Science and Information Conference 2015 July 28-30, 2015 | London, UK

Decoding, Hacking, and Optimizing Societies: Exploring Potential Applications of Human Data Analytics in Sociological Engineering, both Internally and as Offensive Weapons Gregory Maus [email protected]

Abstract—Today’s unprecedented wealth of data on human activities, augmented by proven reliable methods of algorithmically extrapolating personal information from limited data, and the means to store and analyze it opens up new vistas for in-depth understanding of individuals, as well as the potential generation of predictive models for the dynamics of human functions on individual, group, and societal scales. This has already proven to have applications in successfully forecasting behavior, techniques which are only likely to improve. To the extent that the science can move beyond a correlative understanding of the data to a causal understanding of the factors affecting behavior, it will allow new means for (perhaps covertly and deniably) influencing behavior, possibly through long causal chains that could conceal the influence of the manipulator. This offers an immense variety of applications, but this paper will particularly consider them as tools in governmental control over their citizens and as a new form of weaponry. Keywords—cognitive security; computational sociology; big data; sentiment analysis; machine learning; surveillance; privacy

I.

THE AGE OF OMNISCIENCE1

A. Obligatory Panopticon Cliché It is difficult, if not impossible to avoid constantly sharing data about our activities that are recorded and archived. A report by the White House summarizes this, saying “we live in a world of near-ubiquitous data collection.” [1] Whether it be cellphones recording our locations, credit cards and online accounts recording our purchases, cookies recording our movements online, search engines recording our queries, communication companies recording our metadata, social media recording what we say, government forms recording major life events, medical documents recording the details of our health, or any of a hundred other data channels Credit for coining the phrase “Age of Omniscience” as the one we are entering belongs to U.S. Federal Trade Commission Commissioner Julie Brill, who used it in her 2014 presentation in Vienna [65]. 2 As of 2013 56% of the U.S. adult population owned a smartphone and 91% own a cellphone of some kind [66]. The numbers in developing countries are not far off, with 37% of Chinese adults owning a smartphone and 95% owning a cellphone of some kind in 2013; even in Pakistan, which had 1

today (and surely still more awaiting in the data-hungry future,) we are freely giving away an unprecedented amount of information about ourselves. Just as importantly, with the continued fall in costs of data storage and processing, that data is more valuable than ever. Furthermore, as smartphone 2 and social media 3 adoption around the world continue their dramatic rise, in-depth data will be available on an ever-greater swathe of humanity. B. The Human Manifold The explicit data collected about us has also proven effective as a basis for reliably extrapolating implicit personal information. Microsoft researcher Thore Graepel, refers to this as “The Human Manifold,” the grand set of interrelations (probable and actual) between data points and traits that can be used to infer other data points and traits about individuals. He elaborates in a private e-mail: “I picture a cloud of points in a very high dimensional space, which lie on a lower-dimensional manifold – like a plane in 3d space. Hence, knowing some of the attributes/dimensions allows us to infer others – at least in a statistical sense.” Dr. Graepel was one of the three researchers in the joint Cambridge-Microsoft team that produced a widely publicized demonstration of this effect. The team created an algorithm that used Facebook Likes alone to relatively accurately match the study’s 58,000 volunteers to their results on psychometric studies (and certain types of information available from elsewhere in their Facebook accounts) in such fields as stated sexual orientation, personality traits, religion, ethnicity, gender, cigarette and alcohol use, relationship status, and political affiliation [2]. The prediction accuracy varied between traits, but was not insubstantial overall. Members of the team then applied the lowest adoption rate of the 24 developing nations surveyed, 53% of adults owned a cellphone [67]. 3 Internet access rates have been growing everywhere, but still vary significantly around the world. However, social media use is almost ubiquitous among internet users around the world. In the 24 emerging countries that Pew studied, a majority of internet users used social media—except in China, where only 48% did [67].

