1) Electronic items that are given watermarks or social DRM signatures can be electronically stolen/duplicated. This compromises the value of the watermarks and signatures because the extracted data becomes misleading. The data points to the wrong “culprit”. Consider the computer security situation today. A substantial fraction of PCs are compromised with Trojans, viruses and other malware.

It would be possible for “pirates” to steal e-book, music, and video collections. These purloined items could then be redistributed using P2P applications and other sharing systems. If a copyright holder finds one of his or her works being distributed without permission then he or she can consult the watermark or signature. Unfortunately it might only reveal the identity of the person from whom the object was stolen/duplicated. This does not help the copyright holder very much.

2) Watermarks and social DRM signatures will be opposed by some privacy advocates. If the mark is embedded in the electronic object and if it identifies the owner or provides credit card data then there is a danger if the object is stolen/duplicated. The credit card data might leak. Data about personal reading habits might leak. If the signature or watermark is resistant to forgery then the object could be uploaded to the net to provide evidence about the scandalous or outré reading preferences of an individual.

If the watermark or signature is easy to forge then “pirate” sharing networks will probably fill up with duplicated and altered works with designations such as “For the Exclusive Use of Bozo T. Clown”, “Generated for the Private Library of John Q. Public”, “For John Galt – Whoever He Is”, ”Kilroy Was Here – Electronically” or a blank identifier.

3) If watermarks or signatures are used to trace purchasers or library patrons via a database of transactions or accounts then some privacy advocates will object. Many libraries today erase the data linking a patron to the check out status of an item as soon as the item is returned.

If libraries kept a persistent database of transactions to help trace patrons then some privacy advocates will object. Book sellers like Amazon no doubt keep persistent databases of transactions but these databases will eventually be attacked too on privacy grounds by some I think.

You would want to encode information to identify the user and not the book, so the information to encode would be either 38 bits for a Credit Card number (assuming that’s available), 160 bits for GUID (assuming a centralized user database), or ~110 bits for a combination retailer GUID (assuming this can be smaller say 80 bits) and user account identifier (also smaller say 32 bits).

Also, as you noted this would only work for content, where authors/publishers provided the variants so there’s going to be the standard chicken & the egg problem with – building systems for content that doesn’t exist, or creating content for systems that don’t exist… requiring a major player champion to help start creating the ecosystem.

Suppose that an author who desires a watermark creates a collection of textual variants. For example a text might say “red and fuzzy” or “fuzzy and red”. The sentence “Odysseus and his men become trapped in the cave of Cyclops, a one-eyed monster” might have the following variant “Odysseus and his men become trapped in the cave of a one-eyed monster called Cyclops.” The author should create the variants so that they preserve the meaning of the text and are stylistically acceptable.

When N binary variants are created it is possible to uniquely identify 2 to the power of N texts. Twenty textual variants could be used to generate and uniquely identify over one million texts (precisely 1,048,576). The watermark is “hidden” in the text because the existence of the variants is not obvious when examining a single text.

Of course all watermarks are vulnerable to attack. For this style of watermark the adversary would attempt to acquire multiple copies of the text and look for all the differences using a string comparison to try to identify the textual variants. (Technical comment: To try and block this attack a larger pool of variants can be used and an error-correcting-code can be placed on top of the raw data derived from the variants.)

I thought of this scheme many years ago. It is “obvious” to a “practitioner” computer scientist in my opinion and should be unpatentable. However, many obvious techniques are patented and so I would guess that it has been patented multiple times by now.

I can see where this would work extremely well for offering back catalog books from publishers. The low rights price of back catalog items shouldn’t drive the price for offering the service to unreasonable level, and as Baen and Tor have proved offering back catalog books (that are a start of a series) for free can drive sales for current items.

The problems I see with this are:
- This only works well for content that is homogenous in price range (ie offering fiction and textbooks under the same plan would be problematic).

- Finding a critical mass, ie getting enough users and publishers too make the system worthwhile to both.

I myself think social DRM would help in a library situation as a gentle reminder. At times publishers may feel that content will slip into the “free” category, not because users are dishonest but because they may confuse “permanent checkout” with the public domain variety. I know of at least one nonlibrary case where this happened. So, yes, some publishers disliking traditional DRM might enjoy the social DRM option.

Once again, thanks for pointing out the limits of steganographic techniques for text. I’m just curious if perhaps something could happen through a mix of images and a container format. I even wonder about seemingly random sequences of text inserted in inconspicuous places in a book—perhaps varying from library patron to library patron. Is there a chance that such an approach would work? If nothing else, it would address the problem of reconciling the marketing with the need for reflowability. OK, the problem with that approach?

