Algorithmic decision-making

Throughout most of the 20^th century, one leader or ruling party tried to frame the policy visions of the future and then act upon them. Mostly, ideology and simple heuristics were used to accomplish this goal which led to clear slogans, e.g. supporting a free market economy, lowering taxes or investing in education and other public goods. Above all, it was beneficial for two reasons. Firstly, it allowed the political leader(s) to reduce uncertainty by having a long-term agenda. Secondly, it could be used to justify themselves towards their constituents.

However, we are currently witnessing a shift in western politics that has several symptoms. I will focus on three of them. Firstly, the formerly distinguishable profile between right-wing and left-wing parties is eroding. As a result, several European democracies are observing political tendencies towards the “middle” of society, where politicians are trying to appeal to as large a voter basis as possible. Secondly, the threshold of expertise in certain areas has become so high, that only a few people seem to be able to understand specific problems and design solutions. Thirdly, the scope of political agendas, visionary ideas or ideological slogans has been reduced significantly. Previously, the agenda has been designed for the long-run, but now the fast-pacing global environment seems to dictate the political agenda which requires constant updates. The simplest example is the financial crisis.

In Germany, the current chancellor Merkel is responding to this challenge by planning for the short-term and (re)adjusting to current problem(s), even if that means to contradict past comments. In response, the media mostly accuses the incumbent of being “pragmatic, yet without a vision” or “lacking courage”. However, these political judgments are using the wrong frame of analysis. It is not the politician that is lacking courage. It is the circumstances that have become overly complex. The predictability of political visions has been sacrificed in order to ensure short-term predictability. As a result, the interconnectedness of social, economic and political problems does not allow politicians to force their agenda upon reality anymore. The empirical glimpse that experts are getting from using large data-sets and algorithmic analysis shows, that simple recipes and ideological slogans won’t suffice to solve problems. Although far away from being perfect, the analysis of big data exposes the simplicity of ideological slogans, e.g. “raise taxes” or “cut benefits”, on all sides of the political spectrum. The slogans are just not cutting it anymore and people are losing faith in the political system.

In short: the crisis symptoms of the political class, where leaders make false promises or contradict themselves, are most likely due to an overly complex reality that is much more prone to the use of algorithmically based decision-making than to human errors. It is therefore no coincidence that the application of algorithms and big data is increasing at the same time than the political “courage” is decreasing.

If this analysis holds, then there are several conclusions to be drawn. Firstly, the era of individual leaders with big visions is likely to be over. This includes the political trench wars and ideological phrasing of certain problems. So far, visions and agendas have been useful as a political compass. In the current environment, however, they can be harmful artefacts of the time when leaders had to reduce uncertainty by acting under unknown conditions. This was necessary because in a situation with lots of unknown variables you could rely on a clear and desirable vision, using trial and error methods and then hope for the best. Contrary to that, marginal improvements and efficient problem solving will be much simpler in an algorithmic environment. Instead of fighting over the budget, we can actually focus on how to improve the living conditions, distribute wealth, or calculate the most efficient health insurance for a given situation.

Secondly, the algorithmic analysis will have implications for the political process. The hierarchical world of public administrations with its top-down decision-making process will likely be shaken up. A much more flexible model of decision-making is needed to answer how the policy-making process needs to be adjusted to be responsive and accurate? So far, public administrations seem to be unwilling to put themselves under the microscope.

Thirdly, algorithmic decision-making has overall implications for the democratic understanding. If politicians have to respond to problems by using sophisticated mechanisms of empirical analysis instead of doing political “guesswork” that is aligned with their campaign promises, then they will have to change their mind more frequently. Currently, this is considered to be a problem. However, it might actually be a huge advantage. To use the bonmot of the economist John Maynard Keynes who responded the following, when he got challenged for his contradicting theories: “If the facts change, I change my mind. What do you do Sir?”.

However, as citizens or media practitioners, we try to hold politicians to account by using their previous comments or their electoral promises. This is where the problem arises: on the one hand, these complaints are absolutely justified because we have voted for our representatives. On the other hand, it is unfair and harmful, since it only encourages politicians, journalists and citizens to continue with an on-going and well-established masquerade: citizens complaining about characteristics of politicians (greed, corruptibility), journalists using these simplified categories and politicians trying to evoke the illusion that they are actually still in control by responding to “human” categories of being “honest” and showing “leadership quality”. This is a tragedy, because such a democratic theatre does not expand the public discourse. In fact, it only obfuscates the new techno-social challenges. Simple nostalgia and mourning about a less complex past is misleading.

Instead of watching that theatre, we need to address the underlying questions: how can we use algorithms to facilitate problem analysis and decision-making inside institutions? How can we balance this new form of decision-making with the accountability towards the public? And above all, is our current democratic understanding actually well-equipped to address these questions properly? In starting that debate, we can avoid perpetuating the masquerade and begin to understand one of the most vital challenges of the 21^st century.

Behavior change is one of the most difficult things to achieve whether you’re trying to alter consumer purchases or harmful lifestyles. I was able to share my thinking on how mobile apps contribute to the war being waged in becoming our better selves at Editorial Intelligence’s Mobile World Conference in association with Vodafone, The Huffington Post and Channel 4.

