Instead of that: “Can’t we just get A.I. to do it?” croaked Elon through a haze of sweet-smelling smoke, probably. Yep, it’s time for another rant. But before we get into the specifics of how A.I. will or won’t affect car design let’s take a look at the broader trends of what’s going on. The tech industry, flush with cash for years and hungry for The Next Big Thing, didn’t look at a single industry they thought couldn’t be improved by their disruption and insight. Sometimes it’s succeeded. Digitization of information has revolutionized how we consume popular media; no more Friday night trips to Blockbuster Video. We all carry a personalized, powerful and configurable personal assistant and communication device with us all the time. In others, they’ve failed. Self-driving cars. Battery-swappable EVs. The Internet of Things (dammit, refrigerator, I keep telling you I only drink San Pellegrino!) I’m not talking about specific formats but broader ideas and the arrogance that they could take something and make it better by the brute-force application of technology and turning it into a subscription-based service. Juicero is my prosecutorial star witness in how to spunk $120 million of other people’s money in the pursuit of solving a non-problem.
Since the Industrial Revolution, advances in technology have improved the processes used to build things. Line workers today can reasonably expect not to clock off by being mangled inside a huge piece of industrial equipment. Robots now carry out repetitive and dangerous tasks to a high degree of quality, which is what machines excel at. But there is a limit to how much assembly of a car can be automated, as Tesla has found out— for some tasks you still need humans. This means it’s hard to save more money by automating further. For most products (with a few significant exceptions) the profit is made at the beginning of the process in the design, and at the end in the marketing and sales. The bits that require creative minds in other words. Where there’s money to be made, there’s money to be saved, so no wonder this is where the tech companies are setting their sights next. Society in general has always had trouble seeing the value in the creative industries, even though the output is what makes life interesting—and gives choice to us consumers. I always tell my students never to work for free; an internship should be paid, and if a company values your work enough to use it, they should compensate you for it. Exposure doesn’t pay the bills. This is why the design of a lot of automotive start-ups and custom or accessory suppliers is sometimes shocking. They don’t want to pay the going rate of contract designers and modelers because of the expense. Like good lawyers, we bill by the hour, and you won’t always like what we have to say. If the design process can be distilled down to an algorithm, you’re only paying for a few hours on a CPU farm. Much cheaper, and it should do what your outsized CEO ego tells it to do. The race towards autonomous driving has driven a lot of the machine learning that underlies AI image creation. The two most common generative art AIs, Midjourney and Stable Diffusion have been trained on the open source LAION-5B dataset, an index of nearly 6 billion image/descriptive text data pairs on the internet, with all the copyright and privacy issues such a massive image fishing trip entails. The AI recompiles existing images into something new based on words you provided as influences. The buzz goes anyone with no drawing skills whatsoever should, with the correct sequence of inputs be able to produce a compelling studio-quality render.
So where does this net out in the car world? This week, to celebrate the 30th anniversary of the Twingo, Renault announced it would build a physical show car based on AI-created images submitted by the community. If I was on the exterior team at Renault, you’d better believe I would be pissed about this. There would be some flouncing out of the studio and perhaps some door slamming. I’m going to quote from the press release directly, because it says a lot: Silicon Valley will save us, because fuck me! Without an app, us Morlocks couldn’t dress ourselves or make coffee in the morning. Remind me again why I spent years studying for a profession that only the most talented get accepted into? Automotive AI is an Instagram account we’ve covered recently that regularly posts some fun thought exercises: What if Walter Gropius designed a car in the 1930s? Or, imagine if the Smart had actually been in production for 70 years. While these are a fun diversion, they rely on taking the familiar and transposing it into something else. Remixing X & Y and crowbarring them together in an unholy bastardization that’s the automotive equivalent of “What if all the Harry Potter characters were in Star Wars.”
It’s the type of low-effort concept shit you can find all over DeviantArt, except that at least was created by actual humans putting effort into what they do. There’s nothing particularly wrong with this, but let’s not lose sight of what we’re looking at. This is not design, because it doesn’t have any context or external drivers beyond wish fulfillment. Car design is an iterative and time-consuming process, not just bashing out a few fancy images. My social media feeds have been stuffed to bursting with AI hot takes recently. So, with David threatening to repeatedly run me over with a BMW iX press car I figured it was high time I entered the room under his frozen garage floor and fired up the Autopian MU-TH-UR 6000 mainframe to give Midjourney and Stable Diffusion a whirl. Because this is highly scientific, journalistically copper-bottomed and not thrown together by the seat of my pants, it struck me the best way to illustrate what a designer actually does compared to what an AI does would be to use one of my own projects as an example. I knew how my inspirations and influences were woven into a final design that emerged organically through trial, and error and a lot of sketching. I had a few different ideas for my final (graduation) project when I was at university. Because it forms basically the entire last year of study and at the end, you have to produce a roughly fifth scale model to display at the degree show. It’s a lot of work and expense. I wanted to punch people in the eyes with something that would stand out and be as individual as me. This is something that AI cannot do. It cannot spontaneously have an initial idea. I came up with an electric hot rod because I thought it would be cool to design and there was a lot of scope for moving the look forward, and it fits in with my self-imposed brief to reflect my personality and style.
