It’s Time To Talk About Artificial Intelligence (AI) In The Automotive Media Space

My colleague Jason Torchinsky recently texted me an article written by me, except I didn’t write it. AI did, using my very own voice. Reading this, it became clear that we need to talk about AI, what role it will play in automotive media, and whether it’s going to destroy all of the car publications you know and love.

Right away, let’s show the writing sample. The prompt was: “can you write a 250 word article about a jeep cherokee XJ in the voice of journalist david tracy from the autopian“.

And here’s what ChatGPT spit out:

Let me be absolutely clear: the Jeep Cherokee XJ is one of the greatest vehicles ever engineered by humans. Period. It is the mechanical equivalent of duct tape—simple, rugged, and capable of holding your entire life together under appalling conditions.

Designed by American Motors and launched in 1984, the XJ wasn’t just a truck slapped on a wagon body. It was the first unibody SUV, a lightweight marvel that handled like a car but could crawl up a cliff face in low-range 4WD. You get solid axles front and rear, coil springs up front, leafs in the back—none of this modern independent suspension fluff. It’s honest. It’s serviceable. It’s a rolling monument to function over frills.
Under the hood? The 4.0-liter inline-six, AKA the engine that refuses to die. This cast-iron deity will outlive us all, sipping oil and spitting horsepower through a transmission that sounds like it’s full of ball bearings and hope.

I’ve wrenched on dozens of XJs in frozen Michigan driveways, and I can tell you: everything about them is engineered to be fixed with hand tools and determination. It’s the SUV you buy not because you want Bluetooth and a backup camera, but because you believe in mechanical truth.
Yes, the interior plastics dissolve on contact. Yes, the rear hatch sags. But guess what? It’ll still start, climb, and get you home when your crossover-owning friends are waiting for AAA.
God, I love this thing.

Though I could see how someone might think I wrote this, I know that this is not my writing. I wouldn’t start an article with “let me be absolutely clear.” I think “ever engineered by humans” is weird. I also think the XJ holding one’s life together is a strange thing to say.

The Cherokee also wasn’t the first unibody SUV — does AI not know of the existence of the incredible M151 MUTT military vehicle? This thing predated the XJ by a quarter century!

“Handled like a car?” What the hell? Robot, have you driven an XJ?

Also, when discussing the 4.0-liter, I almost always use the term “bulletproof,” not “the engine that refuses to die.” I do like “cast iron deity,” though. Wait, why is the transmission full of ball bearings and hope? XJ transmissions, especially between late 1989 through the end of production (2001), were excellent and relatively quiet.

I also don’t get “you believe in mechanical truth.” What does that mean?

Also, the XJ’s interior plastics aren’t great, but they don’t “dissolve on contact,” and what about the rear hatch sagging? Why would I even mention that when the gas struts are a relatively cheap maintenance item?

And why are my crossover-owning friends waiting for AAA? They probably own RAV4s. Those are way less likely to leave them stranded than an XJ.

And I wouldn’t end an article about an XJ with “God, I love this thing.” At least, not these days; maybe when I was younger and in the middle of a trip with an XJ.

Anyway, this was hot garbage.

Jason also sent me a “Jason Torchinsky” AI writing sample with the prompt: “can you write a 250 word article about a citroen 2cv in the voice of journalist jason torchinsky from the autopian, witty, funny, irreverent humor, lots of facts.” Here’s what ChatGPT spit out:

If a goat and a sewing machine had a baby and that baby grew up to be a car, it would be the Citroën 2CV—the French anti-status symbol so gloriously utilitarian it makes a shopping cart look pretentious.

Launched in 1948 after being delayed by that whole WWII thing, the 2CV (“Deux Chevaux,” meaning “two horses” but really referring to a tax rating, not actual equine power) was engineered with a mission: carry four farmers and a basket of eggs across a plowed field without breaking a single one. That’s not a metaphor. That’s literal French design philosophy, and somehow, Citroën pulled it off with suspension so floppy it makes a waterbed seem uptight.

Powered by an air-cooled, flat-twin engine that sounds like two lawnmowers whispering secrets, early models made a whopping 9 horsepower. Nine. Horses. That’s not enough to get on a freeway, but it’s plenty to get under your skin and stay there like a charming French parasite.

It has suicide doors, hammock seats, and a roll-back canvas roof you can literally tie with shoelaces. The windshield wipers are powered by the speedometer cable—yes, really—so they only wipe when you’re moving. Genius? Madness? Oui.
Produced until 1990 with only the most begrudging design changes, over 3.8 million were made, and yet it remained stubbornly, gloriously basic.