I see two main models that inform how app developers hope to change behaviour: the rational and social learner. Apps that involve logging new data about oneself expose the previously unseen. Apps to track fitness and diet, spending or transport could be employed by anyone with health or efficiency goals. You log calories, see how much you eat, and using this new knowledge, adjust your diet accordingly. In economics 101, rational actors would see this new data about themselves and respond accordingly. This process could work where actual costs exist, but when it comes to complex social relationships like the one we have with food, it seems a bit flat and unrealistic.

Another model involves seeing the data generated by apps as embedded within an aggregate of their users, a.k.a. social networks. Apps that engage in building a world of social learning allow users to see how much other people exercising, spending or commuting. Apps can publish your data out to commingle with others on your social network, or provide a dedicated network of people using that app. Some apps, or settings within apps, allow us to see where we are in the bell curve anonymously. Of course, this can elicit negative feedback when we see our spending is below normal, or exercise above average. Both (1) generating data about ourselves and engaging with it on a personal level, and (2) aggregating data where we see ourselves within a social sphere are effective methods to remember as we move to some case studies on dating, health and financial apps.

Firstly though, let us consider the tool box. Intersecting functionalities are equipping would-be behavior changers with a powerful new means of tradecraft. Apps can draw on social, mobile and local, or ‘SoMoLo’ capabilities. Perhaps a brief explanation is required: Social refers to the ability for apps to integrate with social media platforms or incorporate their elements, mobile refers to the spreading adoption of ever more powerful mobile telephony, and finally, local refers to the ability to tailor content from both actively and passively collected built-in GPS data.

Considering the obvious power of this trifecta, we could assume that app developers would incorporate them all by default, or at least in the majority of their projects. I took an extremely quick pass at coding the top 25 most popular free and paid apps in Google Play, i.e. Android phone marketplace, and the Apple App Store. Based on each app’s description in the store, I determined which of the ‘SoMoLo’ functions it possessed. To be completely honest, this data is not particularly robust and could use a second pass, a deeper dive beyond the product description, better definitions to guide the coding and more coders to generate reliability. However, as a way to gauge the market, it serves.

Surprisingly, you can see that less than a quarter of the most popular apps take advantage of the trifecta. Apple scores below Android and Android’s free apps employ the most ‘SoMoLo,’ with 55 percent of apps falling into that category. Many scholars, members of the popular press, and consumers have pointed out that Apple’s more restrictive regulation of developers and its centralized approach toward approving content reduces innovation. The point being: we have a toolkit of social, mobile and localization technologies that has not penetrated the market, which means both user and developer literacy of how to use them appropriately for behavior change remains nascent.

On to the first case study… In the way that porn has changed the Internet, Grindr has changed mobile apps. Grindr is an example of a technology that perfectly suits the task. Members build profiles usually just a photo and a few details. People rarely use their real names. The home screen shows who is online in your immediate vicinity. It shows who is online within a user designated distance such as 1 mile or more with accuracy to 350 ft. You can view profiles for those who are online and message any with whom you might be interested in coordinating a meet up. A friend and frequent Grindr user told me that it’s a dream realized to have a secret flag over peoples’ heads declaring their orientation and intentions that only other gay men can see. Grindr has hands down changed how a large swathe of people can meet their partners for life or just the night. Some good practices can improve personal security and relieve some of the burden during public meet ups; discussing consent and what you’re into is easier online before a meet up than in an initial face-to-face, and convenience should not be underestimated as a real value for people. There are negative social consequences such as objectifying individuals in an online market, but as an antidote to the bar pick-up culture, or for those in areas without gay clubs and bars, Grindr has ultimately helped a lot of people make connections and change their romantic lives.

Photo attribution: Editorial Intelligence

Engaging in learning through a social app must also be informed by the actual character of the network. An app that is a blockbuster for one community cannot necessarily replicate itself elsewhere without considering the dimensions of the new community. The flirting/dating apps Skout and Skout+ have interfaces that look and function nearly identically to Grindr, except that they are aimed at a general audience which skews more heterosexual. Instances of sexual assault and rape of underage users by adults have given the app the pedophile stigma (see Slate [http://www.slate.com/blogs/xx_factor/2012/06/13/the_social_app_skout_tried_hard_to_keep_teenagers_safe_and_still_failed.html] and/or Jezebel [http://jezebel.com/5917897/this-social-media-startup-accidentally-become-an-app-for-pedophiles]articles). Downloads have suffered accordingly. This is every developer’s worst nightmare, but not a totally unpredictable one given an overly broad network approach that disregards power differentials and adequate safe guards against the victimization of users.

A lot has been said about the potential for so-called mHealth initiatives to tackle some of the biggest public health challenges, specifically obesity and diabetes. Interestingly, Tufano and Karris (2005; http://www.ncbi.nlm.nih.gov/pmc/articles/PMC1550687) contrasted failures of big public health campaigns such as “Five a Day for Better Health” in California and “Active for Life” in the UK against the potential of using mobile health initiatives to change behavior. Looking at the market today, I see two kinds of apps. Those targeted to very specific goals, such as 100 Push Ups, RunTracker and Free from Smoking, while others offer a more general approach to health, including the likes of MyTracks, Healthier and LiveStrong. LiveStrong in particular is a favorite of mine. Medical researchers Halko and Kientz (2010; http://www.springerlink.com/content/37250x67089m2736/?MUD=MP) have said that mobile apps would be more effective in the health arena if they were more tailored to individual personalities. LiveStrong offers a suite of apps for different goals and sets of behaviors with a matching online community and services. You can decide what level of depth to go into with the app and how much to share, or to what aspect of your life to apply the tools. I also appreciate the lack of feminization of diet, fitness, and weight loss. The black color scheme and lack of cuteness makes it gender neutral, perhaps even skews it male. LiveStrong is able to balance social learning that is both personalized and general in user appeal.

