1.5x speed

Planet Money Goes AI

Illustration: Vulture

Would you miss me if I’m gone? That’s the anxious question lurking in the subtext of the latest Planet Money miniseries, a three-parter in the vein of “Planet Money Makes a T-Shirt” and “Planet Money Buys a Superhero” that sees the team embarking on a journey to see if they can create an episode using only generative AI tools. The limitation is taken very seriously, from utilizing a script created by a program trained on the show’s archives to relying on a read generated by an AI voice-over engine that’s been trained on former host Robert Smith’s past recordings. It’s not the first such effort to replicate a podcast using AI tools: The Filmcast crew played with a version of this not too long ago, and the Joe Rogan AI experience exists. But this being a Planet Money project, it’s certainly the most educational effort, with the entire endeavor packaged as a soup-to-nuts learning series in which hosts Jeff Guo and Kenny Malone speak with various AI experts and practitioners who offer some insight into what exactly our automated-media future might look like. Grimes even makes an appearance, sorta.

You should check out the miniseries if you’re at all interested in this stuff, but for now, I’ll cut to the chase. The AI-generated episode isn’t great … but it isn’t bad, either. More importantly, the thing is nearly passable. Had I listened to the final product without context, it’s entirely possible I would’ve thought it was an exceptionally subpar episode that the team phoned in because summer Fridays have started. I would also have thought, Wait, Robert Smith is back? Why does he sound so blitzed?

“What you heard was kind of the best we could do,” said Malone. When we spoke last Friday, he had just returned from a brief vacation off the grid (fitting) and was still trying to sort through his thoughts. On one hand, Malone was despondent about what generating AI is clearly going to do to the future of his/our profession, especially given the rapid rate at which the technology seems to be improving. On the other hand, he continues to be bothered by just how annoying the tools were to use. In his accounting, each step of the process had been a huge struggle, with every artificially generated component requiring a considerable amount of wrestling to get anywhere close to a standard of quality — if they were usable at all to begin with. (Later, it occurred to me that this is probably how my editors feel about first drafts filed by writers all the time.)

Nevertheless, he’s aware it won’t be long before many of those frictions get ironed out, and despite the frustration, Malone was nevertheless fascinated by the fruits of the programs. “It kept generating stuff that was sometimes mediocre and sometimes boring, but then other times it would head off in a direction that was weird but really interesting,” he said. Like, for example, how the program suggested the use of a radio drama as a running thread through an episode. The prompt had contained no such idea.

Generative AI tools are generally talked about as systems that trade in patterns. To oversimplify in explanation, when a particular tool is trained on a model — an archive, a body of work — for the purposes of replication, what it’s broadly doing is constructing a framework out of historical patterns for application on novel prompts and new scenarios. In my mind, the fact that such tools were able to replicate Planet Money’s aesthetic with some relative ease raises a few curious questions about style. What does the automated replicability of a house style illuminate about that style? What’s the line between exercising a house style and being a parody of yourself? Is it unfair to feel that the successful AI automation of a house style somehow … cheapens its value? “What is style but just a set of rules you follow?” said Malone, staring blankly into the Zoom screen. “And does that make us special? I don’t know. Probably not. It’s probably less interesting than we think.”

In the way that the anxious mind does, Malone’s existential spiral has only metastasized over time. “The technology will only get better,” he said. “This is a concern that’s way off in the distance, but it gives me the most anxiety: What is the value proposition of what we do? If what people want from a thing they listen to is a good way to mainline information, it’s going to be much cheaper to produce that very soon, if not already. Why not just feed Wikipedia into ChatGPT, have it spit out a script, and read it back to you in a charming voice? Some portion of listeners already function that way, so presumably they’ll drift away. Then what happens? I guess what’s left is the thing we refer to as the ‘parasocial relationship,’ but I don’t know how important that’s going to be.” Well, there are podcasts whose function is to be artistic works, I offer tentatively, before my own anxious brain remembers it is that very category of podcast that’s already the most squeezed by the recent industry turbulence, and not to mention that some corners are aggressively interested in messing with the terrain around art and artificially generated content.

If you can’t already tell, Malone doesn’t quite know what to think. Of course, neither do I. All I have are childish, fantastical abstractions. I’ve had this image in my head for a while now that we’re all going to end up with a not-so-far future where we all have these little AI bots serving as digital appendages, not unlike little Pokémon running around the internet as our representations. You train your little bot to be more of yourself — feeding them information, speaking with them, editing their output, etc. — and then you let them loose on the world, have them write essays and post things and be horny on main on your behalf. That’s the optimistic sci-fi best-case scenario, probably. The worst and more likely case is that we’ll all be out of a job and thus won’t even have the wages necessary to pay for internet.

It’s cognitively impossible to wrap one’s head around what’s to come. The persistent experience of future shock has become so exponentially shocking it barely registers as anything other than background noise. It’s gotten so bad and overwhelming that I even find it hard to trust how I feel about anything at any given moment. During our chat, Malone brought up a term he wasn’t able to use for fear of overstuffing the miniseries with jargon. He calls it techno-impostor syndrome. “I still believe that I could probably do a better job, or what I believe is a better job, than any one thing the computer was doing, but at the same time, in the back of my head, I’m thinking, But am I?” he said, by way of definition.

I see where he’s going with the concept, but frankly, it’s a little hard for me to relate to at the moment. Lately, I’ve been going through a very human stretch of writer’s block. It’s the kind where I dislike virtually every single sentence I write, including the ones you’re reading right now. All I see is work that’s merely passable. Under these conditions, it’s hard not to be anxious and think that a computer won’t do a better job than me. Or at the very least, that it won’t do a good enough job that’s merely passable. It makes me wonder this: Once everything shakes out, would anyone even notice if I’m gone?

News and Notes

➽ Spotify announced another round of layoffs specifically targeting its podcast division, this time affecting around 200 employees. Claiming it as part of a larger “strategic realignment,” the company is killing the Gimlet Media and Parcast brands, shoving whatever remains into the Spotify Studio umbrella. Gimlet had just won a Peabody and a Pulitzer, but let’s be real: Spotify’s not in the awards business.

It’s a depressing development, but one that’s ultimately unsurprising given how the broader story remains the same. In hindsight, Spotify’s billion-dollar podcast bet only worked out in a few select ways but not in many, many, many others. Facing pressure to improve margins, the company is further cutting down on in-house costs while emphasizing its bet on certain kinds of programming and positioning itself as a YouTube-style platform. Everything else is just collateral damage.

In a statement published on Monday, WGA East, which repped Gimlet and Parcast’s unions, painted a grim picture of what’s been going on over there: “Our unit members and non-unionized colleagues regularly put in late nights, weekends, and holidays to meet grueling deadlines. Sadly, the majority of these employees lost the jobs they toiled at so tirelessly. Our final months were plagued by a lack of direction and transparency, confusion, and announcements that backtracked hours or days after being made.”

Meanwhile, didn’t I hear somewhere recently that the company flew the Ringer’s staff out to Sweden …?

➽ I’m excited for Dreamtown: The Story of Adelantothe latest David Weinberg joint about the rise and fall and rise and fall of a dusty California town …

Planet Money Goes AI