Who are these people who want to talk to their car to make it do everything? Who are they, where are they, and what the hell is their problem? And, perhaps more importantly, why is one of these woefully misguided people in charge of Rivian’s software and human-machine experience? And why are they so eager to deeply integrate AI “agentic” bullshit into cars?
I’m asking these questions because of an interview done on The Verge’s podcast Decoder that featured Wassym Bensaid, Rivian’s Chief Software Officer, the same man who once said that physical buttons for car controls are an “anomaly.” Oy. I can already tell I’m going to be cranky about all of this. Bensaid also has stated that “I think the car is actually a fantastic environment for AI,” and that “The final north star I have is having voice [controls] become the primary means of interaction with the vehicle.” So, it’s pretty clear where Bensaid stands: he wants a car without physical buttons, and an AI that you talk to controlling everything.
To me, such a car sounds like a technological triumph that I would be more than happy to roll off a cliff.
It’s also worth remembering that Rivian is one of those companies that makes you control where your HVAC vents are blowing by swiping at a touch screen with a little picture of a dashboard on it, inches from the actual dashboard vents. You know, like how an idiot would choose to control where air blows.
But let’s get back to this AI agent business that Bensaid is so hot on. This is part of why Rivian is so against integrating Apple CarPlay or Android Auto into their cars, even though so many people seem to want that, with many buyers considering it a requirement for any new car they may buy. From Rivian’s point of view, phone mirroring systems like CarPlay or Android Auto are bad because, according to Bensaid,
“The challenge with screen mirroring solutions is that they take over every single pixel in the car, and that’s not the way we see ourselves interacting with our users.”
… which is just a PR-massaged way of saying they don’t want to give up their screen real estate to a company they don’t control. Instead, Rivian –like a number of other automakers – would much rather you only use their own in-house UX for your interactions with the car, and in Rivian’s case, it looks like this will soon become far more voice-focused. As Bensaid says in the podcast,
“I think we are on the cusp of something really big. When you think about it, you’re in a car, you’re driving, you’re focused on the road. So, in theory, the primary interface with which you should be interacting with the car is actually voice. The only reason that drivers and consumers do not interact with the car through voice is that, to put it really bluntly, the technology has been broken. That’s really the beauty of what we have now with the technology disruption coming with foundational models.”
See, I’d have to disagree here. Even if you’re focused on the road, that doesn’t mean voice should be the “primary interface” with which you should be interacting with your car. Driving is a physical task; the primary interface is, and has always been, and should always be physical. Steering, braking, using muscle memory to move your hand to controls automatically – a good interface between a human and car means the car becomes almost a prosthetic. You don’t need that extra whole layer of cognition to put your desired actions into words at all.
Bensaid doesn’t seem to get this concept at all, and I think makes my argument for me when he describes a voice-controlled trunk-opening process:
The foundational models are providing us this wonderful opportunity to truly have a conversational experience where drivers can interact with the car in human language. I don’t need to tell the car, “Open the frunk.” I can say, “Open the front trunk.” Actually, I can say, “I have a bag in front of the car,” and it will actually open the frunk. I think that completely changes the way you interact with the car.
I agree, it does completely change the way you interact with the car. It makes it worse.
Just think about this for a second. Why would you want a “conversational experience” when it comes to getting your stuff out of the trunk, front or rear? Just think about how you normally get your bag out of a trunk now, in your average modern-ish car. You park, you get out of the car, you walk to the trunk at whatever end of the car your bag is in, and you open the trunk. That’s it. You don’t have to tell the car shit.
The car detects your key in your pocket, you push a little button or latch to open the lid, and you’re done. Telling the car “I have a bag in the front of the car” is just adding a useless step. Will the bag levitate out on its own? No. I mean, for most of us lacking telekinesis skills, no. So you still have to go and physically touch the trunk. What the hell is the point of telling the car you have a bag in the trunk? How much extra computing hardware is needed to process and execute that command? For what? Letting the car know you have a bag? So the AI can send that information to advertisers and you’ll see AI-enhanced duffel bag ads for the next three days? Fuck that.
