This is How Artificial Intelligence (AI) Disrupts the Internet.


Greg Marston, a British voice actor, recently encountered "Connor" online - an A.I.-generated replica of his voice, trained on a recording he made in 2003. This digital clone mimicked his voice, saying things he had never uttered.


At that time, Marston had recorded a session for IBM and had, subsequently, signed a release form permitting the usage of the recording in various ways. However, back then, he could not have foreseen that IBM would utilize his sample for more than just the exact phrases he recorded. Through the power of artificial intelligence, IBM managed to sell Marston's decades-old sample to websites, which were leveraging it to construct a synthetic voice capable of uttering any message. Recently, Marston discovered his voice echoing on the Wimbledon website during the tennis tournament. (IBM acknowledged Marston's concerns and initiated discussions with him directly.)


This situation highlights why many renowned creators in our economy are deeply troubled. We currently face a decline in trust as individuals realize that their contributions to the public sphere may be appropriated, monetized, and potentially used against them in competition. Once this erosion is complete, I fear that our digital public spaces could become even more polluted with unreliable content.


Already, artists are deleting their work from X, formerly known as Twitter, following the company's announcement that it would employ platform data to train its A.I. Similarly, Hollywood writers and actors are on strike, partly out of concern that their work will be fed into A.I. systems that companies could exploit to replace them. News outlets such as The New York Times and CNN have taken measures to prevent A.I. chatbots from scraping their content by adding files to their websites.


Authors are filing lawsuits against A.I. organizations, asserting that their books have been included in the training data of these platforms. In a separate legal proceeding, OpenAI has argued that employing copyrighted data for training A.I. systems falls within the boundaries of "fair use" under copyright law.


While creators of high-quality content contest the use of their work, dubious A.I.-generated content is flooding the public domain. NewsGuard has identified 475 A.I.-generated news and information websites across 14 languages. A.I.-generated music is inundating streaming platforms and generating royalties for scammers. A.I.-generated books, including guides on mushroom foraging that could lead to dangerous mistakes in identifying highly poisonous fungi, are abundant on Amazon. To address this issue, the company is even requesting self-published Kindle authors to disclose whether they used A.I. in their creations.


This situation exemplifies the classic case of the tragedy of the commons, where a shared resource is damaged due to the profit-driven interests of individuals. A common example of this is a public field where cattle graze. When there are no limits in place, each cattle owner possesses an incentive to overgraze, ultimately destroying the land's value for everyone.


The internet also harbors shared spaces, despite its toxic corners. These spaces, such as Wikipedia and Reddit forums, are vital sources for public good, where volunteers generously share knowledge while actively combating malicious actors.


Yet, these shared spaces are currently being overexploited by rapacious tech companies seeking to incorporate all the wisdom, expertise, humor, anecdotes, and advice found within them into their for-profit A.I. systems.


Consider, for instance, the volunteers who build and maintain Wikipedia. They trusted that their work would be used in accordance with the site's terms, which require proper attribution. However, some Wikipedians are now debating whether they have any legal recourse against chatbots that utilize their content without citing the source.


Regulators are grappling with this issue as well. The European Union is contemplating the implementation of the first global restrictions on A.I., which would demand transparency from generative A.I. systems, including summaries of the copyrighted data used for training.


While this would be a positive stride, since many A.I. systems do not fully disclose their training data, transparency alone is inadequate to rebalance the power dynamics between those whose data is exploited and the companies poised to profit from this exploitation.


Tim Friedlander, founder and president of the National Association of Voice Actors, has emphasized the necessity for A.I. companies to adopt ethical standards. He asserts that, regardless of whether individuals are professional actors or simply engage in social media sharing, everyone should possess the fundamental rights of consent, control, and compensation.


Indeed, all individuals require these three essential elements. The option for meaningful consent regarding the inclusion of our online activities in the vast A.I. machinery should not involve the arduous task of locating elusive opt-out buttons, as this seems to be the current trajectory of the industry.


Determining compensation poses a greater challenge, particularly given that most A.I. bots presently operate as free services. However, it is critical to recognize that the A.I. industry intends to generate profits from these systems, and when it eventually does so, it must reckon with those whose creative contributions fueled these gains.


Individuals like Mr. Marston find their livelihoods at stake. He estimates that his A.I. clone has already cost him job opportunities and will significantly impact his future earnings. In collaboration with a lawyer, he is pursuing rightful compensation, asserting, "I never agreed or consented to having my voice cloned, or to witness its release to the public, thereby pitting against myself."


Nevertheless, even those of us who are not directly threatened by A.I. face the prospect of pursuing our artistic endeavors, such as writing novels, composing music, recording TikToks, or sharing humor on social media. Without adequate safeguards against the overexploitation of A.I. data, I am concerned that the very act of creating within the public sphere may lose its purpose. This, undoubtedly, would represent a tragic loss.