/// Facebook Graph Search Is A Disruptive Minefield Of Unintended Consequences
I was not at the Facebook announcement in Menlo Park on Tuesday, so I neither drank the KoolAid nor felt compelled to respond immediately. I’m glad I waited. Facebook’s new Graph Search feature is both a bigger and a smaller story than what most observers were expecting. On the one hand, this will now be an integral part of the product, a third pillar, as Zuckerberg described it, along with the newsfeed and timeline. On the other hand, for consumers, this is somewhat of a product in search of a use-case. I don’t think that users have been clamoring for this—but marketers have.
So I found myself pulled (much as Nate Elliott from Forrester Research wrote in these pages on Wednesday) between thinking that Graph Search is a snore and that it is much more (and a potentially dangerous more for Facebook and its users, as well.) For the service to be a success, and by extension for Facebook to continue to grow and thrive, it has to convince users to change their behavior. This is not an easy thing to do. The search functionality will not work well for users if the data is “sparse,” in the terms of former Google engineer, Lars Rasmussen, who is one of the key architects of Graph Search. Facebook has to get users to provide more signals of the kind that will be picked up by search, at this point this means “Likes” of pages, check-ins to places, shared photos and fully filling out their (now all publicly searchable) profiles.
But Facebook also has to convince users of the utility of searching for “Mexican restaurants that my friends like in Palo Alto.” Although the result is targeted specifically to you (unlike Yelp or to a lesser extent Google) your Facebook friends may not be very good judges of Mexican food—or much else for that matter. I would venture that very few people use Facebook in a strategic enough way to make most of these “social” recommendations necessarily better than less personalized ones. If the majority of your Facebook friends (and their friends) are foodies, than maybe restaurant recommendations will work. But do you also agree with their taste in movies, books, music or (to use one of Rasmussen’s examples) dentists?
There is a difference between ones actual friends, accumulated over time and different phases of your life, and what I call your “content friends.” I find myself using Facebook and Twitter, for instance, in very different ways. My Facebook friends are people I actually know. I don’t necessarily agree with (or care about) their taste in music or food or technology, but I have an affection for them that I want to maintain. On Twitter, on the other hand, I follow people who are into all of the kinds of things that I like to write about. Very few of the 374 people that I follow do I actually know in person, but that’s not the point. I consider them my “content friends.” We’re into the same stuff. They tip me off about new things way before they appear on Mashable. It’s like mainlining pure, early adopter tech intelligence. So when I use sites that filter or curate the content I see, I’m much more likely to use my Twitter account as the starting point.
This is all preamble to the fact that I think Graph Search is indeed important, but the results of Facebook’s search for increased relevance may be both more and less than it intends. Its users may find the utility of searching their own social graph to be hit-or-miss, but they also may find themselves feeling much more exposed in the searches of others than they ever intended to be. Rather than phrase this negatively, however, I want to try to identify the potentially explosive issues, land mines if you will, that Facebook will encounter in its path to build out its third pillar and suggest what it needs to do to avoid or diffuse them.
The very name, Graph Search, shows that Facebook is enthralled with its own internal workings in a way that does not really connect with its users. As much as the company uses the term “social graph,” it is only really developers who think about things in those terms. For users, this is “social search,” plain and simple.
Similarly, at the press conference, Zuckerberg and his engineers talked with obvious pride at making the search of the social graph possible. It is, indeed a tough problem that they have apparently solved, but is it the problem that most users need them to solve most?
Again, most likely, it is the advertisers (and Facebook’s own bottom line) that most need this problem to be solved, but the presentation on Tuesday made it sound like this was all about the users. The problem that users need the most help with is understanding and acting upon the privacy implications of the data they already have on the site and how to feel in control moving forward.
Another contradiction that Facebook will have to face is the asymmetry between how much control users and marketers have to search the social graph and how programmatic the tools are for people to control their own data. To say that users can change the setting on each individual piece of content is not a very generous offer. Facebook knows that most users won’t bother. But just the fact that users could have controlled a bit of content that later they are unhappy about having shared is not enough. Privacy is not a black and white issue. A Woodrow Hartzog and Evan Selinger point out in The Atlantic, there is a big difference between privacy and obscurity. Even if data is technically public it can be obscure and relatively hard to access. Data has physics—it can have friction.
Adi Kamdar at the Electronic Frontier Foundation (EFF) makes the same observation, “Facebook’s Graph Search presents the problem of discoverability. One can have a good balance of privacy and openness if information is available, but not easily discoverable. You might not mind if people specifically interested in you look at your Likes, but you may not want to have a market researcher pull the list and add it to an ad targeting profile.” Discoverability is a form of data lubrication, which is exactly what Facebook intends. But the power of the discoverability needs to be balanced by users’ ability to make their content obscure or actually private in an equally powerful way.
Matt Asay writes in a post on ReadWrite, that he recently took himself off Facebook (in solidarity with his 13-year-old son) and has felt immense relief. “This reprieve from the onslaught of social media and its attendant army of notifications (“Lonn likes your post. He really likes it!”) couldn’t have come at a better time,” he writes. “Now that Facebook’s Graph Search is set to make it even easier for random people that I’ve accepted as ‘friends’ to search my interests and take action based on them. (‘Hey! I like The Hobbit, too! Want to go to the community theatre’s production of Bilbo and Me?’)”
The fear for Facebook is that given the tools to more actively (and conveniently) control their content, users will absent themselves from the commons that Graph search requires to be robust—they will make themselves “sparse.” But the converse is that people will have one too many experiences of being overexposed and, like Asay suggests he is considering doing, “completely remove” themselves from the service.