1|P age www.conference.thesai.org

Science and Information Conference 2015 July 28-30, 2015 | London, UK

the models trained on Facebook Likes to identify the genders and ages of 3.3 million unique Microsoft Live users in the U.S. based on 133 million Bing search queries without a significant loss in accuracy [3]. Though Microsoft Live does not collect information on users’ political and religious views, when the team predicted those views based on the Facebook models they achieved estimates for searchers in each state commensurate with Gallup polls and Pew survey results for each state (assessed by the location associated with the searchers’ IP addresses.) The successful transfer of this model from one platform to another, rather different one without a great loss of accuracy suggests the breadth of possible inferences that can be accurately gained from the models derived from other sources. A major study whose findings could also be valuable in assessing what can be derived about one communication channel or another data source is the Copenhagen Networks Study. This multi-year study combines comprehensive data sets on university student’s social media activity, geolocation, telecommunications, psychographic studies, demographic information, and personality inventories of 1,000 university students [4]. Considering the depth and variety of data being analyzed on this study, analysis of this data could prove something of a Rosetta stone for extrapolating information across data sources. Other studies accurately deriving personal traits from limited digital data abound. The social media-specific studies include predicting Big 5 personality traits from other types of publicly accessible Facebook profile information [5] [6] and predicting personality traits from Twitter [7] [8] [9] [10]. For non-social media studies, many analyses have long found linguistic markers in text that are strongly indicative of various personality traits [11]. Lately, IBM’s Watson in particular has been touted as a tool for identifying personality traits in depth from short writing samples, [12] with researchers going so far as to claim that an accurate portrait of one’s personality (52 traits) can be derived from the textual content of just 200 tweets [13].

C. Reconstructing and Analyzing Social Networks In addition to extrapolating information about individuals, several studies have identified means to identify the structures of social networks. A Rochester study designed an algorithm based on Twitter behavior to accurately infer not only users’ personal relationships, but to accurately predict other users’ physical locations at any given time based on their friends’ activity [14]. The U.S. Naval Postgraduate School’s CORE Lab has also made substantive strides in using open source data to identify and map the hidden social relationships amongst terrorist groups [15], Syrian insurgent groups [16], and FARC guerillas [17]. One unnamed official reviewing the project was sufficiently impressed with the analysis of Syrian opposition groups that he or she noted: “this would have taken an entire intelligence section numerous months to develop the analytical products and potentially years for the intelligence community

to develop the sources and contacts with access to this type of information” [16]. D. Analyzing the Flow of Ideas Other studies have focused less on humans as the unit of interest, per se, but the diffusion of ideas throughout social networks. Due to its highly open-source nature, several studies have analyzed Twitter, identifying many statistically discernable patterns in the diffusion of ideas [18] [19] including cultural variations [20], topical variations [21], and the impact of community structure [22]. Other analyses has found key traits in determining the emotional resonance of content and its impact on content virality [23]. Analysis of movie quote popularity has also found statistically meaningful lexical patterns, a potentially useful finding for those who wish to spread slogans [24]. Several quantitative models have also been established for the diffusion of news articles online, including one study tracking 90 million articles over the course of three months [25] and another tracking 170 million blog postings and news articles over the course of a year, finding deep statistical patterns [26].

II.

FORECASTING

With the data thus gathered and inferred, it becomes possible to formulate predictive models and thus produce forecasts about behaviors. Some successes have already been achieved in this area. A. Individual Behavior Though individual human behavior is often held as an ideal of unpredictability, this view may underestimate the human tendency towards routine. One 3-month study with 50,000 randomly selected mobile phone users using location data whenever users made calls found that human location patterns were 93% predictable, with little age or gender-based variation [27]. There is also substantive evidence that the behavior of an individual’s social peers is strongly predictive of their own behavior in a wide variety of fields [28]. B. Group Behavior The behavior of groups has also been forecasted with some accuracy. The winner of IARPA’s contest in forecasting protest movements achieved fairly high accuracy in predicting demonstrations across Central and South America using open source data with over a week of average lead-time in most cases [29]. Another study developed an algorithm analyzing gang interrelationships to identify key members of street gangs who,

