BuzzFeed CEO Jonah Peretti recently announced a partnership with OpenAI’s ChatGPT artificial intelligence tool in a move he said would create a “new model for digital media.” Artificial intelligence would become “part of our core business” by “enhancing the quiz experience, informing our brainstorming, and personalizing our content for our audience.” In short, “the creative process will increasingly become AI-assisted and technology-enabled.” BuzzFeed’s stock soared over 200% on the news.
While Peretti cautioned against a “dystopian” outlook, that result does not seem far-fetched. BuzzFeed laid off 12% of its workforce last month, only to replace those workers with ChatGPT. One has to wonder if human creativity is being replaced by computer creativity. Will conventional art one day be created by computers from beginning to end? Will the Oscars have different awards for human versus AI-created content?
All of which is interesting, and perhaps a bit frightening. But how does this factor into the world of intellectual property?
A. Protection of AI-Generated Content
Content that is wholly created by AI is not copyrightable or patentable. Copyright law requires the authoring of a creative work that is fixed in a tangible medium. Many recall a famous 2014 case where monkeys got ahold of a photographer’s camera and snapped a number of selfies. The resulting photo became an overnight sensation, and for good reason!
In the end, nobody owned the copyrights on the photographs—not even the monkeys. This is because copyright law requires a human author in order to be eligible for protection. The U.S. Copyright Office put an end to every primate’s dream with a 2014 memo that read “only works created by a human can be copyrighted under United States law, which excludes photographs and artwork created by animals or by machines without human intervention.” Going a step further, and relevant to AI, the Copyright Office noted “[b]ecause copyright law is limited to ‘original intellectual conceptions of the author’, the [copyright] office will refuse to register a claim if it determines that a human being did not create the work.” (Emphasis ours).
B. The same is true for patent protection, where the 2022 case of Thaler v. Vidal held “We, too, conclude that the Patent Act requires an ‘inventor’ to be a natural person.” Content created entirely by artificial intelligence cannot be protected by a patent.
C. Protection of AI Programs Themselves
AI programs themselves are analyzed under a different lens. These programs are typically created by humans who wrote the underlying source code. And protecting these human-developed creations is critical—ChatGPT has been hailed as a revolutionary chatbot that has driven OpenAI’s market cap to $29 billion in just a few months. But how can OpenAI protect its most valuable asset?
i. Copyright/Trade Secrets Protection
For starters, the source code underlying artificial intelligence software can be protected under copyright and trade secret law. Source code is creative, original, and fixed in a tangible medium, and therefore can be registered as a copyright or enjoy common law protection at the moment it is stored in memory. This, of course, assumes the code was written by a human and not a bot (or a monkey).
A computer program can also be protected as a trade secret if it derives value from not being generally known to others, and if the owner uses reasonable efforts to protect its secrecy. ChatGPT is clearly valuable to OpenAI and presumably is kept under proverbial lock and key to protect its secrecy. OpenAI can claim trade secret protection as a result.
ii. Patent Protection
Patentability of software has been a hot topic for the better part of 50 years. The current rule is that software must be more than an “abstract idea” that is implemented in a general-purpose computer. This rule has been criticized by many as vague and inconsistently applied for software inventions generally. However, there appear to be three successful strategies for writing AI-based patent claims that comply with section 101. All are based on the concept that improving the machine learning model is considered a technical improvement eligible for patent protection.
A first strategy is to claim a new (non-conventional) input that improves the performance of the model. As an example, in-vehicle cameras may sense the driver’s attitude and then use the driver’s attitude as a new input to the model, which may improve the output of the navigation system. In such an example, angry drivers may be routed to a scenic route while calm drivers may be routed through the city. The specification should provide support showing that the use of the driver’s attitude results in improved navigation planning.
Another strategy is to claim an improvement to the architecture of the machine learning model. For example, an image recognition system may use a conventional encoder-decoder architecture. In such an example, an improvement to the encoder-decoder architecture (for example, using a new type of skip connection) or the use of a new type of architecture may avoid rejections under 35 U.S.C. § 101.
A third strategy is to claim an improvement to the training process of the machine learning model. The improvement may include generating or using a new type of training data and/or using a new training architecture. For example, autonomous vehicles may be trained by collecting data while navigating through real-world environments. The training process may be improved if synthetic data, such as data from a driving video game, may be used instead of real-world data. This type of improvement may be patentable if it can be shown that the synthetic data yields an improvement in the accuracy of the autonomous driving model, while also reducing the risks of an accident during training.
Dystopian or not, AI is here to stay and tech companies are no doubt formulating their own versions to target a customer’s preferences or otherwise improve their processes. Protecting these innovations will be critical for differentiating competing companies in an already competitive market.