the obvious question: what is the right approach to establishing an appropriate legal framework for treating copyrighted material in AI systems? Well, we can look to a few countries outside of the United States for clues on how to realize the benefits of permitting the use of copyrighted material to train AI. The Japanese government, for example, recently updated its copyright laws to include exemptions of the use of copyrighted works for machine learning. Other countries, including China, Australia, Singapore, Thailand, are looking at making similar changes. Additionally, the European Union recently adopted limited text and data mining exceptions as part of its Copyright Directive and continues to explore further refinements.

Currently, the United States is a world leader in AI but could quickly fall behind if our policies unnecessarily hinder or limit access to data. Setting up a more definite legal or regulatory framework for this type of activity would enable AI to continue to flourish in the United States.

Policymakers and regulators should avoid an all-or-nothing mentality. On one end, loosening copyright standards across the board would strip away the necessary protections content creators rely on. On the other end, imposing overly stringent copyright rules would stifle innovation. Rather, they must strike a balanced approach—one that protects intellectual property rights without impeding the development of AI tools and programs.