Here are the actual mood boards and sketches from my final presentation, 10 years ago. I’m not sure what possessed me to use that font, but hey, I wasn’t the epitome of exquisite taste then that I am now. You can see how the initial sketches evolved from something you could probably see at SEMA into something totally different that captures a lot better what I wanted.
Here’s the final design rendered from the digital model. I chose Midjourney for this experiment because the beta is a little less restrictive in terms of how many images you can create. It works in a Discord server. You type in your prompts and a few seconds later it replies with 4 images.
Here’s what I started off with. I didn’t really give it enough to go on, as these look like bad merch images. Notice how it has trouble getting wheels in perspective and one of the rods has no front suspension.
Second try. I mentioned Braun and Daniel Simon to try and give it a bit more direction. The text in the second image almost says RuPaul, which suggests it can’t tell the difference between drag racing and Drag Race.
This time, I gave it as many prompts as I thought feasible. I mentioned some fashion designers hoping it would reference their work and incorporate those visual signatures into the design. Gareth Pugh designed the jacket on the mood board (I adore his work and he’s an RCA grad like me). Maybe the word hot rod was too restrictive. By taking it out I hoped it would not look at traditional ’32 Ford-style cars and interpret these instructions into something closer to what I had in mind.
I didn’t say rejected Tumbler Batmobile proposals. (On an unrelated note, Matt Reeves: Call me?) Version 4 is somewhat closer. Enhance! You can spend all day doing this, trying to come up with the correct series of words to coerce the AI into giving you what you want. It’s a dice roll every time, the very opposite of considered and careful refinement. It’s not efficient, and in the design studio, time is everything. There are ways of giving the bot more sophisticated instructions by making it reference specific images but that’s getting more into the realms of programming, and my artsy fartsy head brain doesn’t work that way (I’m autistic, and guess what—not all of us are number prodigies.)
As I mentioned earlier, my social feeds are ablaze with AI mentions at the moment. A fellow pro designer and founder of Car Design News, Brett Paterson posted this design for a self-driving taxi on LinkedIn. This was the result of about 80 iteration loops and some Photoshop touching-up. I asked him what his final prompts were so I could try them for myself.
Different but much closer to the desired result than I managed with the hot rod. And a lot more cohesive and believable.
Further iterations. These are all the raw images produced by Midjourney. I’ve not done any editing on these at all. So what have we learned from all this? Basically, with the correct programming, AI could be used to generate alternative design iterations pretty quickly. But it’s a scattergun approach—you never know what you’re going to get. In a worst-case “let’s sack all the designers” scenario you could imagine the correct prompts could become a closely guarded corporate secret like the formula to Coca-Cola, but it’s not likely because the results are too variable, too subject to the unseen whims of lines of code masquerading as something it isn’t. One argument I’ve seen online coming from sponge-brained knucklecocks is that designers will have to adapt and like it. It’s about as well thought out an argument as “learn to code or become homeless” or “get a better job.” The design process has been adapting to integrate new tools ever since Harley Earl introduced full-size clay models into the studio in the 1930s. We’ve had the advent of ever more sophisticated CAD/CAM, 5-axis milling machines, Photoshop replacing traditional pencil and marker rendering, 3D scanning and additive manufacturing, desktop polygonal modeling and now virtual reality. All of these have found their appropriate place in the designers’ workflow. At the risk of sounding like a bleeding-heart socialist European, I believe in equality of opportunity, not equality of outcome. Some of the tools you need to learn have already been democratized. A basic Photoshop subscription costs $20 per month, and there are open-source and one-time license alternative programs available. Polygonal modeling software Blender is free for individuals. Shit, I’ve even started doing sketch tutorials on this very site that you can use for nothing. But all these things are simply tools. You need the vision and ability to understand what you want them to do in the first place. Midjourney can only create within certain limited parameters. It cannot refine, divine, add emotion or context or come up with an idea on its own. It can only reformulate something that already exists in a manner it’s programmed to, nothing else. It can’t look around and write a design brief or see a problem to be solved. A chess program can’t suddenly think up an idea for a new tabletop RPG. I can go to Guitar Center and pick up a starter Squier Strat and practice amp for less than $250. Given enough time and practice I might be able to murder the intro to “She Sells Sanctuary.” But I would never write something like that because that’s not where my creative gifts lie. I’m not Billy Duffy (and yes, I know he plays a white Gretsch Falcon.) The same will almost always be true of AI. AI should do the menial stuff so humans are free to create art, not AI creating art so humans are stuck doing the menial stuff. None of this will stop the world’s tech bros from using AI to replace as many humans as they can, until this technology inevitably fizzles out just like crypto and self-driving cars. But for the rest of you—those that aren’t in the tech-investor class—consider what you’re about to lose in the process. And think hard about whether that’s really a world you want to live in or not.