Financial apps provide us with one final example of behavior change apps. With this category, I want to highlight the development process. I’m also a fellow with IDEO.org, which aims to “design a better world with everyone.” They apply human-centered design to the social sector. Recently, they began development of an app that serves as middle ware between microfinance institutions and their clients to encourage savings. Before beginning market research or production, they sat down and asked the clients how they saved money. Then they brought those same clients a prototype to evaluate. You start by looking at the behavior you’re hoping to change before you attempt any kind of intervention. It’s a basic principle, though often overlooked.

Challenges to using mobile apps for behaviour change are numerous. But fundamentally, we need to design better apps. I see how few apps are opened more than once (somewhere around a quarter of all apps; http://mobithinking.com/mobile-marketing-tools/latest-mobile-stats/e#mobile-app-flops) and I think we have a lot of work to do on passing that first impression test based on interface and design. We have to pass this hurdle before people make the intimate decision to allow pocket-based mobile technologies to improve their behaviours.

This post was adapted from a July 19th post on www.mollyfrancesnorris.com.

Portrait of Alan Turing by Daniel Rogers © Daniel Rogers

“I wish I had,” said Steve Jobs when asked if he had designed the Apple logo as a reverence to the computer pioneer Alan Turing who committed suicide with a poisoned apple. Nevertheless, we can draw a line between the mathematician and contemporary technology companies. Without Turing the development of computer science, the end of World War II and the advances in Artificial Intelligence would have taken much longer.

We celebrate Turing’s centenary and although his name is generally unknown to the wider public, he will enter history books and the pantheon of science right next to Newton or Darwin. The celebration of his centenary might accelerate that process and raise awareness for the sleeping giant. But we need to be careful not to create another simplified genius narrative.

In celebration, the Internet is full of dedicated, reminiscent homepages, blogs, articles and other digital votive offerings. A quick Google search of “Turing” yields 16 million results. The online cemetery procession is led by Google’s doodle rebuilding a version of the so called Turing machine. Said machine consists of a strip with numbers allowing computer scientists to simulate algorithms and understand the limits of computation.

Instead of using a humorous or art-like doodle, as Google usually does for anniversaries of scientists, the Turing doodle was a practical riddle that needed to be solved. In six steps the user could interact with the doodle and eventually spell Google in binary code [0,1] – great start to simultaneously scare and excite people.

Why does he deserve to be in the scientific pantheon? Turing was born in 1912 in Britain, finished his studies in Cambridge with First-Class Honours in Mathematics and became a fellow in Cambridge at the age of 22. During the war, he ran the unit in Bletchley park that was responsible for cracking the code of the German Navals. He developed a machinery to decrypt the enigma codes and thus shortened the war by at least several months providing valuable information about German strategies. His former colleague described him in an interview: “The extraordinary thing is that this quiet man was probably the most important man of his time, except possibly Churchill. Turing did not look like a superstar, he was a very modest man, but Turing was the genius who broke Naval Enigma.”

Before and after the war he worked on several scientific papers about computation and laid the foundation for modern computer science. For him, there was always the possibility that “machines can think”, and he concluded one of his papers with the bold prediction: “We may hope that machines will eventually compete with men in all purely intellectual fields.” Algorithms, AI, software – these are words that have become meaningful due to the quiet work of Alan Turing.

After the war, Turing was convicted for homosexuality, which was a crime under British law until the 1960s. Instead of going to prison, Turing chose a chemical castration procedure and was fed with female hormones. Two years after the treatment, at the age of 42, he apparently committed suicide. In fact, his housekeeper found him in his bed with a half-eaten, presumably poisoned, apple. This act is believed to be a symbolical reference to Snow White, a story that was said to have haunted Turing. Although possibly apocryphal, this narrative offers a clever and imaginative end, fitting a clever and imaginative man. As the Italians say, if it is not true then it is at least well told.

Since the chemical castration and his personal tragedy are probably connected, prime minister Gordon Brown publicly stated an apology, following an online petition in 2009:

“While Mr. Turing was dealt with under the law of the time and we can’t put the clock back, his treatment was of course utterly unfair and I am pleased to have the chance to say how deeply sorry I and we all are for what happened to him.”

However, the apology was not accompanied by an official overturning of the conviction although several people fought for a symbolic clearing of charges. This gesture would have rectified a problematic, albeit legal in the positivist understanding of jurisprudence, entanglement of the British law and the tragedy of one of its most important scientist.

But history is never certain and we should be careful using the simple narrative of a haunted genius that committed suicide as such constructions might be misleading. Professor Copeland, a Turing expert who recently spoke at the University of Oxford, argues that it was not necessarily suicide; murder or an accident are equally as possible. For instance, the apple was never analysed to contain traces of cyanide, Turing was said to be quite cheerful in the weeks and months preceeding his death and his mother remembers a cyanide experiment he was conducting.

Copeland summarises: “In a way we have in modern times been recreating the narrative of Turing’s life, and we have recreated him as an unhappy young man who committed suicide. But the evidence is not there.”

