Artificial Intelligence, Real Responsibility

*What follows is not formal legal advice; it’s been nearly 20 years since I took torts and I have not practiced in the area since.*

I experience a strong, almost visceral aversion to the sense that people are getting away with taking credit for their victories and disclaiming responsibility for their failures. In some ways it makes me even madder when there’s a legitimately viable business attached to it, but the level-playing-field success becomes impossible to separate from the tilted advantage (see Uber’s illegal operations in cities).

I first remember being mad about this with Amazon. They did not collect sales tax consistently even on direct sales until they’d been in business for 20 years. The business model of e-commerce book sales had real potential due to convenience, potential economies of scale, the ability to store books in lower-cost locations, and more, but for 20 years they also benefited by charging customers a lower net price than brick-and-mortar competitors. Customers were, of course, supposed to report those sales on their personal taxes, but many likely did not, meaning that Amazon increased its sales volumes in part by facilitating tax fraud and depriving the United States government of tax revenues, thereby increasing the national deficit. If that’s “part of Jeff Bezos’ genius,” it is so only in the Walter White sense of “genius” in successful evasion of the law. Note also that even as they became one of the largest tech companies in the world, they cited the “complicated nature” of state-by-state taxes as a reason not to pay them.

[As a side note, I’m not mad about duty-free shops; they are explicit that their raison d’etre is to live in the loophole, and because of the physical constraints on their operations, they can only displace a small volume of total brick-and-mortar sales. Additionally, to my knowledge their CEOs have not received the acclaim of “visionary, future-oriented” founders in the way that the tech companies’ leaders have. The one exception might be Chuck Feeney, but his plaudits have been more for his philanthropy than for his ostensible genius.]

My antipathy for the leaders of the largest LLM companies right now feels similar; that they’re partly taking advantage of an evasion of liability that human employees could not. I appreciate Dario Amodei’s efforts to think deeply about the potential and risks of AI. He surely has thought about it harder and for longer than I have, and he seems to be genuinely trying to manage its risks. But he has been hyping AI as “country of geniuses” on the one hand, but also expecting to cover that same AI’s errors with the small-font line “Claude is AI and can make mistakes. Please double-check responses” at the bottom of the user page. And then we, the potential users (and/or victims of bad advice), are asked to “help improve the system” by reporting errors (unpaid labor we perform on behalf of the AI companies), just to make it usable enough so that we can tell our bosses (or clients, or others) that we’re finding ways to be efficient and “use AI.” That sounds like privatizing the benefit and socializing the costs to me.

Numerous AI researchers have commented on how the outputs are non-mechanistic, and therefore more akin to something that is “grown” rather than “built,” more similar to a life form than an object. And therein lies a potential partial solution to the problem.