Re: ACTION-253 ISSUE: 119 and ACTION 208 ISSUE-148 Response signal for "not tracking" and definition for DNT:0

On Sep 12, 2012, at 6:58 , Ed Felten <ed@felten.com> wrote:

> What I'm trying to get at is what statement the user is thought to be making by sending DNT:0 rather than sending nothing.  

As I see it, DNT:0 means

a) I am aware of DNT
b) I may be making DNT:1 requests to some others
c) but I am explicitly not making that request of you (probably as a result of an exception request by you, which I granted).

Whereas unset means I might not be aware of DNT at all.

DNT:0 has a very valid meaning in the exception case, and it might begin to have a meaning if a jurisdiction were to discuss a difference between a 'default' behavior and an explicit signal that the user doesn't mind being tracked (which DNT:0 is).  If, in addition, the 'default' state leads to users constantly being asked questions (e.g. for an exception, consent or approval), then at some point it make become useful for browsers to allow it to be set as their general preference;  but to my read, we're not there yet.



So, it's a total awful mouthful, but I think I agree with David W that the definition of DNT:0 is an explicit "not DNT:1".  "You are not being asked to apply the restrictions that a DNT:1 signal would indicate."




> Maybe the difference is that DNT:0 and unset say the same thing about the user's (non-)request to the current server, but DNT:0 implies that the user might treat *other* servers differently.  But I'm not sure what the rationale is for having users tell one server in a non-specific way what their general policy is toward other servers.
> 
> To put it another way: If I am a User Agent, why would I bother to implement DNT:0 if it is so close in meaning to unset?   Why wouldn't I just use a two-state approach with DNT:1 and unset?  
> 
> On Wed, Sep 12, 2012 at 9:35 AM, David Wainberg <david@networkadvertising.org> wrote:
> Hi Ed,
> 
> My understanding is that there was in the group a desire to distinguish unset from an explicit exception. Others can probably better talk to the rationale and use cases, but I can imagine that it might be desirable for a party to know that it has received an explicit exception from the user rather than passive disregard -- "hey, she really likes us!" 
> 
> One other rationale may have been to address EU consent needs. I'm a little confused on where we stand with that, but I think the most recent statements from EU participants is that we're not going to be able to address that with this. If we are indeed trying to do that, if we're trying to shoehorn into the DNT standard a broader consent beyond DNT on or off, then my points below apply.
> 
> -David
> 
> 
> On 9/11/12 7:08 PM, Ed Felten wrote:
>> Hi, David.  Thanks for taking the time to explain your proposal in detail.
>> 
>> I'm still a bit unclear on your proposed meaning for DNT:0.   In particular, I don't quite understand how DNT:0 would be different from leaving DNT unset.  You say this: "Although in most cases, [DNT:0] is functionally the same as DNT unset, it indicates an affirmative exception granted to the recipient, rather than, for example a UA that does not implement DNT, or a user that does not use DNT:1 at all."   What I don't understand is what the server is allowed/expected to infer from the fact that the UA does or doesn't implement DNT, or that the user sends DNT:1 to other sites but not this one.  
>> 
>> To put it another way, the text I quoted above talks about *why* a user might end up sending DNT:0 instead of unset,  but it doesn't quite say what the presence of DNT:0 is supposed to communicate to the server.   If I'm understanding you correctly, you don't want DNT:0 to be interpreted as granting the user's consent to track.   (If DNT:0 was an explicit grant of consent, that would make it different from DNT unset.)  What could the server do or not do on receiving DNT:0, that would be different from if it received DNT unset?
>> 
>> On Tue, Sep 11, 2012 at 6:16 PM, David Wainberg <david@networkadvertising.org> wrote:
>> Hi All,
>> 
>> I am combining these issues because the problems with both are similar. I am proposing a) that we drop altogether the idea to have an "absolutely not tracking" response, and b) that the meaning of DNT:0 should be limited to "not DNT:1."
>> 
>> Both ideas, because they create distinct, parallel, and barely defined, policies within a policy, create ambiguity and confusion. The result will be reluctance to adopt the standard because of the ambiguous legal risk it creates. They are also both designed to accomplish aims that are beyond the DNT mechanism we are working to define.
>> 
>> First, we should avoid including these policies in the standard just as a matter of good drafting. When the point of the standard, and almost all of the content in it, pertains to one thing -- the meaning of a user's preference not to be tracked (DNT:1) -- tacking on a new definition pertaining to something else -- a user's affirmative consent for something that may or may not be tracking -- just can't be good. Same goes for creating a new state of "not tracking" or "anonymous" that has a meaning apart from the meaning of DNT.
>> 
>> On last week's call we already agreed that defining "not tracking" without defining "tracking" would be a problem. (It also would be a problem if we defined both but with some delta between them; that would be exceedingly confusing to implementers and users.)
>> 
>> We would also create confusion by defining DNT:0 as something more than not DNT:1, because it creates a mismatch between what is regulated under DNT:1 policy and what is then allowed under DNT:0 policy. This is going to cause serious heartburn.
>> 
>> One solution that has been proposed for the status flag is that we go with a term that does not include "tracking," such as "anonymous." But this raises similar issues. The standard addresses tracking. Under the standard, users express a preference not to be tracked, and servers honor that preference per the standard. There are two states: tracking and not tracking. There's no other state required, and to create a third state will degrade the clarity of the policy, and will create confusion for implementers and users.
>> 
>> Moreover, how are we going to define "anonymous" or "pseudonymous" or "foo"? Given that this is an unnecessary appendage anyway, and that we can't even define "tracking" in a "do not track" standard, why do we want to create the problem of now having to define some other state. To include it without a definition would be unacceptable.
>> 
>> Aside from the problems of ambiguity and confusion, both proposals go beyond what the TPWG was chartered to accomplish. The not tracking signal would create a mechanism to aid a small number of sites who, as part of their marketing efforts, are making assertions with regard to their privacy practices. Although it may be laudable that they are doing this, it's not for this standard to get involved in promoting such efforts; they can do so in their marketing materials and in their privacy policies.
>> 
>> Similarly, the DNT:0 definition that has been proposed goes too far. It creates a wholly new policy, beyond DNT:1, and purports to grant consent for practices (e.g. personalization of content) that aren't expressly addressed under the DNT:1 policy. Does this imply that those practices actually are or should be addressed in the standard? Why aren't they called out explicitly?
>> 
>> The TPWG is developing a mechanism for users to express a preference, not a general purpose tool for communicating privacy practices, and not a tool for promoting certain practices over others. The DNT header simply communicates a user's preference not to be tracked. Although we are going slightly beyond that in providing for some indications of whether or how the recipient is honoring the DNT signal, those indications are directly tied to specific aspects of compliance with the DNT:1 signal. These proposals go beyond that.
>> 
>> So, for that reason, and because of the ambiguity and confusion I've explained, we should drop the "absolutely-not-doing-foo" proposal, and we should use the following as a definition for DNT:0.
>> 
>> "DNT:0 means not DNT:1, an express choice that DNT:1 should not apply to the recipient server. Although in most cases, this is functionally the same as DNT unset, it indicates an affirmative exception granted to the recipient, rather than, for example a UA that does not implement DNT, or a user that does not use DNT:1 at all."
>> 
>> If you've gotten this far, thanks for taking the time to plow through my long-winded post. I'm sure everyone will let me know if they have questions.
>> 
>> -David
>> 
>> 
> 
> 

David Singer
Multimedia and Software Standards, Apple Inc.