Without any doubt, Turing was brilliant, he contributed immensely to the developments that are now so familiar when holding a iPhone, using a notebook and surfing the Internet. Regardless of the circumstances of his death, there lies still a bitter irony in the fact that the scientist who contributed most during the war was criminalised by the same nation he tried to preserve with his service during WW II. However, we should also be cautious. Whenever there is a genius to be made, we are inclined to use these categories of a tragic hero (Turing) or an eccentric genius (as in the case of Einstein), mainly to render the incomprehensible complexity of their work tangible. In Turing’s case, his personal life and work becomes more familiar through the tragedy.

The familiarity of the tragic hero narrative may well help cement his story and achievements in the wider public memory. This is not, in itself, problematic. However, this seems to be the beginning of the creation of a genius narrative that inevitably leads to misunderstandings and simplifications. But it also acknowledges that the forgotten genius has materialised. His legacy will have to be continuously analysed and once the personal tragedy is well known, the actual focus on his work and achievements can begin.

The last sentence of his seminal paper on intelligent machines still holds: “We can only see a short distance ahead, but we can see plenty there that needs to be done.”

“YouTube is popular.” There it is, folks. The safest sentence I have written on this blog. With 60 hours of content uploaded every minute and 4 billion page views every day, the pre-eminent video sharing site has found monumental success. But since 2007, what can be less confidently asserted is that YouTube is a champion of user-generated content, a bastion of hope for the layman with a camera or video file.

Of course, a statement like this was tautological when YouTube was created. The only content on YouTube was of the user-generated variety, and so the site fostered the impression that user-generated content was king. But when corporations began to take notice of the video-uploading sensation, they thought two things. First, that YouTube was a threat to their revenues. Second, that YouTube was a new market of opportunity. For both reasons, they began to upload their own, professionally produced content to the site. And since then, a growing body of academics (Kim 2012, for instance) believe that the entrance of corporate, professional content will change the YouTube community’s perception of the amateurs, eventually crowding them out completely.

A doomsday notion, for sure, but certainly one that is testable!

Particularly in the category of music videos. While rock and pop stars have started to funnel videos through YouTube using corporate usernames, amateur musicians have been uploading their work from the beginning. Some of these are called ‘YouTube artists’ based on impressive fan bases, high quality content, and frequent new additions. For me, these ‘YouTube artists’ seemed ripe for a comparison with billboard artists who have signed onto a record label and are thus funded by traditional media.

I decided to look into whether YouTube users respond more or less favorably to the videos of YouTube artists or billboard artists. To judge favorability, I developed a program to collect the comment threads of four videos (2 by YouTube artists, 2 by billboard artists) and used a sentiment analysis technique (SentiStrength, created by Thelwall et. al. 2010) to score each individual comment. I culled the feed to the most recent 300 comments for each video in order to create a manageable amount of data.

Why didn’t I simply use likes and dislikes? Because with that measure, there were only two sides to favorability; I wanted a more granular scale to bring out the various shades of the variable. I also didn’t trust YouTube users enough to know they weren’t simply ‘going with the crowd’ in liking or disliking a video. A comment, however, was a conscious choice to single oneself out (however anonymously) and express something publicly and so was a more reliable indicator of how one actually felt about the video.

The four videos chosen were Alex Goot’s “Sensitivity” (link), Boyce Avenue’s “Find Me” (link), Gavin Degraw’s “Chariot” (link), and Nickelback’s “Leader of Men” (link). Spare me comments about my personal music tastes – these videos were selected simply because they all had view counts within 370,000 of each other, a reasonable equalizer for view counts over 3 million at the time I did this study. The former two were products of YouTube artists, while the latter two were products of billboard artists. To control for genre, both Alex Goot’s video and Gavin Degraw’s video were classified as pop, and both Boyce Avenue’s video and Nickelback’s video were classified as rock.

As you might have guessed from the title of this post, I found that when genre was controlled for, the YouTube artist’s video had significantly more favorable ratings in the comments feed than the billboard artist’s video had. SentiStrength calculated both a positive and a negative score for each comment (scaled from 1 to 5 and -1 to -5 respectively). The comparison amongst the average positive and average negative ratings for a comment on each video can be found below:

The comparison of average positive sentiment scores and average negative sentiment scores from each comment thread

While the difference in average positive ratings of a comment between the two YouTube artists’ videos was not statistically significant, every other comparison amongst the four numbers was. Clearly, the results offered evidence that YouTube artists’ content were not only still appreciated; they may actually be appreciated more so than billboard artists’ content.

Of course, as displayed, Alex Goot’s video was also the most unfavorably received on average based on the comments feed. This was the only statistically significant finding based on the average negative sentiment scores. So Alex Goot’s video seemed to be the most polarizing video, with the second highest average positive score and the highest average negative score. Is this evidence that some YouTube artists are already becoming less appreciated and that user-generated content is doomed?

Most likely not. There were numerous alternative explanations for why Alex Goot’s video received the highest average negative score per comment. For instance, SentiStrength could be blamed for its acknowledged weakness in scoring negative emotions accurately. In addition, a single argument in Alex Goot’s comment thread could be blamed for its ability to compound disapproval and unnaturally bring down the average negativity score of a comment thread; none of the other videos displayed an argument in their most recent 300 comments. Either way, “Sensitivity” did not seem to have been scored the most unfavorably based on Alex Goot’s status as a ‘YouTube artist.’