Has Bensaid never been in a car with a friend, in mid-conversation, continuing as you leave the car and get your shit out of the trunk? Of course he has. We all have. Have you ever wanted to pause mid-conversation and tell your car where your bags are? No. Fuck no.
Part of what seems to be going on here is the mistaken notion that somehow your car needs to be doing the same things you already have a phone for. Listen to this bullshit from Bensaid:
“On top of that, we now have the opportunity with all the agentic framework to truly give people their time back in the car. I hope you tried our Google Calendar agentic integration. You can imagine how the experience will be in the future where you’re driving and can perform operations on your calendar. You should be able to perform operations on your email. In the future with the agent-to-agent integration, you can actually interact with many more apps from your own digital ecosystem.”
Calendar integration? “Perform operations on your email?” Why the fuck would you want your car to be part of that? That’s already what your damn phone is for? And these “agent-to-agent” integrations, that just means that some AI bullshit built into your car is talking to the AI bullshit built into your phone so in the end, what is the AI agent in your car doing other than passing along messages to the phone that’s right fucking there with you and if they just let you have the damn Android Auto or CarPlay you could talk right to it? Or just talk to it as it sits on the seat next to you? What’s the point of all this?
You car doesn’t need to uselessly duplicate all the features of your phone. It’s the wrong tool for that job. Your phone is a good personal assistant tool because it’s the size and shape of a well-worn bar of soap and you can slide it in your pocket, not a 4,000 pound hunk of metal and plastic with wheels. Nobody wants their phone to sprout spindly wheels to you can drive it to work, just as we don’t need to use our cars to answer fucking emails.
Here he is again talking about the car replicating phone jobs:
“You can imagine that in the future, instead of having that mono access to every single app on your car — or honestly, even on your smartphone — you can start aggregating and connecting many of those apps through the agentic framework and have them present a unified user experience.”
Yeah, you can imagine that in the future, Wassym, leave me out of it. Who decided we needed a “unified user experience” via phone and car? Let phone do the phone shit, and let your car do the car shit. Phone mirroring is great: all the phone things: reminders, music playlists, calendar stuff, navigation, whatever are available through your car, but using the same interface you’ve been using all day, all the data and settings and preferences still there like you like them, just accessible on your dashboard. It’s fine. Let the phone have that. The car doesn’t need it.
Bensaid describes another situation that he thinks reinforces Rivian’s decision to duplicate the jobs of your phone, but really does the opposite:
“This is how we’re able to connect the navigation to Google Calendar, for example. I can go to the assistant now and say, “I want to plan a trip from San Francisco to San Diego, and I want to have two charging stops. I want them to be close to an Italian restaurant. I love Italian food.” The assistant would go and play that, and then I’ll say, “Okay, print the summary, add it to my calendar, and then send it as a text to my wife.”
Again, why is the car’s AI doing this? If this was all just handled on your phone, it could be done before you’re even in the car. The car doesn’t need to be in this loop at all.
This, I think, is the root of the problem. Car software people want the data and eyeball-access that phones have, and unless you clumsily try to force it to happen with this redundant and inane car-AI-as-middleman approach, it won’t happen. And that’s fine. Really, all of the AI in the car – if there must be any – should be behind the scenes. Like, why would you want a command to change drive modes? The car can sense your inputs, and if you’re stomping the throttle and brake hard, switch to a sportier mode. If you’re being gentle, go to eco. Just adapt based on the driver’s inputs, seamlessly. That level of machine learning seems fine.
But who the fuck wants an LLM to talk to when you’re driving? I don’t need some untrustworthy AI making decisions for me about what music I want to listen to or opening the damn trunk or adjusting the HVAC. No one wants this, no one needs this. Just stop.
Honestly, based on the rate of change of technology compared to the lifespan of cars, why would you want to be locked into some complex car UX or LLM, anyway? Sure, you can update software, but the hardware isn’t going to get better over time. The average age of a car in America is over 12 years old. Who is using a 12-year old smartphone? All this crap should be modular and easily swappable.
Bensaid noted that for a lot of these interactions, the computing hardware will be local to the car, in addition to using cloud-based resources. Honestly, either way has drawbacks: local hardware will eventually be unable to run more modern software, and cloud-based solutions are dependent on connectivity and the health/desires of the company. They could shut features down at will, or discontinue them, or make them into subscription services. Why are we okay with any of that?