2|P age www.conference.thesai.org

Science and Information Conference 2015 July 28-30, 2015 | London, UK

when persuaded to leave gangs will also cause others to do so and successfully tested said algorithm in Chicago [30]. In addition to group dissolution, group formation has been found to be predictable through Facebook-derived traits, at least in self-selected college project teams [31]. C. Large-Scale Aggregate Outcomes Broader, society-wide trends have also shown forecastable from aggregates of open source data. A U.N. study found that social media sentiment analysis could be used to forecast unemployment figures in the U.S. and Ireland with lead-times of 2-3 months [32]. At least one study has used social media to forecast movie box-office success [33]. Social media forecasting has also shown some promise in forecasting financial markets [34]. D. Unifying Models Thus far, all (publicly discussed) social forecasting models have been relatively specialized, limiting their general applicability. This could be the result of many factors: limited project scopes, academic siloes, data siloes, model transference difficulties, and processing power costs. The political factors associated with privacy concerns could also serve as a barrier to trans-platform data unification, providing an advantage to researchers not dependent on public good will. It should also be noted that there may be broad swathes of individuals that successfully obfuscate their data from collection, particularly those highly concerned with privacy. This may or may not come to represent a blind-spot in analysis, depending on the ability of such systems to make inferences about these individuals or work around them. However, if these factors can be overcome and unified models established, it may open up a wealth of potential opportunities. III.

FROM CORRELATION TO CAUSATION4

It’s a well-known, but no less true cliché that correlation is not causation. Thus, models that rely purely on correlation may be effective in forecasting, but not in “understanding” the actual underlying factors that actually inspire behavior. One way to disentangle the two could be through experiments such as the 2012 Facebook Emotional Contagion study [35]. However, given the public backlash associated with the study’s publication, such studies might be most easily be carried out either in secret or by companies, governments, and 4

Much of the following sections is speculative, wading into consideration of a new field that requires considerable further study to more clearly define its capabilities and limitations. Thus its conclusions must by necessity be taken with an understanding of all the uncertainties inherent in discussing a newly developing field of science. 5 The majority of the examples used in this article are based around influencing a single individual. This type of example is

other organizations less dependent upon the goodwill of their customers, citizens, and publics. A. The Human Sociome If the causation factors can indeed be disentangled from correlation, it could allow for the creation of grand yet granular explanatory models of human behavior en masse, a sort of “human sociome.” It seems likely that such models would be iteratively developed as one clarification builds on top of another, opening doors to new investigations and refinements in turn. A detractor could argue that this is not qualitatively different than previous sociological studies, but this belies the dearth of data and means to process it that limited previous studies to broad assertions often reflecting pre-existing sociological doctrines. In contrast, the omnipresence of modern data collection allows for a more revealing window into sociological dynamics than was ever possible before. B. Quantum Social Butterflies In the classic example of the “butterfly effect” or “quantum butterfly” the flapping of a butterfly’s wings in one part of the world causes a chain of events that determines the path of a storm in the world somewhere else at a later point in the future. The concept has been used to illustrate how small changes in the initial conditions of complicated systems can have complicated and far-reaching impacts on long-term outcomes. The resulting chaos may place limitations on the abilities of forecasting sociological systems. However, this should not be taken as evidence against their fundamental feasibility. The butterfly effect is generally held to have originated with weather forecasting, based on the observation that small differences in the starting parameters of weather models affected their outcomes drastically. Despite this, weather forecast accuracy has been steadily improving over the decades as models improve, with apparently little sign of hitting a plateau [36] [37]. On the contrary, the butterfly effect may be the key to this field’s most promising applications by allowing for subtle, minor adjustments which have probabilities of triggering major differences in outcomes that could be predicted through advanced sociological forecasting. For a straight-forward example 5 of how this could be applied, suppose that an authoritarian government identifies their citizen John as a potential future dissident radical based on the tone of some of his online communications and the fact that according to his phone’s geolocation data he regularly meets used primarily for its relative concreteness and ease of understanding. It should not be taken as a sign that influencing individuals would necessarily be the primary focus of such actions or the vector by which objectives would be achieved. Indeed, the law of large numbers would suggest that the behavior of large groups would be more predictable and thus more predictably influenced.