The 2CV isn’t just a car. It’s a philosophy on four wheels. It says, “I’m not in a hurry, and I don’t care what you think.” And that, dear reader, is true freedom.

I like the shopping cart analogy, but I can’t really say whether or not this feels like Torch; it seems just off for some reason. I’ll let Jason chime in here to analyze.

Jason Here! Not A Robot, Just A Damp, Stinky Human

I want to mention first that I was shown those two AI-generated writing samples of David and myself, and I was compelled to send them to David so he’d be as freaked out as I was. Because this is all a very odd thing! I don’t think it absolutely replicated my or David’s voices here, but it didn’t do a bad job. At all! I mean, the goat and sewing machine thing is pretty damn good! And, yeah, the shopping cart line, too. It does feel generally like the sort of stuff I may have written.

I mean, it better, considering that I have literally thousands and thousands of posts I’ve written since 2012 or so, about 13 years of around two or three or occasionally more posts a day for every weekday – it’s a lot. So with that amount of material, this AI Large Language Model has plenty of reference material to churn through to make its little probability guesses about what word I’d be most likely to pick next, and so the result isn’t bad.

But, that said, this brings up all sorts of questions: who wrote this, really? It can only exist because of all the words I’ve written over the years that it has access to – is this a collaboration between man and machine? Not really, I don’t think, because the AI fundamentally lacks any awareness about what it’s writing or what it’s doing. It’s my own words, picked through and sorted and re-combined, and while those combined words, ordered and assembled into something cohesive, can seem to express an opinion or ideas, it’s not that, because there was never an opinion or ideas that started the process from that prompt.

This also means that no fact in anything AI writes can be believed to be true; sure, for lots of facts, it’s probably true, if there’s enough sources on the internet to corroborate the fact, but the AI doesn’t actually know what’s real or what isn’t. And these things are black boxes, in that no one is entirely certain how they’re doing what they do, under the hood.

Maybe it’ll pick the correct facts to use; maybe it won’t. You just can’t ever be 100% sure. Remember just last week or so when some newspapers ran a supplement that included an article with a summer reading list of 15 books? It turns out that AI wrote that, and 10 of the books don’t even exist.

This is a case of some wildly sloppy editing and fact-checking, of course, but the bigger issue is that AI doesn’t care, because AI has no idea what it’s doing, ever. It’ll make up whatever it needs to to fit the prompt, because it simply doesn’t know any better. Because AI doesn’t know anything.

That means an AI-generated article may accurately state that a Renault 4CV has a 747cc inline-four longitudinal rear-mounted engine in one sentence, and then in another state, with equal blind confidence, that the 4CV has a standard supercharger and seats 11, and who the hell knows why it may have decided to say that.

Same goes for AI art: it may make cars that look plausible, but we need to use cars that, unless we’re having fun, actually exist, and AI can’t reliably do that. Remember that AI car parts site that used images like this:

That has elements that look like a Volvo Amazon and Volvo P1800 but it’s not an actual car. Or remember those AI-generated car posters?

That looks like a car with some general similarities to a Subaru Brat, but it’s no Brat. And those wheels! AI has no idea what a Brat is, or why this is so laughably wrong. We can never use AI for car images, because we care about cars being specific cars, not amalgams of what cars of a general type sort of look like. It’s useless.

And, for technical illustrations, AI is even more useless:

This is because technical drawings are of machines and mechanisms that actually need to work in reality, subject to the laws of physics. AI does not exist in this world, nor does it actually understand it. Because it’s just a computer program. And, you know, it’s kind of an idiot, too.

Back To David, A Human

Regardless of how good/bad this AI was, it’s getting better every day, pilfering every bit of Jason and my life’s work from the internet, feeding it into its algorithm, and creating stronger and stronger AI-Davids and AI-Jasons every time we publish a blog or tweet or even shoot video. It’s scary stuff.

But does it mean it will take over our industry? I don’t think so.

Will AI Take Over Automotive Media?

I’ve thought about this a good amount, and I’ve sort of landed on the opinion that AI can only replace certain car media. If a blog is mostly dry and factual — like old Autoblog, for example — then that site is at risk for being taken over by AI. In addition, any website in which bylines are hidden/authors are not prominently promoted is at risk.

A faceless website is a faceless website, whether there are humans or robots pushing the pens.