What remains is the reassurance that the ethos of YouTube as a promoter of user-generated content is not unfounded (at least for music videos!). This does not mean, however, that YouTube artists produce videos that parallel corporate, professional content in terms of quality. There are more qualities of a video and an artist that users take into account when they are reacting favorably or unfavorably. One such quality that might make up for a difference in quality would be the level of personal relationship an audience feels like it has with a YouTube artist; users frequently appeal to YouTube artists directly, requesting particular covers of songs. No such relationship is evidenced in the comments section of billboard artists’ content.

It is worth ending on a note of caution: what happens if this personal relationship with a YouTube artist dissolves? Billboard artists are clearly separated from YouTube users as a distinct group, and YouTube artists are already delineated from the rest of the YouTube community by virtue of their popularity. If that delineation becomes too palpable in future and there is no line of communication between these lead users and the rest, then YouTube artists might well become classified as professionals, and their content, too. User-generated content would then again be called into question. It will be interesting to see whether this scenario can be avoided, or if people like Kim (2012) are simply predicting a future that hasn’t happened yet.

 “Once we become critical of the assumption that the Web is a neutral repository of information, the structure of the Web becomes much more interesting.” – M.H Jackson, 1997

Absences speak volumes, and yet, interpreting information gaps online has produced only muffled truths. Studies on the geographical origin of Internet content have shown old divides between rich and poor countries repeat themselves online. For example, the vast majority of the shares of Google’s user generated content, academic journal citations and authorship of Wikipedia entries tilt to the wealthier global north. Admittedly, exploring digital landscapes is far less adventurous than the globetrotting variety. However, these journeys allow us to peer not at the contours of physical unknowns, but at the shapes molded by the collective human brain — its associations, biases and desires codified in the traces left behind transacting in cyberspace.

And so I set forth on a small exploratory study of my own. I wanted to make a first pass at charting the informational disparities within one of global citizenry’s largest projects geared towards egalitarianism: the international aid and economic development industry. I chose worldbank.org as the single best available database with worldwide content coverage, peak traffic relative to its competitors and missions to not only create a world free of poverty, but also a website serving as the number one source of information on this cause. An enterprise website is more than the face of an organization, as some in public relations suggest. It is the cognitive expression of the organization’s values and intent. What image of the world would the self-described “knowledge bank” reveal, as mediated by its staff online?

Instead of examining knowledge output per region, I used hyperlink counts as a measure of both volume, but more relevantly, as an indicator of visibility. As a basic organizational element of the Internet, Halavais describes hyperlink networks as increasingly “meaningful, malleable and powerful.” Hyperlink networks are meaningful as mirrors of the human associative brain at work in a digital environment. Once aggregated, hyperlinks record relational connections made by Internet users between information at the unconscious and intentional level. Hyperlink networks are highly malleable in that they are susceptible to being reordered and manipulated by actors within the network once network position is observed and made valuable. (The benefit of restricting the crawl to a single network neutralizes much of this risk since gaming the system through search engine optimization occurs between websites rarely within them.) Finally, hyperlink networks are powerful enforcers of structural relationships between information. As Manuel Castells explained in “Rise of the Network Society,” information is the raw material for much of the work that goes on in the global economy’s new technological paradigm. The current evidence of digital material domination by wealthier countries provides some evidence to this truth.

Using Thelwall’s SocSciBot, the crawl for the hyperlink network began with worldbank.org and cast a net of all pages within two clicks of the homepage. This resulted in 1,180 valid HTML pages featuring 283,787 links and 751 links to specific regional content after removing duplicates, machine-generated and corporate pages. The regional links were discovered through manual filtration based on URL structure that included regional keywords and were then evaluated for content. I ran a quick regression to make sure links weren’t merely a factor of variables related to the organization’s work. I specifically controlled for a region’s population, poverty rate and dollar amount of World Bank financed activities. None of these variables were statistically significant predictors of the number of links to regional content. Seemingly, the study would be able to capture the raw connections between data, free from the obvious distortions brought about by different regional characteristics.

Hyperlinks by Region on WorldBank.org

The findings reveal that Africa’s link count of 148 is nearest to the mean (125) of any other region; Africa is not marginalized within worldbank.org. This is surprising, considering other information geographies have shown Africa pushed to the extreme periphery. In academic knowledge, Switzerland is represented as three times the size of the entire continent of Africa using the JCR database. In geo-tagged Wikipedia articles, there is more than twice as much content about France as there is about Africa (according to one study). World Bank’s web provides a counter example to a digitally excluded Africa.

Elsewhere, the comparison between the East Asia Pacific region versus South Asia presents a notable contradiction. East Asia is underrepresented, relative to the sample, especially when compared to South Asia. Both regions host a behemoth, China and India respectively, and have similarly high populations, poverty count and amount of financed activities. And yet, World Bank staff appear more willing to link to South Asian than East Asian content. This disparity brings up awkward questions about preferred narratives and politicization within the development industry. Is the questionability of China and other Asian Tigers as a recipient for development financing echoed online? Or does South Asia just prove more salient in development knowledge? It is hard to know what this absence is saying. But as I said, this is just the initial sketch of a new territory, the informational landscape of international development online.

For more information, see this article.

Who should get to decide how long Facebook should keep personal data or how they should deploy targeted advertising? The future of privacy will be decided by little tweaks in the phrasing of regulation terminology and users are mostly unaware of these changes. The recently proposed alterations in the privacy rules will facilitate data processing procedures and are deemed problematic by the Irish Data Protection Commission.

