<wseltzer> present=

<wseltzer> in meeting business use cases or ecosystem needs?

<wseltzer> https://w3c.github.io/web-advertising/dashboard/

<GarrettJohnson> 3 cheers for the general mute. Hip

<GarrettJohnson> Hip hurray!

Agenda-curation, introductions

<wseltzer> Max_Gendler: data governance team at New York Times

<wseltzer> Moe_Ismail: Observing

PrivacyCG report from vF2F

/me at 98% trying to get into webex

<wseltzer> Erik: Second vF2F last week

<wseltzer> ... 4 hours over 2 days

<wseltzer> ... we discussed a variety of work items: incubations that have some level of browser vendor support

<wseltzer> ... get active discussions going

<wseltzer> ... recognize some cross-participation between groups

<wseltzer> ... Where you want to have a more technical conversation, come to PrivacyCG

<wseltzer> ... Also discussed proposals: new ideas with structured thoughts

<wseltzer> ... some of those might turn into work items

<wseltzer> ... If that sounds interesting, check out privacycg.github.io

<wseltzer> https://privacycg.github.io/

<ErikAnderson> https://github.com/privacycg/meetings/tree/master/2020/09-virtual

<wseltzer> wseltzer: also some active breakout sessions afterwards

<wseltzer> btsavage: in our use cases document, 2 separate use cases

<wseltzer> ... aggregate measurement: how many total conoversions

<wseltzer> ... optimization: training ML algorithms to serve relevant ads

<wseltzer> ... As we've previously described, we ran a test ...

<wseltzer> ... We talk lots about measurement, less about optimization

<wseltzer> ... Help to clarify that these are separate use cases

<wseltzer> ... John Wilander clarified that Safari team is not trying to solve optimization, focused only on aggregate measurement

<kleber> Breakout session notes: first day https://github.com/privacycg/meetings/blob/master/2020/09-virtual/09-16-minutes.md#breakout-session-on-pcmads

<scribe> scribe: Karen

<jrosewell> btsavage: Reminder of work by Garrett Johnson : https://docs.google.com/presentation/d/1juu6UBguR7ru1Rhfyor9knFBFsyLNP6FmIiXcJKyJZM/edit#slide=id.g604c6519b1_0_1

<wseltzer> ... on day 2, that was described as a matter of timing, not now.

bleparmentier: I just wanted to say
... I 100 percent agree with Ben on that
... we have always, together with different issues

<kleber> Breakout session notes: second day https://github.com/privacycg/meetings/blob/master/2020/09-virtual/09-17-minutes.md#breakout-session---pcmconversionad-topics-discussion-continues

bleparmentier: when we talk about measurement we were thinking about optimization too
... hard to do @ without measurement
... we have issues with Differential Privacy
... blog post
... exactly on this subject

<AramZS> Link blog post please?

bleparmentier: why advertising world is not well suited
... explain further som eo fo our concerns
... even more important where actors can act when info is passed
... if actor is not able to differentiate from an ad
... huge incentive to put a lot of ads to optimize the page
... optimizaiton is not just manipulation
... some hype
... sorry I could not join those calls as it was late
... wanted to pile onto what was said
... important use cases
... why we are proposing our measurement proposal like that

Wendy: thank you
... Aram asks if you can share a link to the blog post

AramZS: once again
... as far as i know, maybe something different from what Ben is referring to
... don't know if 50% number is
... environment being tested
... situation that Brave talked about
... operating on this level was more successful for everyone
... note that I don't necessarily think that stat is the most useful one
... the optimization case is one to consider
... no Apple or Mozilla people to speak

<bleparmentier> The blog post link https://github.com/Pl-Mrcy/privacysandbox-reporting-analyses/blob/master/differential-privacy-for-online-advertising.md

AramZS: but making ability to optimize less effective
... legit item to discuss
... fine line between optimization and manipulation of populations; we need opt. and to successfully deploy
... can talk about algorithmic redlining

<joshua_koran> note the Brave study did not control for seasonality, ad format, or advertiser mix on the puplisher - thus comparing results in t1 to t0

AramZS: find a middle ground when we put together a use case document
... get as close as possible working with engineers
... but need to compromise
... way things work is not sufficiently private for browser engineers and users
... no hard line that it's most reasonable
... looking for a compromise
... that's it for me

