At most there is a choice between ‘less discordance vs more’.

But where does that get us?

We might at a push associate red with poisonous berries, and green with edible vegetables, and perhaps figure out that red wasn’t safe whereas green was.

But you still have to decide what behaviours/responses they’re supposed to produce in the observer.

If I see a green light in front of me, why should I not interpret it to mean “Remain here at this sign of safe harbour, for venturing further into the unknown is fraught with risk”. And if I see a red light, why do I not interpret this as a clear warning sign that danger lies nearby and will befall me if I do not quickly proceed?

So, similarly, if I see a switch with choices between degrees of order, and the bulb is broken, how would I guess which configuration should result in the light being turned on?

Artificial light is unnatural so I might infer that disorder brings illumination.

Alternatively, I might believe that illumination was positive and that order was positive, and assume that one implied an affinity with the other. Conversely, I might infer that change was positive and required flow from positive to negative, thus in order to produce light I needed to create disorder as a reason for it to come into being.

If I guess wrongly I may get electrocuted when I change the bulb.

The switch design is also plainly broken in its state of least disorder. So well is it camouflaged that I have little idea that it exists in three horizontal sections, nor that the central section may be shifted either to the left or the right.

Perhaps, in the darkness, if I had a bright light to hand I might figure out how to operate it?

Ideally, a switch should be a miniature model of that which it controls. It should also be usable in the conditions in which it is expected to be useful, i.e. visible in the darkness, recognisable in appearance, and intuitively operable. Even if at the sacrifice of aesthetic.

However, yes, as controls over those who should be discouraged from operating a light, they are effective. Light-off - switch camouflaged (usable only by cognoscenti). Light-on - discordant, visible and jarring, but still not necessarily completely obvious as a light switch, although an apparent mechanical solution to a reduction in discordancy does appear to be worth a try.

Comparably, one could imagine hidden light switches behind opaque white flush panels. Completely invisible except when the switch is on, when bright leds illuminate an outline of a conventional switch. Thus switches may be turned off by all (by a single press), but turned on only by those who know where they are (and tap twice in the right place). We can ignore the finger print problem - this only being an analogy.

So, it is not ‘obvious when they are switched on’. It is more a case of a solution to turning them off being somewhat intuitive only after they have been identified as being light switches and only after their current configuration has been ascertained as being in the ‘on’ state (probably because the lights that one would expect them to control are currently illuminated and no other switch has been located).