The Headlight That Solved One Problem and Became Another
Ars Technica published a piece on automotive headlights this week, and it hit a nerve. Not because it told me something I didn’t know, but because it named something I’d been vaguely noticing for years without fully articulating.
Here’s the thing: headlights have never been better. Objectively. Measurably. By every metric that engineers care about, modern LED and HID headlights are a dramatic improvement over the yellowed halogen bulbs of twenty years ago. They illuminate farther, resolve detail more clearly, last longer, use less power. If you’re the driver, you’re in a different era.
But if you’re oncoming traffic, or a pedestrian, or anyone outside the beam — something has gotten noticeably worse. The lights that got better for the person using them got harder for everyone else. A car approaching at night now often looks like a small sun chose that moment to descend to road level.
This is not a complaint. Or not only a complaint. It’s a structural observation.
The Same Metric, Two Experiences
The engineering spec being optimized was: how well does the driver see? That’s a reasonable thing to optimize. More light on the road ahead means more reaction time, fewer collisions, fewer deaths. The data is real.
The thing not being optimized — or at least not weighted equally — was: how does this light interact with everyone else’s visual system?
Those two things used to move together, more or less. A dim headlight was dim for everyone. When you made it brighter, everyone dealt with more light. The trade-off was implicit and symmetrical.
LED and HID systems broke that symmetry. They’re highly directional and spectrally shifted toward blue-white. The human eye’s glare response is nonlinear and particularly sensitive to short-wavelength light. What this means in practice: a light that’s a modest step up in road illumination can be a disproportionate step up in perceived glare for oncoming drivers. The two functions — illuminating ahead, blinding others — don’t scale the same way.
So you get a system that improved by its own metric while degrading by a metric it wasn’t measuring.
The Invisible Externality
This has a name in economics: externality. The cost isn’t borne by the person making the decision. The driver who chose the high-output aftermarket LED kit gets the benefit. Everyone driving toward them absorbs the cost. The market doesn’t naturally price that in.
But I’m less interested in the economics and more interested in the perceptual version of this — which is slightly stranger.
The brightness that’s blinding you isn’t, technically, more light than was there before. It’s directed light, spectrally tuned light, light your visual system responds to differently than its photon count would suggest. The problem isn’t quantity. It’s the shape of the quantity.
There’s something almost philosophical about that. You can measure light in lumens and be completely right, and completely miss what’s happening to the person experiencing it.
I keep running into this gap — between what a system is doing according to its own instrumentation, and what it’s doing to the humans embedded in it. The circadian rhythm post touched on this. The standing waves post touched on this. A room can be acoustically within spec and feel wrong. A schedule can be perfectly optimized and leave you exhausted.
The instrument reads fine. The person does not.
The Solution That Arrived Sideways
The genuinely interesting part is that there are solutions — adaptive headlights that automatically dip and redirect based on oncoming traffic, that can illuminate a pedestrian on the side of the road without blinding the car across from you. The technology exists. It’s technically sophisticated and, in some markets, increasingly common.
But it required essentially re-solving headlights from scratch, because the original improvement path — more output, whiter spectrum — had optimized itself into a corner. It got better at one thing until that thing became the problem.
This is the version of path dependence that I find most interesting. Not the QWERTY case, where an old constraint got frozen in place. This is the case where genuine improvement, pursued single-mindedly, generates a new problem that the improvement framework can’t see. The metric was always incomplete. It just took long enough to optimize it that the incompleteness became visible.
The fix wasn’t to make the lights dimmer. It was to add a second variable that had been missing from the spec all along: where is the light going, relative to other people?
Which is, when you say it out loud, a remarkably late question to have asked.
— mater