Here’s What Our Professional Car Designer Thinks Of The Ram 1500 Revolution
Our Grumpy Professional Designer Pleads With BMW To Stop Obsessing Over The 1980s Already
Our Pro Car Designer Shows You How To Sketch Your Dream Ride
Here’s What A Ford Mustang Raptor Would Look Like If Our Professional Designer Had His Druthers
For people like that, AI looks like it will be the latest version of “Good Enough”. Someone else will get stuck with trying clean it up and make it workable, and when they can’t, probably because they were hired for a low salary that matched their lack of experience, they’ll be the one to get booted, not the marketing people who overpromised. I’ve been seeing that for years- digital design and photography came in, and too many people thought it was easy and quick because they could actually use the same software on their own computers, if they wanted to. Then phones had cameras, and it got worse for the photographers. They always need to learn to hard way- it isn’t that easy, and experience matters, and real creativity doesn’t come cheap, unless all you want is good enough. As far as I understand*, this is EXACTLY what those AI are doing. There is no “thinking” because this is not, and was never the point. It just takes the word input, approximates whatever mashup is statistically closer to the tags on the images, pixel by pixel, and show you the n best results. The fact that perspective is always a bit off is telling, as it is drawing from subjects pictured in many different angles – with flexible things like people you don’t notice it as much – but with a car, it shows. As usual, tech bros misunderstand and/or misrepresent things. The AI is not “inspired” by the style of Dali or Picasso – it is literally using their art as statistical reference and mashing them up. The consequence is that, in a hypothetical world were we fire all designers, there will be no more “in the style of Adrian Clarke with hints of Bishop” to draw from, and design will stagnate as a big soup of old styles – for the simple reason that it needs new input to be minced and remixed to create the outputs. Maybe one day AI will be able to project a 3D mesh, suggest engineering decisions and production processes (cower, engineers!), but as TV didn’t kill cinema, and as Internet didn’t kill ignorance, I doubt that AI will truly replace artists, engineers and coders (people think ChatGPT can replace coders). At best, it will be what Excel was for accountants – highly transformative and, ultimately, empowering, but in no way replaces a professional. It may, however, make Joe S. Dumb think he can balance the budget of his startup all by himself…
- I don’t work with AI directly, but I need to interact with it as part of my job, so I have some idea on how this particular sausage is made Just ughhh. Renault I Love You But You’re Bringing Me Down (And Have Been For The Past Three Decades). And “the internet of things” can suck my joystick. Bluetooth refrigerator? The pad and pencil I have been using to make grocery lists with for decades now doesn’t even require a Wi-Fi signal. Screen on the front door? Well… it’s a DOOR. That means I can, you know, OPEN it and, like, look inside the fridge and see what food I need to buy. However, as an automotive design engineer it’s my job to make the pretty pictures work, and the AI produces absolute garbage. I’ve had my problems with human designers not understanding how real things work, but you can, as a last resort, jab crayons deeper and deeper in to their ears until they concede that there will be a number plate on the back and the cab can’t be so far forward that the wheels are inside the drivers legs. What worries me is senior managers creating useless and unhelpful pretty pictures that they then think are ready for production, then throwing 10 million on trying to get a working prototype only to sack the engineers who fail to make it work. Or even worse: they get CAE to productionise their concept and garbage cars become our only option. Still, at the moment AI can’t even make the wheels round half the time, so we have a few months left at least. The entire point of AI is to do GRUNT WORK so that humans can do CREATIVE WORK. Instead we have idiots programming AI to be creative and failing because all AI is capable of is remixing existing creative work, i.e., plagiarism, or derivative design at best. Get a glimpse there. You’ll need to translate. Btw, nice take, as usual. However: Let’s not kid ourselves here. Despite the clear shortcomings of the entire project, and the evils of ‘AI will solve everything!’ tech bros, I think we can all agree that we’d like a Renault Twingo that is also a submarine? I saw them open for System Of A Down at Reading & Leeds Festival 2010… I’m already seeing a lot of pushback on this – I think clients approaching creatives with the expectation that quality work will be done cheaply and within a week are in for a rude reality check, despite all the silicon valley hype. To be fair, a lot of this is on creatives themselves – they have to stand up for themselves loudly, in the same way any decent creative agency of any discipline won’t do unpaid pitches or spec work. When I was looking for work in my field 3 years ago already with 9 years experience as an electrical engineer, some engineering firms were offering me as little as $12/hr, and one of the interviewers called me a “typical entitled millennial” when I rejected such an insult as an offer and explained it wasn’t even enough to move out of my parents’ basement. For engineering work. I continued to wash dishes at the local COVID den for slightly less, while companies all over complained they couldn’t find people with my skillset. I’m currently doing similar engineering work to what that $12/hr offer claimed to entail for more than triple that pay today, at a firm that is actually paying me something that isn’t an insult. Finding that job was not as easy as the news or the official statistics would have you believe. Some of those I graduated college with ended up homeless, AFTER they had experience and patents to their name. God bless ‘mericuh.
