Tantra Analyst

Generative AI (Gen AI) can be a godsend tool for content creators. The creative possibilities are only limited by human imagination. But it is not without significant challenges. The concerns around copyright and licensing are already giving jitters to users, and the questions around the authenticity of the content are putting the believability of any content on the internet in peril.

However, some clear trends and best practices are emerging to tackle these challenges. Some forward-looking organizations, such as Adobe, have adopted them proactively and are leading the way.

Copyright/Licensing Challenges of Gen AI And The Menace Of Fake Imagery

Gen AI models require enormous amounts of data to train. This data includes pictures, videos, text, audio and other media. The Gen AI model is only as good as its training data. Currently, there is generally no mechanism to share or transparency around what inputs Gen AI models have used. That absence creates uncertainty regarding copyright or licensing risks. The ongoing court cases against Gen AI startups like Midjourney, Stable Diffusion and Open AI as well as giants like Microsoft are likely only the beginning. As AI-generated content usage expands and awareness about the inputs increases, these challenges, if not addressed, could grow exponentially. The risks will be magnitudes greater for Gen AI models created by social media and other companies with access to vast depositories of public images and other data.

The uncertainty regarding legality could discourage Gen AI usage, especially for enterprises. They may face legal action, financial distress, operational disruptions, damage to their brands and more.

At the same time, we are also starting to see the menace of fake imagery. Currently, identifying fake imagery is painstaking manual work that requires experts. Even that is done after the imagery goes viral and the damage is done. This threat will likely also increase significantly as Gen AI tools become more advanced and easily accessible.

How To Mitigate Copyright And Licensing Risks

The most effective solution to manage copyright and licensing risks is to use clean inputs, i.e., use data to which you have complete rights for Gen AI model training. If you use third-party sources, maintain complete transparency about the source, what terms and conditions they have signed, and so on. For example, note whether they explicitly agreed to Gen AI use and whether the licensing for their content is without any restrictions (e.g., whether it is export-controlled).

Not many Gen AI companies can boast that they have full rights to their training data. The traditional content creators, aggregators or distributors have a distinct advantage here. A great example of such a model is Adobe’s Firefly, which Adobe claims is trained on its vast collection of licensed stock content (or public domain and openly licensed content). Getty Images also claims it uses its own creative content and data for its image generation service. I haven’t seen social media companies or other major enterprises claim that.

Another solution to mitigate the copyright and licensing risks is for Gen AI model companies to indemnify customers for any risks. This acts as an insurance policy that seems logical but is not without complexities. For example, indemnity likely can’t be unlimited because of financial reasons. So, users must be careful about the coverage limits, restrictions and exclusions.

Many Gen AI players, including Google, Microsoft, Amazon, Open AI, IBM, Adobe and others, have recently announced indemnification of varying degrees for their models. Many have one or more limitations with lots of fine print. Some have offered it proactively, and others retroactively. It seems Adobe is among the first to offer it proactively. Smaller players without deep pockets typically offer nothing.

So, then, the question boils down to this: Is either of the options—licensed inputs or indemnity—sufficient on its own, or do you need both? The first minimizes the risks, and the second provides peace of mind. But clearly, with so much at stake, both are needed. However, currently, very few players support both.

How To Fight Against Fake Imagery

There is a concerted effort in the industry to identify fake imagery. Even though many companies started working independently, they all seem to be aligning behind a common approach that involves embedding credentials into the content. This is akin to the nutrition label used for food products.

Two entities are currently working together to spearhead that effort. The first is a standards group called the Coalition for Content Provenance and Authenticity (C2PA), with steering members including Adobe, Intel, Microsoft and Truepic. It develops the specifications. The second is a large industry forum, Content Authentication Initiative CAI, that implements those standards. Creators can easily incorporate the CAI implementations into their content.

As I discussed on Yahoo! Finance, both entities together have members across a broad spectrum of industries, including Adobe, Nikon, Canon, Sony, The New York Times, the Associated Press, the BBC, Reuters, The Wall Street Journal, Qualcomm, Intel, Microsoft and more. Recently, Google joined C2PA, and Meta announced that it will adopt C2PA standards to identify AI-generated content on its platforms.

In summary, the Gen AI revolution has started, but so have the challenges. Early trends and best practices are emerging to address and future-proof against those challenges. It is incumbent on the users, especially enterprises, to be aware of the pitfalls, make the right choices and be prepared for their Gen AI journey.

The information provided here is not legal advice and does not purport to be a substitute for advice of counsel on any specific matter. For legal advice, you should consult with an attorney concerning your specific situation.

Prakash Sangam is the founder and principal at Tantra Analyst, a leading boutique research and advisory firm. He is a recognized expert in 5G, Wi-Fi, AI, Cloud and IoT. To read articles like this and get an up-to-date analysis of the latest mobile and tech industry news, sign-up for our monthly newsletter at TantraAnalyst.com/Newsletter, or listen to our Tantra’s Mantra podcast.