I guess I should note that there seem to be plenty of people in China that like talking to their cars and AI in general. I guess this is just one of those cases where 500 million plus people are wrong and I’m right. It happens, it’s okay.
We’re going down a bad path. I’m sure Mr.Bensaid is a wonderful man, a smart man and probably a very tender, generous lover, but I think he is woefully misguided when it comes to how people – real, actual people, not AI-deluded dillholes – want to actually interact with their cars.
I’m going to go out on a limb here, but I still think that people want to drive their cars, not have tedious conversations with them. That’s what friends are for, after all.
(top images: YouTube/The Verge, Rivian)
This sounds like something a child would want after watching too many Marvel movies.
“Deploy Drink Holder!”
don’t give them ideas.
I love the idea of having a conversation with my passenger and casually mentioning that something’s in the frunk brings the possibility of the built-in unturnoffable agentic AI helpfully opening the frunk for me because it’s hallucinating I’m in park while I’m driving 70 mph.
God this AI bullshit is getting exhausting, and it’s so, so, so incredibly stupid.
You know why they’re wanting to build a shit ton of data centers? So they can process all the “AI Agent” interactions we’ll all apparently be having with our cars. You know what data centers use a lot of? Electricity. So all those AI commands have some underlying energy and therefore monetary cost associated with them as your command gets calculated through some LLM until it spits out an answer.
You know what doesn’t have any monetary cost whatsoever once the hardware’s made? A damn button. You can press it millions of times over the life of the vehicle, and it will never cost you another cent. Open the trunk to your heart’s content without accruing any AI usage fees.
The reason these asshats are so bent on this version of the future is that they think they’ll convince you to pay for a subscription to their AI bullshit that will cover the cost of data centers and AI doing a bunch of functions that you used to do for free with a goddamn button.
Which, incidentally, is exactly what big corporations are realizing is a terrible fucking way to spend money. We’re just starting to hear from companies like Uber that the money they’re spending on AI isn’t making them any faster or better, it’s just going towards paying AI to do shit that they already have tools for. Seriously. How many people are having AI run through costly computations so they can send a text prompt to get it to average out a column of data instead of just slapping that shit in Excel?? That’s the kind of thing people are using AI for right now, and it’s complete (very expensive) nonsense.
LLM’s can do some neat things, but everything the current crop of CEO’s pitch it for is shit like this, where it’s an extra unnecessary step that requires time, computation power, and money for no benefit.
great points.
I have only used AI twice ever. And both times were for you wonderful people. First was to help format a sonnet for a Sonett, and ended up using one from England, after many unsatisfactory results, and still took at least 4 revisions till acceptable. The second was to generate a Gonzo image jumping VW busses with Happy clouds in view for Jason, and the first result was just too F’n funny, I sent it.
These are electric cars we’re talking about, if the AI’s housed locally you’d be able to watch your available range plummet with every noise being “thought” through as a possible instruction.
Well said! A connected LLM is useful to them because it’s a way to soak the consumer in perpetuity. But wait, there’s more!
They will of course want to directly charge you for using it, but even if it’s “free” (and it won’t be), everything you say, every vocal or written prompt you submit, is personal data that you’ve surrendered forever and it has value. Your data will be harvested to build a virtual model of you as an individual based on your bonafides, interests, locations, proclivities and vices. This information is sold to data brokers who have already amassed by piecemeal a stunningly large amount of info about you from other sources like your phone, your smart TV, your doorbell camera, your connected appliances, your ISP, your cell data provider, flock cams, public records, social media, it goes on and on. The data brokers will then package and sell “you” and others to marketers, insurance companies, credit card issuers, domestic and foreign governments, basically anybody with an interest in your demographic and a check that will clear. And crucially, your data will also eventually find its way to…
…other LLMs. These LLMs can aggregate every piece of information in the known universe about you, finally putting Artificial Intelligence to use doing the only thing any of its investors care about, which is forming an extremely accurate forecast about your ability to pay off a loan, your likelihood of outliving your term life insurance, and your receptivity to a timeshare pitch.
