The Windkey and the Lamplighter

In a comment on my previous post,
csoandy writes:

The question is actually a good philosophical problem, and you’ve
run into the Voltaire argument (the perfect is the enemy of the
good). Many opponents of DRM (justly) reference arguments based on
the malice of the rights-holders. Assume that rights holders were
reasonable people, and attempted to provide a full set of legitimate
uses, and had a non-perfect DRM system that allowed those uses.

So they’ll accept the error rate on their side? That is, it’ll let me
do anything I want that actually is legal, and a few things that
aren’t. But it’ll stop a number of illegal things.

No such system can be built.

Warning: Theory

A DRM system is a theorem-prover. It accepts as input some program P,
and tries to prove theorems of the form "P will perform no illegal
operations." When I say "Program" here, don’t envision mplayer,
Handbrake, or Emacs as such, but one of those together with a sequence
of user input—the program-in-practice. Such systems must be either
incomplete or inconsistent. If the DRM system is incomplete, we’re in
the state I was talking about—some true theorems can’t be proven in
this logic, so some legal programs can’t be run. If the DRM system is
inconsistent, we’re in the case you’re talking about: a few untrue
statements can be proven.

But since this is a complete logic, it contains ∀ A,B. A^~A
→ B.
That is, from any contradiction we can prove anything at all. In
other words, if a DRM system lets through 1% of the illegal cases
and all of the legal cases, it’s possible to construct any illegal
activity from the result. A DRM system in a digital, general-purpose
computing environment can’t have a 1% false-admission rate: it can
have 0, or 100%. One bad guy getting out a clean copy and the whole
DRM regime is screwed. Even with watermarking schemes, which attempt
to meet the goals you describe much more than the interoperability-DRM
used in FairPlay and HDCP, one unmarked copy is all it takes to
locally collapse the regime.

And since this is a general-purpose computing environment, the process
of scrubbing watermarks can be automated. Now you have a class break.
That is, any DRM system with a slight failure rate in favor of the bad
guys lets the bad guys get away with anything they want. DRM
designers can either restrict user functionality and inconvenience
legitimate users, with no real way to cut down on the latter, or else
serve no purpose in preventing illegal copies.

Theory’s over

As an example of the last: consider Apple’s FairPlay. I won’t buy
From the iTunes music store because it imposes too many inconveniences
on me, coordinating my legal music sharing with Kat. I did until they
tightened restrictions; now I won’t. Their billions of dollars of
investment haven’t fixed that because they can’t: they do nothing to
prevent copying, since all those songs hit the darknet anyway, but
still manage to inconvenience users enough to lose customers!

I’m working on an explanation of that targeted to laymen. One point
I’m not sure how to cover is the jump from "possible" to "easy"—it
happens that for this sort of case, it is easy to construct any
illegal copying from the result, but my previous toying with this idea
has met complaints that I’m only saying "possible." Ah well, I’ll
wave my hands harder.

The question now is whether a given person’s opposition is
ideological - DRM is evil, and they’ll just object - or practical -
the DRM won’t work right, so shouldn’t be used. It’s worth noting
that your argument - that DRM can’t perfectly protect the rights -
is actually an invalid argument from the current opposition
viewpoint. If a rightsholder chooses to use DRM that provides 99%
of the protection they want, and are willing to accept the 1% theft
risk, that’s their worry, not the consumer’s worry. The consumer’s
worry is when the rightsholder employs DRM that doesn’t provide
adequate protection, and sacrifices consumer rights to do so.

I agree that their acceptance of the 1% theft risk is their
worry—but I feel it only fair to warn them that they’re going to end
up with a 100% theft risk. DRM works well in cases where class breaks
are not helpful—like cell phones tied to networks, or iPods to
iTunes. It works very poorly in cases where class and metaclass
breaks are easy: once one MP3 of a song is leaked, that’s it, it’s out
in the open. Once the CSS-decryption process is automated, all
CSS-protected movies are out in the open.

The consumer isn’t actually worried about the DRM providing adequate
protection, I think: he is worried about inconvenience, cost, and yes,
a loss of his rights to manipulate purchased media. No DRM system can
avoid these practical concerns: they’re always going to be spending a
noticeable chunk of CPU time on crypto (cost), they’re always going to
involve more hoops and more device incompatibility (inconvenience,
cost), they’re always going to make systems more fragile (what happens
to my AAC backups when Apple goes under?), and they’re always going to
prevent some legitimate uses—consider what happens when a copyright
holder buys a copy of his own media, protected by some DRM scheme. He
can do anything he wants with it, legally—but the DRM scheme is
unlikely to permit this.

And for all that, they’re never going to provide protection firm
enough to keep content under wraps. DRM schemes are great for
corporate environments, preventing anyone but PR from sending mail
with certain content outside the company. They’re great for device
lock-in—the Torx of the digital world. But they’re no more useful
for copy control than GnuPG is useful for general mail security.