If you have used the internet sometime over the last year or two, the odds you saw or interacted with content generated by artificial intelligence (AI) are almost certain. Art pieces, memes, news stories, videos, Instagram comments and much more are all being churned out by AI programs at a constant rate – some labeled as such, but most passed off as created by a human.
The text generator ChatGPT and the image generator DALL-E, created by the research organization OpenAI, are two of the most well-known programs, but companies such as Google and Microsoft are creating their own versions as well. The absurd images they produce, like Taylor Swift throwing for a touchdown, can be funny and the answers they print out for homework you put off until the last minute can be helpful, but these AI tools are raising many new, still unanswered legal questions for intellectual property (IP) lawyers, such as Domenica Hartman of Hartman Global IP Law.
“There are very serious concerns with AI,” said Hartman. “Probably the biggest concern is the bias that’s inherent in AI, as the way it behaves depends on the design of the creator for the algorithms being used. Will AI prohibit or minimize thought outside the boundaries of the algorithms used? I also question the security of AI. Who is collecting the data? What is it being used for?”
Hartman’s questions are all based around how generative AI works. Put simply, generative AI are mathematical models that are fed a vast amount of images or text called data sets. These collections often include human labels that tell the AI, for example, that a pile of orange pixels arranged in a certain way is a cat.
The program reads that data set an uncountable number of times. Eventually, it becomes able to identify patterns in words or images on its own – it “knows” what a cat is, what dancing is, what a basketball is because it has seen them so many times.
When you ask DALL-E to make an image of a dancing orange cat dribbling a basketball, it will produce an approximation by predicting what that might look like based on the patterns it associates with each of those concepts.
This process, called machine learning or “training,” is where the questions start to arise for intellectual property experts. The simplest questions, and perhaps most controversial ones at the moment, have to do with the data sets used to train the AI.
Many artists, for example, take issue with the very start of the process. Some AI creators train their programs on data collected without asking the original artists for permission. Some artists equate that with IP theft, or plagiarism – using their work for another person’s profit.
Another common concern is that AI generators are inherently biased, as the outputs they create are based on data sets and mathematical weights selected by their creators. The companies behind the most popular AI tools often keep their specific data sets and training methods as closely guarded secrets. When you ask ChatGPT a question, no one but OpenAI knows how it reached the answer.
Using AI programs also produces valuable data. People around the world ask millions of questions, write millions of prompts, and generate millions of images. Advertisers, for example, would love to know what brands you are asking ChatGPT about in order to help refine their marketing. OpenAI says it does sell your data for marketing purposes but does note that it shares your content “with a select group of trusted service providers that help [it] provide [it’s] services.” Other AI service providers offer fewer guarantees.
Hartman noted there have been some attempts at using AI to examine patent applications, a process that, put simply, determines if an invention and its application is original and within the bounds of the law.
“The results have not compared well to a human examination,” Hartman said. “The human mind and reasoning cannot be replaced.”
The United States Patent and Trademark Office (USPTO) is working to find a middle ground, with AI tools assisting human attorneys rather than replacing them.
“Our trademarks organization is also looking to develop AI capabilities in image searching, classification, and identification of goods,” said Kathi Vidal, under secretary of commerce for intellectual property and director of the USPTO in a 2023 blog. “[This] will better assist our more than 750 trademark examining attorneys examining applications for trademark registrations.”
Generative AI is still very new, but more and more industries are incorporating it into their business models. As this happens, most of the questions raised do not yet have answers, with courts having yet to rule on most aspects of the developing technology. Attorneys and the USPTO will keep a watchful eye on the fledgling technology as it continues to grow and challenge new boundaries of IP rights.
To learn more about Hartman Global IP Law, visit hartmanglobal-ip.com.
