The May 7 ARDG meeting in Washington, D.C., featured a presentation from Scott Craver, a PhD candidate in the Electrical Engineering department at Princeton University. Scott discussed the state of academic research on embedding and attacking watermarks, discussed the process of research, and mentioned several specific statistical attacks. He also noted that there are general attacks on watermarking systems which do not require attacking the watermark itself, and mentioned several attacks which attempt to conceal the presence of a watermark from a detector without removing the watermark itself. He emphasized the need for peer review of watermarking technologies, and concluded that the current situation "favors analysis".
Alan Davidson of CDT gave a presentation about consumer requirements in which he suggested that consumers' rights must not be down-rezzed (here, a joking term for "rounded down"). This led to an interesting exchange about the "rounding problem".
Suppose a user receives a movie to which a movie studio has attached a policy such as "View for 48 Hours". If an analog marking system can only express four policies -- "Copy Never", "Copy Once", "Copy No More", and "Copy Freely" -- there is no policy which faithfully reflects the movie studio's original policy.
One could then ask which of those four states is the most faithful to the original policy. "Copy Never" is somewhat more restrictive than the original, and "Copy Freely" is somewhat less restrictive. Alan argued that it was most appropriate to round up in order to ensure that consumers never lost out as a result of an analog conversion. In effect, motion picture industry representatives who opposed him argued that the reverse was true -- that it was most appropriate to round down in order to ensure that movie studios never lost out. I was troubled to hear that the movie studios appeared to believe that a policy of rounding down was somehow logically necessary or logically preferable to a policy of rounding up. As far as I could tell, "Copy Never" (more restrictive) is just as inaccurate a paraphrase of "Copy for 48 Hours" as is "Copy Once" (less restrictive). It's simply that one choice is more protective of publishers and one is more protective of users.
Alan presented another example: if the original policy is "Copy 3 Times", two possible paraphrases are "Copy Freely" (less restrictive) and "Copy Once" (more restrictive). If the policy is paraphrased as "Copy Freely", the new policy permits things which would previously have been prohibited; if it's paraphrased as "Copy Once", the new policy prohibits things which would previously have been permitted. Again, I see no strictly logical basis upon which one paraphrase could be preferred to the other. Both are inaccurate.
It would be unfortunate indeed if MPAA presented the situation as though there were no choice to be made between alternatives like these when paraphrasing policies. Nobody at the ARDG has presented -- in the context of MPAA's view on the policies which ought to be expressible in an analog signal -- a way to avoid having to make such a decision.
In a subsequent conversation with Brad Hunt, I pointed out that, outside the ARDG, MPAA tends to point to the existence of the analog hole when DRM technologies are criticized for thwarting fair uses. For instance, in the 2600 litigation and in the Library of Congress DMCA rulemakings, movie studios suggested that people who needed to make fair uses which were prevented by CSS could use analog to make those uses. If MPAA hopes to do something which closes the analog hole, will that particular safety valve evaporate?
A presentation on a technology called CGMS/A+ (supposed to be a next generation version of CGMS/A) was met with a wide variety of criticisms. CGMS/A+ would add a visible "rights assertion mark" -- a logo in the corner of a viewer's screen -- to programming which had previously been marked with CGMS/A (a non-watermark label in a well-known portion of the signal which is not normally viewable, nearby any closed-captioning information which might be present). The claim was made that this visible mark would be difficult to remove, and devices might make certain assumptions about CGMS/A data based on the presence of a rights assertion mark.
The reasoning behind the proposal to add the rights assertion mark is that it is fairly easy to remove (and somewhat easy to alter) CGMS/A data. The mark might function as a signal that CGMS/A information had originally been present and had subsequently been removed by somebody. However, it did not appear to many participants that CGMS/A+ would be broadly useful. One reason is simply that many important devices like computers typically don't respond to CGMS/A in any way, and aren't legally required to. So trying to preserve CGMS/A marking can't do anything to affect how these devices behave.
Jim Burger presented an interesting set of requirements for discussion, prompting others vigorously to discuss them.
We hope Jim's document will be made available to the public soon.