Before satisfying your curiosity about what the two largest “generative AI” machines have to say about copyrighting their texts, it’s worth understanding why this is an important question. Copyright protection for human creators is one of the few government responsibilities specifically mentioned in the U.S. Constitution, right up there with “collect taxes” and “provide for the common defense” (for brevity, I am ignoring other, equally-important intellectual property, IP, such as patents and trademarks). Article I provides federal responsibility “To promote the Progress … by securing for limited Times to Authors … the exclusive Right to their respective Writings” — in effect stating that copyright protection for people is a human right.
During the Enlightenment, protecting the ownership rights of human authors was considered a major public good because it respected individual rights and encouraged more authorship. It’s sometimes forgotten that, previously, almost anyone’s creations could simply be taken away from them and used or destroyed by a king or lord. The important principle was ‘If you, a human person, write it, it belongs to you’ — and this has logically been extended to all sorts of creative works from art to computer software and from individuals to groups of individuals, such as corporations. In some part, as a result of this protection for creative people, copyright advocacy groups estimate that over 10 percent of the U.S. economy — or over $1.5 trillion of our Gross National Product (GNP) — is related to copyright.
So, copyright protection is, by almost any measure, a pretty big deal.
And yet, no one could suppose that in 1787 anyone was thinking about a spinning jenny or a steam engine or any of their descendants, being an “author” with its own “writings.” Through “generative artificial intelligence” (meaning it “generates” written or spoken text, hereafter GAI) computer machines like ChatGPT, New Bing and many others are now generating text. No less than Bill Gates has recently described this technology as revolutionary, and it is generating important questions for the future of copyright.
Immediate and complex copyright questions have already arisen. For example, if a GAI machine creates a text very similar to a copyrighted text, does the owner of the machine — or even the machine itself — own their text and can they hold its copyright? How similar does the GAI’s text have to be in order to be infringing? This is particularly significant because — unlike human copycats — it’s not difficult to program GAI to go right up to the border of infringement but not cross it … thousands of times.
More widespread and immediate questions have arisen over AI-generated art and music that looks or sounds very similar to existing copyrighted creations. And these questions do not even include the complex question of what is called the “fair use” of copyrighted material, which may include such things as scholarship and non-profit uses.
For years, the UN’s World Intellectual Property Organization (WIPO), the U.S. Copyright Office (USCO) and other groups have been studying the emerging impact of AI on the world of copyright. The recent advent of consumer-facing, generative AI has thrust all of this into a sense of urgency.
In 2017, the USCO wrote that a U.S. copyright can only be issued to a work “created by a human being” and that it will reject a copyright application “if a human being did not create the work.” In February of this year, the USCO revoked a previously-issued copyright for images that it concluded had actually been created by GAI without human involvement; then in March, USCO issued a Federal Register Notice on GAI and copyright reiterating and amplifying its position that human involvement is essential for a copyright to exist. In April, the Office will launch a major initiative to further examine how copyright and AI-generated content fit together.
Meanwhile, WIPO created what it calls “a conversation” on the global impact of AI on all types of IP, including copyright, in 2019 which has included scholars, industry representatives, civil society and government agencies from around the world. As might be expected from such a multi-national, multi-topical, multi-stakeholder effort, the result has been creative and far-ranging — and it points to many unanswered questions, such as whether an AI-creation can be copyrighted.
The USCO and WIPO efforts are obviously quite different, although they both suggest at least three different perplexing policy debates that are rapidly emerging: (1) trying to define and enforce infringement in an interconnected, multinational world in which machines can generate thousands or even millions of look-alike materials that are carefully calculated to lie just outside of infringing; (2) trying to define when there is sufficient human involvement in AI-generated material to justify the USCO’s classification of it as copyrightable and, separately, which humans played the creative role and thus hold the copyright to the AI-generated materials; and, perhaps most important, (3) is there a case in which a machine could or should actually own the copyright for something that it created on its own without significant human involvement?
This last question, examined in some of WIPO’s wide-ranging discussions, has obvious and enormous implications for all of copyright and much more: If a machine has a right to own a copyright on material that it alone created, what other rights should/could that machine hold?
So, what do the machines have to say?
My queries were made on March 19, 2023, and both replied “yes” when asked if I could use their replies in this editorial.
ChatGPT replied “As an AI language model, I do not hold any copyright over the content I generate. The copyright … belongs to the person or entity who owns the software or platform that is running me.”
New Bing replied “According to the USCO, only the owner of copyright in a work has the right to authorize someone else … Accordingly, you cannot claim copyright … unless you have the owner’s consent. So, you cannot claim copyright to the text that I write for you … you cannot … use it for commercial purposes without my consent.”
Now it begins.
Roger Cochetti provides consulting and advisory services in Washington, D.C. He was a senior executive with Communications Satellite Corporation (COMSAT) from 1981 through 1994. He also directed internet public policy for IBM from 1994 through 2000 and later served as Senior Vice-President & Chief Policy Officer for VeriSign and Group Policy Director for CompTIA. He served on the State Department’s Advisory Committee on International Communications and Information Policy during the Bush and Obama administrations, has testified on internet policy issues numerous times and served on advisory committees to the FTC and various UN agencies. He is the author of the Mobile Satellite Communications Handbook.