When dispersed challenges to their corporate decisions manifest themselves, Facebook can usually rely on a natural network effect because of its impressive member base. Users rarely object because they are either unaware,  or the alternative to opt-out is not feasible. For large parts of our social life, the network has become almost indispensable. However, this “Matthew effect” – the phenomenon that the rich become richer – is problematic if the company uses its weight to silently push the privacy rules. The chance to vote in favour or against the new privacy rules could be regarded as a step into the right direction if it were not so obviously flawed.

Did Facebook voluntarily become more democratic? Of course not. The referendum on the new privacy rules is the result of active campaigning by the advocacy site “Europe versus Facebook”. Led by the Austrian law student Max Schrems (previously involved in a debate about the destiny of his user profile on Facebook) the site is fighting for transparency, minimization of data collection and enhanced regulation to facilitate platform migration. After more than 7000 comments on their governance page, Facebook decided to let the entire community vote for themselves. Every user has the chance to vote within one week from the 1st of June until the 8th of June.

More specifically, the vote is a farce for two reasons.

Firstly, in a democratic election people will be informed about when and where to vote. On the Facebook site governance users can find the proposed and existent versions. In order to be binding, a total of 30% of Facebook users need to vote, otherwise the voting result will only be considered “consultative”. That necessitates a total of 300 million votes. One day before the election ends, 290,000 users have voted. Putting it simply: the quorum will not be reached. Facebook has designed this outcome. Only users that “like” the “Facebook site governance” will be notified about the voting process.

Facebook should not be exempted from a proper voting procedure. Especially given the fact that the company has arguably transformed into a virtual state that privately governs its virtual territory. Legitimacy in this space can only be the result of a democratic and transparent process. Facebook clearly failed to design an appropriate voting procedure.

If the democratic ideal would have been taken seriously, the voting ballot should have been placed on top of each user profile. You cannot force people to vote, but you can draw their attention to it. Instead of trying to control the outcome, Facebook could have gained legitimacy by designing a democratic voting procedure that serves as a model for future regulation processes. This virtual state needs to take their self-declared citizens more seriously. Although it is clearly unlikely that private companies will be governed by user votes, the potential should not simply be wasted by setting up Potemkin villages. Contrary to the historical example in Russia, this is not just a flawed digital referendum, it is also plain obvious that it is flawed. It will therefore yield greater negative PR ramifications than the previous silent changes.

A real chance for user empowerment has passed. Even worse, the fact that only a fraction of the users is aware of the vote and the statement that the new rules will be automatically implemented if people continue to use the site, reveal a problematic understanding of the governance and user ideal in the largest virtual state. Far from being democratic, this behaviour resembles much more the policy of a semi-authoritarian state.

The Google insight chart below shows the relative volume of ‘privacy’ and ‘piracy’ in news headlines since 2008. What we see here is that often after an upward blip in the public salience of piracy, there is a corresponding upward blip in the public salience of privacy (there are, however, spikes in privacy that are seemingly unrelated to piracy salience). This fits with a conception that the public salience of privacy (as measured by media attention) is driven by privacy advocates responding to specific campaigns (like anti-piracy measures) which are threats to privacy.

It is not totally clear what the time-lag here is, however, I adjusted the data to account for a two month lag in privacy salience to quick-check the hypothesis. This leads to a fairly notable correlation (.28) that is statistically significant (p = 0.0000).
A quick regression tells us that, essentially, for every one point increase in the salience of piracy now we can expect a .44 point increase in the salience of privacy two months down the road.

When discussing the internet, the economic concept that seems to have made the largest dispersion into popular discourse is the concept of ‘network effects’. “Facebook  is unconquerable because of positive network externalities.” “Product X must reach the tipping point so that network effects can take over,” etc.

This sub-field is often treated as if it is more-or-less mature (and it may well be), but there is a serious theoretical ambiguity within the economic literature about exactly what a network effect is based upon.

The original incarnation of the term deals with a good whose value increases as the number of nodes on the network increases. The traditional example is the fax machine. As more people received and hooked up their fax machines, my fax machine (or potential fax machine) increased in value because the number of people I could contact with the device increased. This is a node-based network effect.

But there is a difference between this node-based effect and a quantity-based effect. For instance, imagine a good whose value is not connected to how many other people use the good but to how much of the good exists out in the world. This quantity could be held by a few large entities, or distributed among many smaller other entities. The important factor is: no matter to whom a good is sold—even if they already own the good and are buying a second or a third—the value one places on the good increases as this good is bought. I call this quantity-based good ‘Facebook’ and the node-based good ‘fax machines’ for simplicity from now on.

But why ‘Facebook’? It seems that the reason Facebook is valuable is because all of one’s friends are on it. This would make it a traditional node-based network effect good, and no different from ‘fax machines’.

Here we reach the weaknesses of economic theory: all economic theory is necessarily a simplification. According to Ariel Rubinstein (2006) economic theorists are fable tellers (which is not a bad job, to be sure). They simplify life, create models that in some instances have absurd results, and hope that their models can still be true-enough-to-life to tell people something interesting and valuable about the world they live in.