Arnaud: I think the point from Aram is very interesting
... example he gives that happened in GDR enabled companies
... by publishers trying to make money
... is really about targeting and not about measurement problematics
... that is Ben's point
... you might lose some business if you lose targeting
... but if you lose measurement capability, it's a whole other story
... also about ability to do frequency capping and real time reporting
... and all experiments made by publishers and other actors
... are completely set aside; only focus on user targeting level
... it's about measurement
... what is proposed by Google; really about measurement and what we should talk about
... not the user level targeting capability

<AramZS> This is the situation I was talking about: https://techcrunch.com/2020/07/24/data-from-dutch-public-broadcaster-shows-the-value-of-ditching-creepy-ads/

<AramZS> for ref

Kris: I was going to say
... I agree with Aram
... we do need to approach this at a compromise level
... consumers and users don't feel like data being collected on them and targeted to them can be creepy because they don't understand what is going on
... Also fully agree we need to look at optimization as well as measurement
... but argue that part of that should be argued from user POV rather than Publisher POV
... users are more engaged in converting when ads are more effective for them
... getting ads that users want to see or are interested in is better than giving them ads they are not at all interested in
... structure this in more user centric POV
... rather than going after it as advertising industry v. browser industry
... that is a losing argument
... we should focus on what is most effective for users
... that is my two cents

<wseltzer> https://github.com/w3c/web-advertising/blob/master/support_for_advertising_use_cases.md

Wendy: thank Ben for calling our attention back to the use cases

<arnaud_blanchard> (not a reproach to the scribe, but the minutes do not reflect at all what I just said)

Wendy: I had proposed in next agendum as well
... and for distinguishing among different use cases and the different technical requirements they might have
... and to Kris' point, there is a user needs section in the document
... that might benefit from more looking at by user advocates
... to see if we have captured needs of different users
... and how we should refer to privacy and content access considerations
... to see things that advertising supports
... don't want to leap too far ahead to next agendum
... several more people queuing up
... we should go back to the use case document
... now that we have had lots of technical conversations about proposals
... and see if high level needs need more detail

<joshua_koran> @Arnaud I think your point was that while targeting can help focus marketer budgets, the most important value for marketers of data collection/processing is measuring what is effective, so that they can do a better match.

Aram: I want to note
... we have been talking about user tracking
... including Google 50% study
... that is not nec optimization
... agree these two things are mechanically different

<joshua_koran> @Kris this measurement seems to also help your point of ensuring we are doing a better match of content to people, that end users would believe is relevant

Aram: current implementation, measuremment and optimization requires level of detail to user or person
... to run optimization systems as they are now
... note Ben did a great job talking about certain limits in which noise can be introduced and systems can still be implemented
... concern is I understand they are different things
... optmization...use data
... and possibly separate those interest
... and find ways to address separately

Paul: I wanted to intervene
... we are talking about optimization to be used for manipulation
... what do we consider manipulation
... how do we draw the line; how do we define manipulation here?

AramZS: I can be quick
... when we talk about manipulation, we are talking about towards audiences
... a reporter searching for QANON
... found themselves being targeted for pseudoscience ads
... sent them further down the hole and created a reality of advertising around them
... that is manipulative to further a viewpoint
... gave example of algorithm redlining
... a person of color for example, regarding job search; jobs might be shown only to white people

<kleber> The quotes about manipulation from the break-out session included:

AramZS: that is the style of manipulation I'm talking about to be clear

<kleber> Tanvi Vyas (Mozilla): The downside is that you can target people, manipulate them, in ways that compromise democracy, in ways where you realize things about people that is more than they know about themselves Wilander: Agreed — news bubbles, radicalization, anti-vaxers, etc

Paul: Thank you
... how do we draw the line in detail
... can a human understand what trespasses the line we feel is acceptable

btsavage: I was going to talk about something else, but interesting topic
... I think we would benefit from having an open, honest, and constructive dialogue
... about our concerns with optimization
... for web advertising
... hard to come up with solutions without clearing defining the problem to solve
... If Apple invests millions of dollars into an evocative, emotional advertising campaign
... and puts on TV, billboards, digital
... if they don't use user targeting, does that mean it is not manipulation?
... would that campaign change beliefs or sell products?
... why does that not mean it's manipulative?
... not sure how to distinguish
... between ad targeting and manipulation
... if we are worried about filter bubbles
... surely this happens with contextual advertising
... let's say I'm on Breitbart or Fox News, I would see advertising that confirms my beliefs

