Reversible Measurement: Is It Possible?

[A Puzzlement]

Reversible computation is a somewhat well-known topic. (Quantum computers, for instance, must use reversible gates).

Apparently, though, quantum measurements can be reversible too. This also means you could recover the original state by “unmeasuring” the system. Imagine being able to “see” a photon without absorbing it!

So in theory should it be possible to have a robot or lifeform that interacts with its environment in a completely reversible way, dissipating no heat? Or at least one that comes arbitrarily close. (Of course, the environment would have to be fine-tuned, which I presume is why we don’t see reversible creatures everywhere).

Then again, a reversible lifeform might look so different from anything we’re familiar with that we’d completely miss it if it was there.

 

[PeterDonis]

This also means you could recover the original state by “unmeasuring” the system. Imagine being able to “see” a photon without absorbing it!

That's not what would happen. "Unmeasuring" would reverse the state of the measuring device and the environment as well as the measured system; so the result would be that you "unsee" and "unabsorb" the photon and have no memory of observing it.

 

[Khashishi]

A reversible system is basically deterministic, so any "life form" would be acting like a machine, doing what it does with no surprises. Most definitions of life include some kind of metabolism, which is generally irreversible.

 

[Strilanc]

The reversibility they're talking about in the linked document has a non-negligible chance of failing.

What they're doing is a bit like taking a Bayesian prior and coming up with a test that has at least one outcome that will cause the prior to become a desired posterior. But if you run the test and get a different outcome (which must be possible unless the prior is already the desired posterior), you will get the wrong posterior instead. You can repeat the process, trying to craft a second test that has an outcome that will cause the first posterior to update into the target posterior, but this second test will have an even lower chance of success. Even with an infinite branching sequence of optimal tests prepared, the chance of not reaching the target posterior will stay bounded away from zero. The alternative would be like... if in reality you could make a drug work by doing weirder and weirder studies, instead of by designing the drug correctly.

The more useful your initial measurement is, the more likely it is that the reversing operation will fail. The more likely it is that the reversing operation succeeds, the less information you can get out of the measurement. These two effects will perfectly balance each other in a way that ultimately adds up to "useful measurement is irreversible".