Even in the first network goods papers the theorists recognized that the node-based concept was a simplification. Obviously adding a family member to the fax network is more valuable than adding a stranger (see Farrell and Klemprerer). Also, sometimes nuisance nodes pop up whose addition to the network degrades quality. But the general trends hold—as more people use the fax network, more value is placed on owning a fax machine.

So there is no purely node-based network effect good. Likewise, there is also no good that has utility curves that look like those I assigned to the good called ‘Facebook’. But by treating these extreme cases as theoretically possible ‘corner solutions’, we can attempt to be fable tellers discussing the relative importance of the different factors by talking about moving away from one corner and towards another.

Facebook is valuable for me because of how flushed out my friends’ Facebook pages are and how often we reconnect. So Facebook (though still partially based on node-based network effects) is more valuable to me the more time my friends spend on Facebook. If my friends each spent an hour a month on Facebook, it would not be worth as much as it is when they spend 10+ hours a month. But fax machines are a more ‘bursty’ kind of communication. The value of a fax machine is of having one turned on and left alone in the house.

The fax machine is clearly worth more to me if people use it more (it is not a corner solution), but this effect is not as important as it was when considering the value I place on Facebook. Unless you never check your fax machine, the fact that you are not proactively reestablishing connections through fax does not really degrade how much I value the machine. It sits idly in my house, and alerts me when it needs attention—the quantity of nodes is more important. Facebook requires updating and continuously refreshing contacts/uploading photos—nodes are still important, but the quantity of time spent online is more important than it was for fax machines.

Therefore, if we think of network effects as being on a continuum, extending from one extreme where only nodes matter to the other where only quantity matters, we might find that goods online are located in a different part of the continuum than the goods we were comparing them to before the internet.

This has implications on how firms treat consumers already in the network. If only the number of nodes is important, the competition will be more fierce over the marginal consumer who is indifferent about joining the network. If the quantity purchased is relatively more important, firms will also take into account the effects of the additional quantity purchased by inframarginal consumers (those ‘not on the margin’ or already in the network).

The specifics of the model are a bit mathematically complex, but the difference between these cases comes down to the reverberation of the increased number of nodes/quantity. The firm obviously still cares about the inframarginal consumers in the ‘fax machine’ model. But the firm benefits more from a new customer than from a customer buying a second fax machine because of the increase in value that other consumers enjoy–part of which the firm can now extract.

On the other end of the spectrum, the value created by the increase in quantity purchased is the same if someone buys a second good or if someone enters the market for the first time. This can explain why some network good providers spend relatively more time catering to their current customers (who are supposedly ‘locked in’) than they would be predicted to.

The real concern is that this differentiation has not been recognized in the network effects literature. Or, perhaps more fundamentally: should the things on this continuum even be called network effects anymore, or should they just be subsumed into a more general  category like ‘consumption externalities’? The distinction is important because there are times when it will lead to different predicted results in models because the change in inframarginal demand was previously not considered as affecting the value of the good for others.

Theoretical economists may only be fable tellers, but people seem to be paying attention to this specific story. It is worth properly defining the factors upon which it is built.

Having a Facebook page is becoming more and more of a liability. Surely we’ve heard it all before, though. Journalists, authors, bloggers, and even occasionally incredulous Masters’ students love talking about the potential negative Facebook effects, from loss of self-esteem to increased anxiety or jealousy.

But there’s a much more tangible one: you can be expelled or not hired based on what is posted on your Facebook wall.

This is hardly a newsflash depending on how jaded you are about invasions of privacy, but the situation has gotten worse. In 2009, the University of Oxford’s student population discovered that their administration was using Facebook to spy on their pictures and subsequently fine them for any wrongful behavior. As recently as four days ago, a report surfaced that three 14-year-old girls were expelled based on wall posts in which they jokingly discussed who they would like to kill. I’m assuming the school had never heard of childish games like MFK before.

Through 2011-12, cases emerged where employers would ask for one’s Facebook account as part of their application process. Recently, this exercise has been taken to its extreme. In a disturbing new case, Maryland’s Department of Corrections forces job candidates to scroll through their Facebook profile while a recruiter looks over their shoulder, in what is being called a compromise brokered by the American Civil Liberties Union; recruiters initially wanted candidates to simply disclose their usernames and passwords.

Let me be clear: this is not your average “Facebook leads to negative things” post. Because at this point, it is easy enough for one to believe that most Facebook users know the potential risks of using the site. People must also know that regardless of how inaccurate these judgments turn out to be, they continue to be made. The problem is the illusion of control that everyone assumes we have over our own Facebook accounts, and how we could always directly prevent situations that befall those unfortunate few. Of course, to the extent that we do have control, we should exercise it by posting content with an awareness for potential consequences, refraining from extending friendship invites to everyone and their mother, and putting friends in ‘lists’ to which reasonably differentiated privacy settings are applied.

However, complications arise at each step of this process that neither academic researchers nor policymakers seem equipped enough to handle. It is widely believed that Facebook became popular because it took less time and effort for people to communicate with each other when they wanted to (see Baron 2007, for instance). But in order for them to place people into lists, this time and effort would need to be increased, particularly with a large, unlisted, existing pool of friends. (Incidentally, Google’s “Circles” innovation, Google Plus’s raison d’être, seems to show the benefits of starting from scratch with an organizational schema like this, but switching to Google Plus would also require more time and effort.)