Don’t forget Palantir and their desire to surveil you for the benefit of fascism everywhere!
I honestly think that’s where this leads. Running these AI data centers is too expensive for even the most advanced ad-generation and targeting efforts, but if you can get some sweet DoD (or foreign nations’ equivalent) contracts, you can cover your billions in operating costs.
And you can make everyone “safe” from things like dissident thought.
You know what’s even better than a button to open the trunk? A lever with a cable like my 1st gen accord had. And another one to open the gas door and a 3rd to unlock the hood. I also like levers with relay rods to open the doors, inside and out. When the battery dies, you are not locked out (or in).
I am amused by the parallels to industrial automation a few decades back.
As computers got more powerful and networking got faster, suddenly making colorful, fancy, animated UI graphics became the cool thing to do. You have a boiler? Animate flames on the graphics. You have a centrifugal pump? Animate the impeller spinning.
Unfortunately, all those neat graphics failed to capture the very essence of what the human-machine interface was for – communication of information. The neat flames and fun colors became distractions that slowed down how people ingested the information being shown to them, delaying their responses and corrective actions. The industry course-corrected and pushed for simplified, near-monochromatic graphics that conveyed the information necessary and nothing more.
I feel like a large part of automotive controls have already been whittled down to what is necessary over the century-plus that humans have been driving, and this new focus on software-defined vehicles and AI integration is just us creating new distractions by trying to include dubious capabilities, and in the somewhat near future the automotive industry will course-correct back to a more simplified setup. At least I really hope so.
I’m sorry, Dave. I’m afraid I can’t do that.
“I want to plan a trip from San Francisco to San Diego…print the summary…” Does the car have a printer in this scenario? Is it printing out MapQuest directions for you? We don’t need to bring ink and paper into this.
A friend of mine is in the Air Force and he told me the first job he had when he got to his new base was refilling the printer in a cargo jet because the plane couldn’t take off unless it had printer paper. Yes, a C-17 gave the pilots a “PC LOAD LETTER” error message.
What if you wind up talking to your skeptical friend?
“We don’t need to bring ink and paper into this.”
Reminds me of this:
“I don’t need a receipt for a doughnut. I’ll just give you the money, and you give me the doughnut. End of transaction! We don’t need to bring ink and paper into this! I can’t imagine a scenario where I’d have to prove that I bought a doughnut. Some skeptical friend…’Don’t even act like I didn’t buy that doughnut! I’ve got the documentation right here! Oh, wait, it’s back home, in the file. Under d…for doughnut.”
-Mitch Hedberg
I want to talk to my car, in the abstract.
Like my dog, if it starts actually answering me in plain english, it’s really gonna fuck up the whole dynamic.
I regularly speak with my car and the cars I encounter on my occasional drives. I cannot repeat the words and phrases I use on a public forum.
Im guessing it was “**** YOU, YOU ****** *********, GO **** YOURSELF, etc,etc.”
And many adjacent to those.
Some comedian I can’t think of had a great bit about how it’s good your dog can’t talk because if it could all it would ever say is “HI I’M CHARLIE HI I’M CHARLIE HI I’M CHARLIE HI I’M CHARLIE HI I’M CHARLIE HI I’M CHARLIE HI I’M CHARLIE”
I wonder if the people who think this streamlines their life grew up having people around them do all the stuff they don’t want to do.
If they arrive somewhere and there’s not a person waiting to lug all their shit around they get a hollow feeling in their gut. Now they still get to tell someone to get their bag.
Hopefully soon they’ll have a robot in the trunk Motocompo-style to actually do their menial tasks.
Just to be clear, driver, I left your bag at the last stop for .0000000000074 btc. I got bills to pay.
The Chinese are communists. They’re used to everything around them listening to them
Also, were I you, I’d be careful crossing a street if there are any Rivians in sight.
If we get to the point that AI can do all the things that these tech bros want AI to do, most of the population isn’t going to be able to afford to buy a Rivian and they also won’t have a job to commute to.
Couldn’t agree with you more, Jason
If only there were a series of simple buttons, dials, or levers one could push/turn/slide to make things they way you want them…..
