AI companies want to self-regulate, but they’ve already proven themselves unfit

Keith Kupferschmid

Submitted photo/InsideSources

AI companies that put short-term gains ahead of long-term legitimacy and the rule of law hide the evidence rather than explain their actions and denigrate the human creative spirit rather than respect it. In (Maya) Angelou’s words, they have shown us who they are. We should believe them.”

Tell Us What You Think

What are your thoughts on artificial intelligence? Tell us in 300 words or less at yourvalley.net/letters, or email your comments, name, address and contact number to AzOpinions@iniusa.org.

Related

AI industry insiders voice alarms to NYC council as local governments take up safety concerns

Opinion

AI following old script of delayed prevention

Opinion

Arizonans must proactively protect themselves as AI rapidly evolves

Opinion

Living with AI: Power, risk and responsibility

OPINION — Concerns about possible dangers of artificial intelligence are everywhere — from fear of costly data centers, to chatbots that devastate users’ mental health, to the industrial-scale theft of creative works that made these products possible. 

What’s less clear is how to address them.

Congress and state capitols are awash in legislative proposals. Government regulators are busy passing standards and releasing white papers. In the copyright arena where we focus, more than 140 cases have been filed in federal court by creators of all sizes and types to address mass infringement by a number of giant AI developers and platforms.  

So-called AI “accelerationists” reject all this, claiming the most effective way to regulate AI would be to let the AI companies do it themselves, backed up by a White House determined to “win” AI, whatever the cost. Recently in Washington, top AI executives announced a new AI “constitution” where they agreed to police themselves. 

The idea that Big AI would effectively regulate itself was never very persuasive. New documents unearthed in copyright litigation against the AI giants make clear self-regulation could never, ever work. 

As author and poet Maya Angelou famously said, “When someone shows you who they are, believe them the first time.” 

Well, in a slew of documents released in litigation against these companies by book authors and news outlets, the leaders of America’s dominant AI companies tell us over and over again they will not follow the rules. 

The documents discuss the need for large volumes of books and other text, what they dehumanizingly call “data,” to train new models and reflect an intentional, knowing decision to get it from illegal piracy websites and then hide the crime. “We trained GPT-3 on pirated stuff! No sharing that!”  

The documents make clear the problem wasn’t difficulty finding legal ways to acquire training data, they simply didn’t want to pay.

OpenAI founder Greg Brockman acknowledged, “We can also purchase a lot of books, but they are expensive … so we haven’t prioritized.”

It’s no surprise Brockman, who testified that he was in this business to earn “gazillions,” also approved circumventing internet paywalls to get access to even carefully protected training materials without pay. 

And later, when it became clear training on “copyrighted data from sketchy Russian website” would create legal and public relations risk, OpenAI launched Project Clear to scrub references to their illegal piracy from company systems. 

The filings also show an almost cartoonish disdain for authors, their work and the long-term health of human culture and the global information ecosystem. OpenAI’s policy director acknowledged that “our work in this area will make people unemployed.” 

In addition to being staggeringly offensive, those concessions have a great deal of legal significance: When illegal copying substitutes for the products of human labor, the AI companies’ “fair use” defense is far more likely to fail. They cut deeper than that, reflecting a callous profits-over-all attitude that reveals just how little these companies care about their project’s effect on creators and others.  

At one point, an AI executive warned that by displacing human authors and journalists and making it harder to find quality works for training in the future, “Our AI content strategy has started a ‘doom loop’ that will hurt the performance of our models and the entire web at the same time,” but the company continued nonetheless! What kind of leader chooses short-term gains when the cost is long-term harm to their products and potentially the entire web?  

Another OpenAI executive was deeply committed to the replacement of human authors to the extent he fantasized about George R.R. Martin, a plaintiff in the lawsuit, dying, saying he would “rest easy knowing  ... GPT-5 will autocomplete his series.” Another time, OpenAI researcher Tarun Gogineni mused about “the death of reader” as “machines created slop for more machines,” dismissing author warnings about lost jobs and damage to the culture as “acceptable economic disruption.” 

Together, the documents reflect an industry that even in its earliest days had already committed to a move-fast-and-break-things approach, with fantasies of gazillion-dollar IPOs replacing any serious commitment to responsible development, copyright and other laws, and even the long-term performance of their own products. 

AI companies that put short-term gains ahead of long-term legitimacy and the rule of law hide the evidence rather than explain their actions and denigrate the human creative spirit rather than respect it. In Angelou’s words, they have shown us who they are. 

We should believe them. 

Keith Kupferschmid is the chief executive officer of the Copyright Alliance. He wrote this for InsideSources.com. Please submit comments at yourvalley.net/letters or email them to AzOpinions@iniusa.org. We are committed to publishing a wide variety of reader opinions, as long as they meet our Civility Guidelines.