3|P age www.conference.thesai.org

Science and Information Conference 2015 July 28-30, 2015 | London, UK

with other dissident radicals, all of which fits with quantitatively validated models of dangerous radicalization. Worse, based on others’ reactions to him he’s been shown to be charismatic. Such a figure could become a nucleus for a group opposed to the government: a butterfly with a greater than average probability of causing a storm of trouble for the authorities. Fortunately for the government, based on the psychological profile developed from his data, his passions and ambitions could be redirected in a more favorable direction if he were more invested in the status-quo. Thus, happenstances are engineered to incorporate John into the establishment: perhaps he is sent a job offer from a government agency (where his behavior can be more easily monitored and/or adjusted), maybe he is sent advertisements for low-cost higher education (granting him a sense of economic opportunity), or his media newsfeeds are subtly tweaked to show him the violent excesses of revolutionaries against his ethnic group. If one or more of these methods succeed, John’s moderation could remove the nucleus of a nascent group of radicals, or even provide the core for a new group of government supporters. Thus, a few calculated small measures dramatically altered the probable course of John’s future, turning him from a potential dangerous enemy to a productive ally. With enough butterfly nets, a government might thusly pin the wings of a potential future revolutionary movement before it ever has the chance to get off the ground—but more of that in “Social Optimization.” C. How (Easy it is) to Win Friends and Influence People Some of the influence exerted through these methods could be covert, concealed by both the subtlety of adjustment and the length of the causal chain between the “nudge” and the desired outcome. We will return to the latter in the next section. There have been countless studies on altering behavior through subconscious influence over the decades, so for the sake of brevity, we will focus on selected studies in subtle online influence. The 2012 Facebook Emotional Contagion Study demonstrates the impact of the availability of friends’ emotional expressions on one’s own emotional expressions (as well as how altering the algorithm that prioritizes these expressions can in turn alter the emotional contagion) [35]. In a more subtle change to the environment, the administrators of the online multiplayer game League of Legends experimented with presenting various tips during matches (24 different tips across 5 different categories) in three different font colors (red, blue, and white) in three different locations (in the loading screens, in-game, and both) in random combinations against control group games that were given no such tips. Through hundreds of thousands of matches, it was found that precisely how the tips were presented had strong statistically significant impacts on player in-game attitudes, verbal abuse, and offensive language, even if the tips were just presented briefly before the matches started, often with significant variance in impact depending on the font color [38]. Showing an impact on opinion formation, a 2013 doubleblind study demonstrated that slight rearrangement of search

engine results could dramatically influence subjects’ views of political candidates when they had no prior knowledge of them, with the manipulation being unnoticed by the vast majority of subjects, an influence worth considering in light of the fact that 67.6% of the test subjects reported previously using search engines to learn about political candidates [39]. Taken together, these studies reinforce the suggestion that minor tweaks to the digital environment can impact behavior in foreseeable ways without the knowledge of the subjects. D. Setting up the Dominoes With sufficient understanding of sociological dynamics and knowledge of “local” conditions, it becomes conceivable that one could subtly influence events indirectly. This could be useful either to conceal one’s objectives or influence, or if one’s resources in the target space are limited. To illustrate, let’s return to the example of our protodissident John. It turns out that John might be suspicious of a government job offer, ads for higher education, or any news stories allowed by the state-run media. However, based on his psychological profile (derived from online activity/writing and geo-location data cross-referenced with tens of thousands of similar cases,) it is calculated as probable that his anger towards the government stems from a mix of father issues, unemployment, and sexual frustration. By resolving one or more of these, the chances of John’s radicalization would be diminished. One of the reasons that John’s father issues translate so cleanly into anti-governmental anger is that the father works for the government, which grants the government access to his work records (after navigating some bureaucracy of course, depending on the authority that the monitoring organization is granted.) From these reports, coupled with arrest records, they determine that the father has alcoholism issues that regularly affect his work performance and thus probably his home-life. Thus they subtly encourage him to become involved with an alcoholism assistance group: placing ads for them in searches, sending fliers to be hung up in the work place, mailing them to his address, etc. They could even act one step further removed by subtly nudging members of such groups into making presentations at locations that he is known to frequent. Alternatively, the government could nudge the father away from indulgence by simply making it subtly less convenient or less tempting. By increasing the probability that his father’s alcoholism will be resolved (and perhaps other issues through similar methods,) John has a higher chance of reconciling with him and thus a healthier figure and so less subconscious inclination towards anti-government sentiments. To resolve the sexual frustration, the government might analyze similar cases and determine which factors would contribute to resolution. Maybe a minor change to his behavior would increase his courting success, offering a suggestion through an anonymous friend of a friend or person at a club or bar. Perhaps there are a number of women he knows who have expressed a secret interest in him through one or more channels the government monitors, in which case a tip-off or a subconscious nudge to be more direct with the feelings might