This is the place where I wish I were a mathematician, statistician or logician. I’m not even sure which one I need to be to solve the Facebook privacy equation, but I have a simple question. Given the structure of the Facebook social network and the rules in place, is it provable—one way or the other—that a piece of a user’s content will be seen ONLY by the friends they intend to see it? I’m thinking of the Randi Zuckerberg situation over the holidays where a personal photo inadvertently got posted publicly by a “friend of a friend.”
My strong hunch (and I welcome expert feedback on this) is that it can be provable to be impossible. Unintended consequences are a natural part of large networks. I think this is true and I think that the engineers at Facebook well know it. If it is indeed possible to prove the converse, they should post the proof for peer review. Otherwise, I have to say that the privacy assurances, particularly now that a large chunk of user content including profiles and photos are searchable, are disingenuous.
Facebook’s contention that nothing will come up in a search that has not been shared publicly with the person searching may be technically true. And making it true may well be a monumental engineering problem that it is proud to have solved. But the fact is that most users do not realize all that they have shared, or have shared things in one context and never intended them to be shared in another. As I have suggested above, without tools to make users aware of and in control of all they have shared, there is no way that there will not be many embarrassing and even life-destroying revelations that result from the everyday use of Graph Search.
What will make Graph Search more useful to and less problematic for users will be the efforts of developers both inside and outside of Facebook. As Zuckerberg et al. suggest, searching the social graph is filled with possibilities, but I don’t think that it is universally useful. Different use cases need to be considered and built out. Risks need to be assessed and confronted. This is a big project that will benefit from a diversity of opinions and approaches.
The problem for developers who want to work on the Facebook platform is not only that the social network keeps changing the rules, but that it requires a quid pro quo for access to its APIs. As TechCrunch reported on Friday, Facebook just cut off acccess to its Find Friends API from social messaging app Voxer because the startup was taking data from the network and not giving content back. According to the story, “If a startup shares back content such as photos or Open Graph stories, Facebook says they’ll be free to use its Find Friends Data. I believe that’s because Facebook can monetize that content with news feed ads. If a messaging app doesn’t share anything back and qualifies as a competitor [and] it will only be able to use Facebook’s login system.”
The reason why Voxer does not “give back” is that users have no interest in posting their private conversations on Facebook. Duh! But the reason Facebook is tightening up on Voxer right now is because they have launched its own messaging app. So Facebook was not a “competitor” when Voxer started, but finds itself to be one now.
Maintaining a robust enough developer ecosystem to make best use of Graph Search for users will require more assurances from Facebook that startups efforts will not be voided as it rolls out its own apps. But Facebook also has to look at what it requires from developers in order to have reliable API access to Graph Search. Especially with search which is individualized by definition, public postings don’t make much sense. No question it will be in Facebook’s interest to foster a diverse community of developers trying to pull value from Graph Search. For that to happen it will have to create trust, transparency and some new rules for API access.
Perhaps the thorniest challenge of all is the issue of ambiguous intent—of irony. Unlike computer code, which needs to be clear and consistent in order to run at all, human language and behavior is inherently complicated and filled with multiple, often conflicting, meanings. So if you search the social graph for “people who like Chuck Norris,” for instance, you may get people who both legitimately “like” the reportedly homophobic action movie star, others who have “liked” him with an ironic wink but certainly don’t consider themselves homophobic and still others (younger folks particularly) who invoke his name because it’s “random” and they’ve seen his name in online memes but don’t even really know who he is. What is the value of such a search?
Facebook’s assumption, built into its algorithms, might be that this problem sorts itself out because people who like Pabst Blue Ribbon earnestly have other friends who do as well, and the same is true for PBR among hipsters in Williamsburg. But, in reality, if you are from Wichita and move to Williamsburg, you could have friends from both camps.
And this is not just a matter of noisy results, but of potentially embarrassing disclosure, as well. In a recent story on Gizmodo, These People Are Now Sharing Horrible Things About Themselves Thanks to Facebook Search, Graph Search autocompletes some really nasty stuff and calls up profiles of people who have “liked” things that in retrospect they shouldn’t have.
I described a similar problem in relation to Facebook’s sponsored stories feature that led to and embarrassing association for a user with a 50-gallon drum of personal lubricant. The physics of data, indeed!
Irony and “randomness” are so enmeshed in youth culture at this point that it is no wonder that younger users have been finding Facebook less relevant to their lives and more annoying than their less ironic parents. This is a big problem for Facebook’s future. Zuckerberg’s empire is beginning to seem like the literal minded dad who doesn’t get the joke!
Facebook move into search is filled with potential and pitfalls. How well it works out has to do with how well it addresses the issues I have raised above (and undoubtably others I have missed.) I don’t see Graph Search as much of a challenge, at this point, to Google. Google already has some social signals in its algorithm and will add more. Facebook and Google will have different value for users, but I don’t see Facebook’s social personalization surpassing Google’s contextual personalization any time soon, if ever.
The real question for Facebook is how well it can use these search capabilities to deliver users to advertisers without alienating those users and causing them to flee the platform. The related question is how generous Facebook will be with their third-party developers who attempt to deliver relevance to users through the Graph Search APIs. Business as usual will cause developer flight as well.
In the mean time, what can users do as Facebook ramps up its search capabilities? Most importantly, make sure you are OK with what you have posted and liked and shared and remove or add friction where necessary. Below is Facebook’s own tutorial video about how to control what shows up in Graph Search. The network effects of the activities of friends and friends of friends is downplayed here in an attempt to make this as simple as possible. The EFF has posted some additional tips for how to protect your privacy from Facebook’s graph search, if you want a second opinion.