<Jordan> Let's not conflate content or search result optimization with optimization of ads, the former of which can lead to negative manipulation, the latter of which is much moreso around optimizing which commercial products you see ads for

<AramZS> Here is the Pseudoscience article - https://themarkup.org/coronavirus/2020/04/23/want-to-find-a-misinformed-public-facebooks-already-done-it

btsavage: I wonder if filter advertising is not @
... not sure if Privacy CG or this group is the right group for the discussion

Wendy: Thanks, Ben

<bleparmentier> +1

Wendy: maybe that topic is another one to queue up for our virtual F2F
... to get some of the broader participation and perhaps research inputs into that

<joshua_koran> +1

<arnaud_blanchard> I agree with Paul and Ben

bleparmentier: I agree with what Ben just said

<AramZS> And I wrote very briefly about the line between broad and specific user targeting, in theory, though I will not claim to be an expert, but specifically about the line Ben refers to on a thread here: https://twitter.com/Chronotope/status/1270500232569278464

bleparmentier: and go back to point
... about optimization
... in current proposal on Sparrow
... not target single user
... concern with privacy and not follow user across the web
... I don't see anything that blocks the issue you are raising
... if you went to QANON web site, you will be private, but you will be targeted with ads
... agree with Ben
... what has been flagged as an issue is not at all what is served here
... my impression is creepy ads
... how can we serve
... can do manipulative ads on contextual, on TV
... none of proposals I've seen here
... back to optimization
... talkinga bout ability to just choose between two publishers
... and two ads from same publishers
... maybe one is not available; at bottom of page
... this is optimization
... very restricted, because of cohorts
... limiting the user
... not target with sophisticated ads
... know which ads is interacted with by the user
... this is something necessary
... so on TV you watch ad
... ad on newspaper you see the ad
... when you do ad on web site
... you need to be able to measure and optimize it
... ability with cohort
... to learn what is working
... and not a single user

<GarrettJohnson> Here is a summary of studies on the value of a cookie (in the current environment). Twitter: https://twitter.com/garjoh_canuck/status/1166037247386370048?s=20 Google Doc: https://docs.google.com/presentation/d/1juu6UBguR7ru1Rhfyor9knFBFsyLNP6FmIiXcJKyJZM/edit?usp=sharing (I've posted this here before.)

bleparmentier: it's very important; with cohorts we should never do anything that would be done in a bad way
... publisher side, I don't see why it's manipulative and why it's an issue

Wendy: thanks, I have closed the queue after Kris
... to get to other topics as well

apascoe: Calling back to
... what some people earlier were saying
... I think the optimization thing is straight up critical
... we see same thing at NextRoll
... not significantly drop publisher revenues
... but we are conflating optimization with user tracking
... maybe be different with big players with user posts
... as pure marketing tech company
... we straight up strip out user IDs,
... we care about high level features that go into those models
... I think we could potentially tease those things apart
... despite talking about publisher revenues, look at user experience
... if publishers cannot make revenue, there will be less content
... or publishers will try to protect that revenue
... and we will see more paywalls and required logins
... so users have to provide some sort of stable identifier
... then we see identity providers brokering identities to publishers for better optimization

<jrosewell> Here's another good link from Garrett on the issues apascoe just mentioned around market concentration - https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3477686

apascoe: if we go too far, it will undercut what we are trying to do here with these proposals

James: a very high level thing from Privacy CG
... it confirms for me that we just don't know what we are designing for
... endorse suggestion that TPAC F2F focus on these issues
... not repeat mistake
... if you are a hammer every problem looks like a nail
... need more diverse set of use cases ...
... transparency and consent framework
... invite other organizations to spread knowledge
... and come back to these problems and to Ben's point have a better idea about what we are designing the future for

Ben: I understand, not ....optimization
... bring some folks in to talk about this
... If people are open
... there are journalists and privacy advocates to reach out to and speak to their concerns
... that optimization is digging deeper into the demographic targeting
... and causing lift and that is a problem
... if people are interested, I can speak with journalists at The Post

Brian: I think that

<btsavage> I would love to listen to these people you mentioned Aram

<joshua_koran> @Aram if particular demographic targeting is the core issue, this seems distinct from optimization (which as Ben mentioned) might work with only contextual targeting

Brian: we have to assume that optimization is a fundamental part of advertisint
... point of showing someone an ad is to manipulate their thinking

<wseltzer> [Aram, thanks, that sounds to me like valuable input]

Brian: but if a user wants to understand why they are seeing something, then they have a way not to see it
... if we have anonymity throughout ecosystem
... how do we figure if an ad campaign is inappropriately redlining or targeting or breaking the law?