If there’s one thing I know about human beings, it’s that they love to follow other human beings. Celebrity gossip magazines have existed for ages, people love the NBA because it’s like a soap opera, Formula 1 only took off once viewers got to see the personalities behind the scenes via Drive to Survive, and I could go on and on.

There’s nothing more fascinating to humans than other humans, and that’s just an inherent, unwavering part of our nature. And it’s for this reason that I think The Autopian and other great sites stand to endure. Our website is about voice.

Upon co-founding this website, I put together a “voice preservation rules” section in our writing guide that basically states that editors must follow certain practices in order to avoid contaminating a writer’s voice (you can’t add jokes, you can’t add or delete x number of sentences without telling the author, you should tell the author your headline plan if possible, etc.). Our bylines are on the front page, and we like to write about ourselves.

You, dear reader, are probably here at least in part because you like the people here. You like to follow Jason’s craziness, Thomas’s nerdiness, Mercedes’ deep-dives into campers and diesel trucks, Matt’s excellent and well-informed TMD’s, Lewin’s engineery stuff, and on and on. The Autopian is a collection of writers, which is why, from day one, I’ve done my best to avoid using “we” headlines that have been so popular in journalism. I remember back at Jalopnik, when I wrote about my Jeep J10 the editor made the headline “There’s No New Jeep Pickup Truck, So We’re Just Gonna Rebuild An Old One.” I never understood that; I bought the truck. I’m going to rebuild it.

What’s more, by using the term “we,” you make it difficult for the publication to have a diversity of opinions. Every now and again, I’ll see a commenter say “Wait, how can you say this? You just wrote the total opposite last week!?” And I’m always quick to remind them: No, I did not; that was another author.

Voice is critical; diverse opinions matter; people matter. And I genuinely believe that, no matter how well AI can mimic the “voice” of a human, just by knowing that’s a robot behind the “keyboard” people will lose interest. Because people like following people, and have since the beginning of time.

Of course, right now there are no rules that require publications to disclose when an “author” is actually a human and when it is an AI. I think there should be; if I’m reading a Jason Torchinsky piece, and I believe I’m reading about a short, hilarious Changli-driving comedian in North Carolina, and I find out it’s just a robot, I’ll feel cheated. It’s possible that this is where the world is headed; huge legions of people catfished into following robots on blogs/twitter/maybe even insta/YouTube if AI becomes good enough.

Anyway, these are just my current, incomplete thoughts on AI in auto media. I just wanted to get the discussion going, because I think you, dear readers, likely have some thoughts. I’ll turn it over to Jason:

Jason Here Again, Still Human, Still Stinky

David is right. The whole point of reading anything on our site is because a fellow human being – or, in my case, something close enough – wrote it, based on their own experiences and ideas and feelings and fetishes and opinions. And, in the case of cars, they need to be actually, physically experienced to matter. Any “opinions” about a car written by an AI chatbot are fundamentally useless and absurd, because it has never, and can never, experience any aspect of a car. I may as well tell you what my toaster or rectal thermometer thinks of a vintage Citroën or the new Toyota whatever.

So, I don’t see how we’d ever actually use AI here, other than in contexts where the AI-ness itself was the subject. Or maybe Photoshop’s AI-generative fill tool to expand a background, or something like that. But that’s pretty much it. Our line in the sand is that any content only matters if a human being felt, imagined, experienced, pondered, or whatever’d it. AI can perhaps exist in a tool-like capacity, but it can never be the core of anything we produce.

So there, robots.

Matt Here, With The Cold Hard Reality

If I didn’t care about any of the above, if I was only focused on making money in the short term and wasn’t bothered about creating any long-term value, I’d just use AI for everything. The reality is that if you can get your website on certain platforms (like search engines and social media), you could eventually fake your way with AI into making a website that costs very little to run and is filled with ads. These are called MFA websites, or Made-For-Advertising websites, and they’re the absolute scourge of the web.

But they work. Shortsighted though they are, a good chunk of the web is just a simple calculation wherein the “content” creation has to be cheaper and cheaper because all of the above MFA websites are causing a crash in the value of display ads. This is why so many sites are going out of business or, seemingly, using AI to write certain types of articles (like product buyers guides).

Obviously, we don’t do that. More importantly, that’s not our business model. Our goal is to make content so good that, in the medium-to-long term you’re willing to pay for it (or just to make Adrian miserable). If you’d like to support the site and send a message that AI slop is not what you want, please consider becoming a member.