Tech for cars has gotten ridiculous and is only getting worse. Three years ago when I bought my Ford, the salesperson very carefully signed me up for the Ford app and showed me a bit about using it. A coupe weeks later the dealership called me to make sure he had shown it to me because Ford had notified them that I hadn’t logged on and apparently that’s an important metric. I said, “Yeah, you showed it to me, but I have no use for it.” I still haven’t used it.
Oh, and this Rivian guy is drinking some seriously good punch if he actually believes what he’s saying.
I think it’s because just about any car at this point is really good at being a car. Yet capitalism demands innovation with every model year. If they weren’t focused on this BS, really what else is there? (other than figuring out how to lower prices).
You’re probably right, but it seems to me that they could work on improving quality. My Ford has been subject to 8 or 9 recalls, though admittedly most of those were software issues. But quality isn’t sexy, I guess.
oh for sure, that’s what they should be working on. But yeah, I don’t think too many people appreciate cars that don’t get recalled, plus I doubt Marketing is gonna go for “our cars don’t break as much anymore!” if only…
There is room for improvement in the cars themselves.
The problem is that they don’t care. The future isn’t in hardware, it’s in connected services, customer interactions, and whatever terrifying idea the Silly Cons have this month.
right we are in the dystopian place where R&D is going toward screen-controlled air vents while at the same time companies forget how to do gas pedals right. (that’s the most egregious example I can think of, but far from the only).
Don’t forget door handles.
Any mechanical product has room for incremental improvements, some clever or specialized design tricks.
And you know, I want a hatchback door that’s designed to hold/lock itself open in a way where it doesn’t fall closed when parked on an incline. I don’t want it to be motorized or automated, just thought-out.
“Thought out” is I want from legacy industrial design companies with decades of institutional knowledge. Have every lesson and trick down as reference material and people checking all the details.
It’s in the realm of “how I imagined adults/corporations worked as a kid”, that there’s some kind of practiced professionalism guiding everything. Maybe that was how it worked before business schools and startup bros took over.
Ah right though, improvements to new cars. Accommodate mobile devices from a hardware perspective instead of making redundant ones and overly-integrating things with software, etc.
I’ve been fighting this USB microphone for ages now because I can’t get the proprietary configuration software to work with my tech. I don’t want a car to be like that. Build the physical hardware with physical features, then give me access to it with whatever software/systems I’m using…
Just as well off, as my Ford app can’t do anything I want it to. The only thing I want it to do is have my car interior at a comfortable temp when I leave for work. In the year and a half I’ve owned it, it’s managed to do that once.
remote start from my keyfob does that as well as anything.
Truly, no one wants this.
I hate talking to computers, sooo much… it – it – the f – it -fla – flames. Flames, on the side of my face, breathing, breathless, heaving breaths. Heaving breaths.
Despite all this, apparently this company was still way better at software than VW.
(There’s no such thing as good software.)
Telling the car that I have a bag in the trunk and the trunk opening is an incredible feat of speech recognition, context recognition, reduction of motives, etc for a machine to parse out and then respond exactly how you want.
It’s not a feature though.
What if I mention to someone in the car, while driving, that I have a bag in the frunk – does the frunk fly open?
As it turns out, it also costs money to do that LLM processing! So now, instead of millions of buttons being pressed millions of times each to open trunks for free, now you have those same trillions of cumulative button presses being multiplied by the cost of an LLM prompt.
I’m sure that won’t lead to cascading costs that are completely infeasible for absolutely no benefit compared to the free button pressing.
Don’t forget the astronomical electricity and water usage needed for all that.
Yup – every LLM interaction has a real cost in the real world – in water, electricity, required infrastructure, etc. And those costs don’t decrease with scale, so as they ramp up the number of users and the interactions with each user, those real world costs only climb.
Isn’t Rivian also the car company that spent some big sum of cash licensing Unreal Engine and/or Epic Games for its dashboard animations?
Plus, trying to give your CEO a $450 millionish pay package doesn’t exactly engender good feelings from the existing user base and potential customers, does it?
