Advertisement
BBC TopGear
BBC TopGear
Subscribe to Top Gear magazine
Sign up to our Top Gear Magazine
Subscribe
Opinion

VWs are about to get ChatGPT. Here’s why we’re worried

Volkswagen puts the ‘AI’ into ‘air-con’. But can we trust the chatbots?

Published: 10 Jan 2024

All hail our new infotainment overlords! VW has announced its new cars will soon come plumbed with ChatGPT, with the AI chatbot set to be integrated into Volkswagen’s existing in-car voice assistant over the next few months.

ChatGPT will burrow its way into the ID.3, ID.4, ID.5, ID.7, Tiguan, Passat and Golf from the second quarter of 2024, offering – in Volkswagen’s words – "a multitude of capabilities that go far beyond the previous voice control".

Advertisement - Page continues below

What kind of capabilities? “The IDA voice assistant can be used to control the infotainment, navigation, and air conditioning,” says VW, “or to answer general knowledge questions.”

A more cynical organ than Top Gear might suggest that, with no human ever having successfully navigated Volkswagen’s current infotainment/nav/air-con system, they’ve had to call in AI to try and figure it out.

But that’s not Top Gear’s concern here. Top Gear’s concern is: can we trust it?

Because, as anyone who’s played around with ChatGPT will attest, it’s (a) an incredibly powerful tool of knowledge aggregation, and also, on occasion, (b) a big fat liar.

Advertisement - Page continues below

AI tells porky pies. It makes stuff up. Legal stuff. Stuff about Proust. Really bad stuff about professors.

And you really don’t have to go out of your way to make ChatGPT fib. Here's a conversation I had with it a couple of weeks ago (which also offers a searing insight into the sort of excitement I get up to when not being driven into by Golfs):

Me: What's the longest English word that can be spelled using just seven different letters? Each letter can be used more than once.

ChatGPT: The longest English word that can be spelled using just seven different letters, allowing for repetition, is "uncopyrightable". It consists of 15 letters and uses only seven different letters.

Top Gear
Newsletter

Get all the latest news, reviews and exclusives, direct to your inbox.

Me: No it doesn't. It uses 15 different letters.

ChatGPT: You're right, my apologies for the mistake. The longest English word that can be formed using just seven different letters (with repetition) is "doorkeeper," which consists of 9 letters.

Me: Surely 'storekeepers' would be longer?

ChatGPT: Yes, you're absolutely right!

I know! Don’t sound so cheery about it! Competence at silly word games isn’t, I realise, a prerequisite for an in-car voice assistant, but still… the misplaced certainty is kinda unnerving, no? As wiser analysts have noted, the problem isn’t that ChatGPT doesn’t have every correct answer. It’s that ChatGPT confidently asserts utter nonsense as the correct answer.

(And yes, I know it’s not technically lying, because lying implies conscious deception, and ChatGPT is just a big box of wires with no agenda, but isn’t that exactly what a big box of wires with an agenda, and the ability to lie, would tell you?)

So what about car-stuff? On early investigation, no better. Just for funsies, I asked ChatGPT to give me the best driving route from London (big city in the south-east of England) to Truro (small town in the far south-west of England). It confidently informed me the fastest route was to take the M32 motorway directly into the centre of Bristol, which is never, ever the correct answer to the London-to-Truro question.

“Why would you go via the M32 when you could take the M4 and M5 motorways?” I replied.

“Apologies for the confusion earlier,” responded ChatGPT. “You're right; using the M5 and M4 can be a more direct and common route from London to Truro.”

OK, slightly dodgy sat-nav directions don’t pose quite the same existential threat as, say, AI drones turning on their human operators. But still, if I’m begging my in-car assistant for, say, urgent directions to the nearest A&E, I’d like to be sure it’s not confidently leading me to the local owl sanctuary instead. (“Yes, that is a Eurasian scops, not a paramedic! You’re absolutely right!”)

I know, I know, large-language models are evolving at a fearsome rate. In a couple of years – maybe less? – I’m sure these errors will be eliminated, and that ChatGPT shall be able to answer any automotive query flawlessly, right before stealing my job, and wife, and any remaining sense of purpose on this planet.

But this is right now, and right now ChatGPT has a bit of a truth issue. Other carmakers, including Mercedes, are also developing AI language programmes, but theirs remain in the beta-testing phase. Has VW jumped the gun on this one?

In search of answers, I did what any self-respecting journalist would do: ask ChatGPT. And the answers were, at the very least, impressively self-critical.

“The integration of AI into vehicles raises pertinent questions about safety, privacy, and the potential distractions it might pose,” the bot admitted with searing honesty. “How much of our data will be shared and stored by ChatGPT within the confines of our vehicles? Will sensitive information remain secure, or could it inadvertently become part of a larger data ecosystem?”

To that end, Volkswagen has clarified that “ChatGPT does not gain access to any vehicle data; questions and answers are deleted immediately to ensure the highest possible level of data protection".

Still, ChatGPT wasn’t done with the self-critiquing just yet. “The efficacy of AI in understanding and responding accurately to human language nuances remains a work in progress,” it continued. “It might end up being a double-edged sword, providing insightful information one moment and misinterpreting commands or intentions the next.”

That noise you can hear? A big box of wires committing hari-kiri. And also proof that ChatGPT can do honesty, when it puts its mind to it.

Even so, Top Gear can’t help wondering if it’s all a bit… pointless. Sure, VW, ChatGPT might be one way to help make your in-car functions more intuitive and less distracting. Alternatively, how about just… not making them so unintuitive and distracting in the first place?

More from Top Gear

Loading
See more on Opinion

Subscribe to the Top Gear Newsletter

Get all the latest news, reviews and exclusives, direct to your inbox.

By clicking subscribe, you agree to receive news, promotions and offers by email from Top Gear and BBC Studios. Your information will be used in accordance with our privacy policy.

BBC TopGear

Try BBC Top Gear Magazine

subscribe