Thanks again, Jim.

David

I also agree that there should be a mix of models as well. Most of what I am pointing out is that there are substantial problems with picking any one model as THE model. It won’t work for all content, but most of the models you propose will work for some content/publishers. For instance the publishers that are comfortable with Social DRM (with no encryption backing) are also going to be comfortable with no DRM. At best current Social DRM only provide a gentle reminder not to the share the book, so to give their content out like that requires that publishers to be in a space where they can trust their users. For various reasons not all publishers are there yet (or will be there for quite some time – advances in the music industry not withstanding).

BTW The lack of progress of steganographic techniques for watermarking text is not for lack of trying (on my and plenty of other’s part).

I’ll start working on the response to Permanent Checkout now….

On the matter of the EFF plan, keep in mind that (1) it’s just one of several possible models, (2) publishers could decide whether to participate, (3 there could be some kind of plain-text mark to show that content was covered, and (4) Amazon, although admittedly it’s subsidizing the practice, is doing a pretty good job of price standardization at $10 for bestsellers. If enough people publishing industry don’t want the EFF proposal because of the complexities of pricing, etc., there are other ways of avoiding use of DRM for library use—if nothing else, then through timed browsing via the Web.

As for Social DRM, keep mind it’s already working fine for The Pragmatic Programmer people on technical books. Would it work with other kinds? Libraries and publishers need to find out. Like you, I heartily think we should not just trust “theory.” Publishers could experiment with social DRM in a library context even without digital watermarking used. Remember, too, that OverDrive is already letting people burn CDs for their private use. That’s happening even without Social DRM (not sure about what digital watermarking if any might be present).

Meanwhile, who knows—perhaps solutions to the technical problems (beyond use of PDF!) can be found.
Finally there is the possibility of expirable Web access; I could be wrong but you yourself may have brought that up.

One way or another, library users need alternatives to traditional DRM, which comes with its own problems, as end users keep complaining. It’s such a burden on library and private users that the “Hate DRM” factor might be balancing out any sales preserved through the anti-piracy proofing. Remember, DRM’s problems and potential problems don’t and won’t just result from user inconvenience but also from the eBabel challenges that inclusion of a related “protection” standard for ePub would require. The result could be an expensive delay in spreading around ePub.

Also keep in mind that, as noted, traditional DRM is bypassable, as shown by the existence of you-know-what software that makes s mockery of Microsoft Reader’s DRM. And yet we don’t see Microsoft Reader books pirated all over the place. Maybe users are more honest than the traditionalists would give them credit. Instead of theorizing, both the library and commercial sides need some Real World tests.

Keep speaking up, Jim. Although we disagree, I find your comments useful and constructive.

Thanks,
David

Assumptions – the user reason for Social DRM is to have the content non-encrypted (ie regular DRM), so that the user can move content from format to format and machine to machine as the user desires/needs. The Publisher reason for Social DRM is that the watermark will contain sufficient personal information to discourage the user from sharing (acct/credit card/social security etc..) or baring that will have sufficient information to track down the original infringer.

Unfortunately the current watermarking techniques are fragile and don’t provide publishers sufficient safeguards since they are too easily removed. First the obvious technique of embedding the watermark in the content is that it is easily handled with most editors. There are more advanced steganographic techniques, but these use presentation information of the more complex formats (PDF, Word, etc), but any format transition, either within the format (changing page size), or to a different format will garble/strip the watermark.

So first off – Collective Licensing.

The EFF proposal calls for this to be voluntary for both the consumer and the publisher. Unfortunately this would fail in practice unless all of the publishers were part of the “voluntary” collective. If there were any content that was not part of the collective licensing agreement, then it would place a burden on the users to determine if the content that they just received or decided to share was covered or not. However, given the sophistication of most end users regarding copyright issues, this just isn’t going to happen. While this may push most hold out publishers to become part of the voluntary collective in order to be compensated (how voluntary is groupthink anyway:) ), the remaining holdouts will end up being a RIAA redux with lawsuits and DMCA takedown notices creating the same nightmare that’s currently happening with music that collective licensing is supposed to fix.

Also the same preconditions for the Music industry do not hold for the publishing industry. Music compensation is largely homogenous, but unfortunately there is a larger degree of variance in pricing in the book world ( just think large segments – current Fiction, Textbooks, Industry texts/reports for starters).