The most popular solution has simply been to educate people about the risks associated with Facebook use, and certainly, that measure should still be taken. But even in those places where the issue has been over-saturated, incidents such as those mentioned above are still prone to exist. They may arise because Facebook changes its privacy settings, forcing every user to re-visit any and all hard work they already put into cultivating their relationships with the proper viewing privileges (do I want everyone to see my cover photo as well as my profile picture with Timeline?). They may arise because apps that we use are becoming more precocious and hold their services at ransom to us, the payment being the publishing rights to our walls (can I really not play Words with Friends without having everyone see how nerdy I am?).

But mostly, they may arise because of our Facebook friends.

Numerous academic studies assume that because we can control who we are friends with, we can thereby control the content of their posts (most recently in Chou & Edge 2012). These studies are not in touch with the realities of social pressure, both on and offline. Can we control who we are Facebook friends with? Probably not. We confirm Facebook friends all the time to avoid unpleasant situations, such as the one where an office neighbor asks why I have not confirmed his friend request yet. And can we truly censor the content of their posts? Technically, we can with a delete command. But realistically, we cannot because the post may have already received a number of likes and comments from people who knew it was there and would notice its disappearance. And even if they had not noticed, the original poster certainly would have.

In this way, Facebook is even worse than real life. Nowhere in real life do one’s friends make such a consistent and indelible impression on how that person is viewed by others. And it is bad enough when it is a Facebook friend who is judging you, but when it is a school, an office, or even the government, the consequences tend to be much more dire.

Clearly, the whole story is not being told unless it is mentioned that interactions with your Facebook friends can also provide numerous benefits, from social support to an awareness of news and events in the world around you. But in no way do these interactions need to be public to be effective. Facebook users should be able to decide more efficiently which of their wall content gets to be viewed publicly.

We have passed the point where Facebook profiles are generally considered private. That point continues to recede into the distance as Facebook inspires more and more public displays and invites more and more public organizations to interact with its members. And while the culture surrounding Facebook use has changed to one that accepts ‘going public,’ that acceptance is not often accompanied by excitement but with resignation.

Because every day we give ourselves over to Facebook is another day that our characters may be soiled by an errant post, video, image or behavior that receives a little too much publicity.

I wrote a piece recently for The Atlantic entitled, “It’s Not All About You: What Privacy Advocates Don’t Get About Data Tracking on the Web.” The thesis of this piece was, essentially, the power that massive user and behavioral data gives private corporation (and the asymmetry between them and everyone else) is more important ramification of this new ecology than any individual level breech of personal information. Creepily targeted ads aren’t really the downside, insane levels of influence and control are.

I think you get the gist of the article from the opening paragraphs (I am learning about this journalistic dont-bury-the-lead structure thing):

Jonathan Zittrain noted last summer, “If what you are getting online is for free, you are not the customer, you are the product.” This is just a fact: The Internet of free platforms, free services and free content is wholly subsidized by targeted advertising, the efficacy (and thus profitability) of which relies on collecting and mining user data. We experience this commodification of our attention everyday in virtually everything we do online, whether it’s searching, checking email, using Facebook or reading The Atlantic Technology section on this site. That is to say, right now you are a product.

Most of us, myself included, have not come to terms with what it means to “be the product.” In searching for a framework to make sense of this new dynamic, often we rely on well established pre-digital notions of privacy. The privacy discourse frames the issue in an ego-centric manner, as a bargain between consumers and companies: the company will know x, y and z about me and in exchange I get free email, good recommendations, and a plethora of convenient services. But the bargain that we are making is a collective one, and the costs will be felt at a societal scale. When we think in terms of power, it is clear we are getting a raw deal: we grant private entities — with no interest in the public good and no public accountability — greater powers of persuasion than anyone has ever had before and in exchange we get free email.

But I am writing this post because I feel compelled (by some comments from people I respect on twitter, solon and Harley Geiger) to clarify a thing or two about my position. First a caveat or two: 1) I greatly respect the editorial staff at The Atlantic; 2) I understand the click-driving nature of titling pieces.

I think my piece was mis-titled. Generally, free-lancers don’t title their own pieces – the editors do that based on their own set of criteria and the editorial goals and standards of the publication.

I wrote a piece about reframing privacy/data tracking debates in terms of power. I think the power dimension is under discussed in public fora, and is not the first thing average internet users think about when they worry about people using their personal data. As I said, “the bargain that we are making is a collective one, and the costs will be felt at a societal scale.” I think this societal level concern is a bigger deal than the personal privacy (or at least as important) and should be reflected as such in the public discourse.

The title of the piece (again, not my doing) would indicate that I think that privacy advocates don’t understand the power aspect of data. But of course they do. Anyone who spends any time thinking and working on these issues knows how powerful this data can be, and thoughtful people (if so motivated, as the great folks working in this field are) will draw out the logical societal scale implications of that power. So let me repeat. I DO NOT think that privacy advocates don’t get that big data is about power as much as privacy.

I think when privacy advocates make their case to everyone else – the non wonk, non compsci, non info sci etc. set – they frame the issue in terms of individual level considerations more than in terms of these larger scale imbalances that are created. I think this is a strategic mistake. Not that the individual level stuff isn’t a good sell, but I think there is value at coming at the issue from both angles. It illustrates that this new phenomenon is bad on a variety of levels at different scales and has more significant implications that strike at the root of our individual autonomy ethos. The power discourse is a useful counter argument to those who say that the data-for-free-services deal is a fair trade.