Generated

Despite my optimism for Generative AI and its floor-raising use cases, we might have already peaked in AI hype. As much as Wall Street analysts and venture capital firms seem detached from regular people, they’ve already started to sound the alarm on AI overinvestment, citing that the sector’s valuations cannot be justified given projected productivity gains, and that it’d take an order of magnitude more of AIs-replacing-human-labor efficiencies to earn breakeven returns on the money already poured into the space. In other words, everyone already knows it’s a bubble, but nobody wants to predict when it’ll burst.

Maybe it’s because so much of tech is experimenting and incorporating Gen AI into its products and content, but there’s a certain look and feel—a level of genericness and plainness—which have become telltale signs of this artificial generation. In the past, the weird human hands and overly flat coloring gave away the machine; now the flaws are a bit more subtle, pictures that are just too clean or include nonsensical details that are less evident on the first pass1. Textual generation suffers from similar signs: the writing has almost an aggressively neutral tone and cadence, going almost exactly “by the book” but missing a lot of a human author’s personality and voice. It’s why the tool works so well for high-school-level essays—most kids haven’t developed strong, distinctive writing styles yet, and sound just as formulaic as ChatGPT.

Then again, these are well-known and expected use cases. The challenge in the coming years is finding, and acclimating, to Gen AI fluff in all areas that aren’t exactly known for high-quality output that previously required some human effort. Self-published Kindle e-books, say. Or comment threads on X/Twitter. AI automation is driving down the cost of these different avenues of spam, and I have to think that much like these platforms are working to shore up their detection and filtering, akin to email providers’ eternal battle against spam.

The other concern is that it’s easy to identify simple and lazy applications of Gen AI, but good uses are much harder to notice and may already be deployed to wreak havoc. Like movie and television visual effects, while bad CGI stands out as fake and unnatural, good CGI doesn’t get noticed, since the entire point is to seamlessly blend in with the rest of the scene. For a bit, art and photography competitions had—and still kinda do have—trouble distinguishing between real photos and paintings versus AI-created ones, particularly since participants keep on submitting disqualifying images to make a point. Some AI-powered email responses sound just like what someone would quickly type out on their phone.

Perhaps we’re fretting too much about whether these outputs are manually created, as opposed to crafted with AI assistance. I guess there’s a sense of cheapening the artwork or prose when it’s iterated out of large language models, but at least for now, their outputs still require aggressive human curation and tweaking. But just as there’s still value in trivia in the age of Google…even if AI takes over the rote creation of audio and video, images and text, there’ll still be plenty of space for, well, all the non-rote art.