The R3/X is the most interesting upcoming car/EV that I can think of. But if it’s crammed full of things I neither want nor need, I’d buy a Mazda instead. You can still get some of those with actual instruments in the dash and plenty of buttons, as well as with modest screens that don’t dominate the interior and force the driver to interact with them for every single little thing. The CX30 and Mazda 3 couple/hatchback both offer that kind of experience, as well as better-than-class-average driving dynamics, interiors, and sheetmetal.
Plus, they weren’t brought into existence with the help of Saudi Arabian sovereign weath funds, should that sort of thing matter to you.
Your tech overlords want it Jason. So you’ll have it, or you’ve already bought your last car.
Well, unless you go with a Slate!
I’m absolutely certain Torch hasn’t bought his last car, but there’s a nonzero chance it’s already been built.
To be fair to him, “AI-deluded dillholes” is pretty much Rivian’s existing customer base.
That being said, they do need to expand it if they’re going to offer smaller and cheaper cars.
I’m stealing that.
Me: Open the frunk bay door, Hal.
Rivian: I’m sorry, Dave, I’m afraid I can’t do that.
Me: And stop calling me Dave!
I was at a tech conference yesterday and had an interesting experience. The conference was full of AI talks (because of course it was) which I mostly skipped, but I did go to one about hiring people in an AI world where they can instantly look up an answer to most questions. Early in the presentation they did a straw poll of the audience about how many AI skeptics, wait-n-seers, and true believers there were.
The result was a handful of skeptics, a auditoriumful of wait-n-seers, and a couple of very uncertain true believers. They said when they gave the same talk 3 months ago at a different conference it was more like 20-60-20%. It seems like most of the people being forced to use this at their jobs are figuring out two things: 1) It ain’t all it’s cracked up to be but 2) It does have it’s uses. None of which are in your car.
None of this even gets to two basic issues:
not everyone can talk (the hearing impaired can still drive…)not everyone WANTS to talk in their vehicle!!!I don’t care about what the tech is behind it – but if you are driving, and have an infant, that has just fallen asleep in your car, the ABSOLUTE LAST thing you want to do is make ANY noise. Let alone having a voice, that isn’t the soothing voice of a parent, confirming you need to turn the AC on.
Meanwhile the stereo is playing your favorite song/podcast/news report – How is AI going to hear your voice? Of if the windows are down? Or a big noisy diesel smoketruck/Harley Davidson/”I WANT EVERYONE TO HEAR MY MUSIC” driver/Ambulance is next to you at the light?
What then?
Now I want to know what the car does when it hears “Whoop That Trick”
On Android Auto I press a button, the audio pauses and the fan for the HVAC lowers to a minimum speed. I actually do like voice dialing. I would also like voice texting while driving, except I generally have to say the text about 10 times before it actually understands what I’m saying.
So in order to make this work, you need to press a button.
A button which that guy wants to eliminate.
And your voice to text doesn’t even work properly.
See how little sense it all makes?
I mean, it’s a ‘button’ on the touchscreen.
And that guy thinks that’s too much.
According to him – it’s better that you talk to your car, which uploads your command via the airwaves to an AI datacenter how many miles away – which then decides what you want to do and to do it or not – and records that data, including where and when you made that command, data which you have no control over – then sends that command back to your car to be done.
See how stupid and invasive it all is?
So, I’m not saying I want to talk to my car for everything. However, for certain things, I do find it useful, especially since I’m a delivery driver and have to make a lot of calls from my car.
His statement that AA or CP takes over every pixel on the dash doesn’t have to be true, you could have one segment for AA/CP, other segments for Rivian’s nonsense.
Ok but how are the visually impaired drivers gonna adjust the vents through the touchscreen?
“…visually impaired drivers…”
I’m just gonna leave that right there for you to consider.
/s.
This moron is emblematic of how little success has to do with talent and intellect. I’ll stop there as that’s about the nicest thing I can think to write.
Didn’t HAL in 2001 teach us anything?
Never trust a Human?
Like not being in a computer-controlled vehicle when the computer was programmed with hidden goals that are against your best interests?
Don’t worry, unlike HAL’s “keep the mission a secret” thing, Rivian’s AI will only be designed to extract your personal information, encourage you to engage with it as much as possible, and whatever else Rivian can think of asking it to do!