Kris: in terms of manipulation
... advertising and marketing is manipulative, so is information
... problem is not that it is manipulative so much as a user understanding as to why they are being targeted
... when an ad pulls at their heart strings, they know it's an ad
... if it's contextual, they understand how it's linked when they went to a certain page
... but problem is when ad is targeted and they don't understand where it came from
... goes beyond a boundary of what they can understand
... that is what a lot of data privacy folks focus on

Bigger picture: Are we capturing key use cases? Do we see gaps

Kris: more about understanding why users are being targeted

Wendy: This leads into the next item on the agenda
... Looking at our bigger picture
... previous discussion was helpful
... maybe separate things into phases
... one is describe the user cases and what is required for optimization for a use case
... and then thinking about concerns the way that use case currently operates
... ways it would operate; ways to address those concerns
... commit useful parts without the harmful parts
... easier to say than find answers

<angelina> i want to point out that advertisers (and agencies) - most/predominantly are looking at this granular data to append to user level data in their database...and that most info is to gain insights on how to improve...and NOT manipulate users...they believe they are "influencing".

<bleparmentier> But the questions are not asked! I 100% agree with ben, WHAT DO WE WANT TO SOLVE?

Wendy: use case demand and express concerns about privacy or societal aspects of it
... and look for optimal solutions
... to that

<angelina> i mean are NOT appending user data in their database

Wendy: and get many views around the table
... Angelina, you were trying to queue before

Angelina: Having been on agency side, working with hundreds of advertisers in my career
... advertisers are not looking to manipulate but to provide insight to provide the best messaging and approach
... having all this data and doing stuff wtih it
... is not a correct perception of what they are doing
... they want numbers and data to inform
... at least within platform they are using
... they are not collecting data for a massive profile of their users

Wendy: Thanks

TPAC planning ahead, issues and agenda

<wseltzer> https://github.com/w3c/web-advertising/blob/master/meetings/TPAC2020.md

Wendy: Let's look a little bit at the virtual F2F meeting and TPAC generally
... we have two days of a virtual F2F, similar to much of thinking borrowed from the Privacy CG; two, four-hour sessions
... take time to plan ahead for topics and invite outside speakers
... I have already added optimization; questions of privacy manipulation to list of suggestions there
... welcome other suggestions
... and look to sort those and choose specific times
... time slots for those
... other opportunity we hae

s/have

scribe: is breakout sessions
... anyone can make proposals for a session
... and invite participation across the W3C community
... if some other items come up
... and it would be great to hear about what someone in Immersive Web or Web Audio thinks about these ideas
... the breakouts is an opportunity to float ideas
... any other topics?

Wendall: we talked about different modes for user agent- browser thing
... idea coming up again
... modes to browser that facilitate movie delivery, other copyright laden content like sports

<Jordan> We've spent 45 minutes discussing the use case of optimizing the delivery of ads, and what have we accomplished exactly? Have we narrowed down what we're trying to solve for? No. +1 to Ben's suggestion to narrow down the problem statement. In the meantime, does anyone believe the AppStore is not optimizing the display of apps?

Wendall: this detente we are trying to develop
... the stakeholder perspective, the balanced nuance
... we could talk about different modes
... I would like to talk about that, willing to facilitate that discussion

Wendy: Thanks
... browser modes and media modes as a shorthand?

Wendall: yes, seems that this is already propped up in practice; an idea we have been playing with inside Verizon

bleparmentier: it would be interesting to have a session
... stating what are the problems with 3d party id and what we think we are doing to solve them
... lay out why we think this is exactly the issue
... of the effort
... what we are proposing; think this is very important
... an interesting meeting topic

Ben: I would like to host a break out session on web views
... this topic has come up a number of times
... cross device and cross user agent ads conversion paths
... and how frequently those move between app and web view
... Aram mentioned difficulty of showing an ad
... for example Wash Post ad opened in web view
... versus other browser or other apps use like Snapchat or Twitter
... in-app web viewers is an interesting use case
... and app-bound domains
... is important to discuss
... app bound domains

<Zakim> wseltzer, you wanted to discuss dedicated issues

Ben: limit extent to which mobile apps can interact with web apps
... rich topic of conversation

Wendy: As we are developing these topics we should start issue threads for them
... to capture the background and explanatory material
... so discussion is not limited to the amount of time in the virtual F2F but can start beforehand and continue afterwards
... Privacy CG demonstrated that quite well; starting discussions from threads that started to tease out the question
... invite proponents to start an issue in Github

James: should that be this group's Github repository?

