There’s a lovely small detail in BMW’s Neue Klasse safety engineering that interested me rather more than the breathlessly inevitable claims in the press pack about computing power.
If the car detects a cyclist approaching while you start to open the door, it can briefly stop the electronic lock releasing. You pull the handle. Nothing happens for a second or so. Cyclist passes. Door opens.
It is in itself a tiny intervention, but it tells a story about the way we have designed automotive interfaces for years. The usual approach would be to detect the cyclist, flash something, perhaps in the mirror, make a noise and hope the person opening the door notices the warning, understands it and reacts before depositing said cyclist across the road, GoPro and all.
We have become very good at warnings. Cars bong, flash, vibrate and illuminate little amber hieroglyphics all over the place. Much of the time this is eminently sensible. But every warning also represents a decision about responsibility: the system knows something, tells the dumb distracted human, then leaves the same fallible human to sort it out.
In this case BMW appears to have asked a more pertinent question. Given that the car already knows there is an immediate hazard, and given that delaying the door by perhaps a second has almost no meaningful cost, why involve the human in that particular bit of the decision at all?
We spent quite a lot of time thinking about this during the Human–AI Systems course I took this year with Cambridge University. Discussions about automation and Ai tend to drift towards capability: what can the machine do, how clever is it, how much should we automate? The problem is usually much more prosaic. Who is actually best placed to do each part of the job?
Here, the person still decides to open the door. The car does not second-guess where they are going or decide they should remain imprisoned in a BMW for their own good. It simply refuses to execute that intention for the brief period in which doing so would be obviously dangerous. That feels like good judgement.
And it connects to something I like finding in much broader experience work. When a problem becomes visible at an interface, the natural response is to improve the interface. Make the warning clearer. Increase the contrast. Rewrite the copy. Add another confirmation step. It’s catnip on social media where junior designers are endlessly posting ‘better’ forms and processes.
Sometimes all of that work is downstream of the more useful question: why have we made this the user’s problem?
A better warning cannot compensate for badly allocated responsibility.
There will obviously be limits to this. Systems that intervene too readily become infuriating, unpredictable and eventually distrusted, we can all think of examples where this is the case. But this particular example is wonderfully clean: immediate hazard, high confidence, trivial delay, potentially nasty consequence.
So yes, BMW’s new computing architecture is probably very clever but not as pleasing as the fact that someone used a little of it to make one warning unnecessary.
AI: Ai was used to correct a smattering of the most egregious grammatical errors.