Wendy: yes, that is a good place for them
... or point to another discussion happening in another group
... provide the links

James: quite a few came out of the Privacy CG; will link to them
... want to raise a subject for discussion
... we touched on some
... this simplistic definition of first, third party and web browser
... people's trust varies in different situations
... trust needs to be understood; far too simplistic and leads to unintended consequences

Kris: +1 to what James said
... talk about first party and third party context
... sets is great, but also interested to hear what Apple is thinking about
... in blocking CNAME cloaking
... like to know more about logic they will be using

Wendy: lots of good suggestions and discussion
... a packed agenda by the time we identify all of this
... if there are things people want to tee up for discussion sooner, please send notes to mailing list with those requests for the weekly meetings here
... that takes us to the end of this call
... reminder to register for TPAC meeting
... period of virtual meetings is free of charge
... we ask people to register
... and make sure you re getting updates on the logistics of the meeting
... and the breakout scheduling
... and overview materials being made available

<AramZS> Can we link the TPAC stuff in here please?

Wendy: if you are newer to the W3C community, TPAC is good opp
... to see how W3C operates and how community as a whole works in break out sessions
... look at other meetings and opporutnities

https://www.w3.org/wiki/TPAC/2020/SessionIdeas#Public_Breakout

https://www.w3.org/2020/10/TPAC/group-schedule.html

https://www.w3.org/2020/10/TPAC/registration.html

For help with registration, please contact me

<wseltzer> Karen: apologies for the clunky registration system

<wseltzer> ... you can "register" for one group and "observe" multiple others

<wseltzer> ... If you have a W3C account already, should be able to register

<wseltzer> ... if you want to invite colleagues or guests, contact Karen or Wendy

<wseltzer> [adjourned]

<hober> Belated regrets; in a TAG vF2F today

<wseltzer> thanks hober

This is scribe.perl Revision of Date 
Check for newer version at http://dev.w3.org/cvsweb/~checkout~/2002/scribe/

Guessing input format: Irssi_ISO8601_Log_Text_Format (score 1.00)

Succeeded: s/documentation tool/optimization too/
Succeeded: s/privacy/Differential Privacy/
Succeeded: s/light/late/
Succeeded: s/optimization...manipulation/optimizaiton is not just manipulation/
Succeeded: s/for optimization...populations/between optimization and manipulation of populations/
Succeeded: s/frequency/frequency capping/
Succeeded: s/compromised/compromise/
Succeeded: s/@:/AramZS:/
Succeeded: s/user/pseudoscience/
Succeeded: s/gave/gave example of/
Succeeded: s/@:/btsavage:/
Succeeded: s/@/restricted, because of cohorts/
Succeeded: s/demographic/demographic targeting/
Succeeded: s/echo/eco/
WARNING: Bad s/// command: s/have
Succeeded: s/problems/problems with 3d party id/
Succeeded: s/@/app-bound/
Succeeded: s/see name/CNAME/
Succeeded: s/Britebart/Breitbart/
Present: wbaker_ ErikAnderson Karen kris_chapman imeyers seanbedford joelstach mlerra joshua_koran jrosewell Mike_Pisula_Xaxis blassey pbannist kleber arnaud_blanchard sharkey pl_mrcy lbasdevant btsavage bleparmentier arnoldrw AramZS gendler scottlow pedro_alvarado charlieharrison dkwestbr hcai ionel
Found Scribe: Karen
Inferring ScribeNick: Karen

WARNING: No meeting chair found!
You should specify the meeting chair like this:
<dbooth> Chair: dbooth


WARNING: No date found!  Assuming today.  (Hint: Specify
the W3C IRC log URL, and the date will be determined from that.)
Or specify the date like this:
<dbooth> Date: 12 Sep 2002

People with action items: 

WARNING: IRC log location not specified!  (You can ignore this 
warning if you do not want the generated minutes to contain 
a link to the original IRC log.)