4|P age www.conference.thesai.org

Science and Information Conference 2015 July 28-30, 2015 | London, UK

open up romantic avenues. Maybe there are many compatible women in close geographic proximity, but with whom he rarely interacts, inviting the fabrication of “coincidences” to increase their time around each other. One or more of these might succeed, offering John the increased probability for a potential outlet for his passions. Finally, John getting a job would still give him hope for the future and a sense of investment in the current social order. Firstly, John’s newly stabilized home and romantic life might contribute to the drive he needs to successfully acquire a job. If they’re not enough on their own, the government can, once again, cross-reference similar cases to identify possible solutions. Employment seminars, centers, or groups could have their marketing materials sent to him. If John has a relative or old friend with access to a job, subtly encouraging them to reconnect might spark a conversation. If he has an online resume it might be subtly sponsored to firms with a history of hiring employees like him. If John’s employment can be ensured, maybe it could improve his father’s pride, reducing his drinking, or perhaps it could boost John’s confidence to increase his courting chances. One might note that doing all of this for John would be extremely labor-intensive if performed by a human being, particularly if multiplied by thousands of potential high-risk dissidents. That’s where automation comes in. IV.

AUTOMATED INFLUENCE

A. Persuasion ex Machina It would seem intuitively plausible that the resulting sociological models will be rather complex, possibly even too complex for any single human being to understand in their entirety, especially as they continue to advance. Thus it may be most effective to implement many of the suggestions implied by the models with automated algorithmic agents. Even if the automated agents do not have this advantage, they are likely to still have advantages in cost and the ability to rapidly detect and respond to shifts in the broader social environment through constant trawling of relevant data. Of course, the (occasionally humorous) limitations of modern ad-bots should be kept in mind when projecting shortterm applications of this technology, but these faults should not be considered inherent in automated influence itself. In a joint study conducted by Telenor Research and MIT’s Media Lab, researchers created an algorithm to select marketing targets for a mobile network operator in Asia, based on analysis of call metadata and social network analysis. The algorithm then competed with the firm’s human marketing team (which was operating under best practices) to select the target customers most likely to purchase a phone plan upon receiving a marketing text message a common campaign practice in the market (sample size 200,000 in the treatment group and 50,000 in the top group that the marketers selected.) The algorithm vastly outperformed the marketers, achieving a 6.42% initial conversion rate (compared to the marketer’s 0.5%) and a second month plan renewal rate of 98% of converts (compared

to 37% of converts in the marketers’ selection) [40]. Thus, despite the humans getting the picking first based on their formulae and having the opportunity to be even more selective than the machine, it was 34 times as effective in selecting loyal customers. The authors of the paper also note that they were not given access to data about previous marketing campaigns and previously persuaded adopters, claiming “The conversion rates are likely to have been better if we had” [40]. In an e-mail conversation with the authors asking about study repeatability they asserted that they had tested the methods in other markets and achieved similar results, though they could not reveal information on the specifics. In addition to finding the ideal subjects for general marketing persuasion, research has also made significant progress in using open source data to identify the online users most susceptible to bots, potentially providing a vector with which to seed desired content [41]. Before one scoffs at the foolishness of those who can be influenced by automated agents, it should be noted that even supposedly tightly-guarded population segments can apparently be influenced, even infiltrated by blatantly deceptive online personalities. In 2010 security consultant Thomas Ryan fabricated an online persona named Robin Sage with a large number of intentionally glaring red flags in “her” stated background: a name that was a direct reference to a special forces training exercise, an implausibly long period of experience for someone so young, etc. Despite this, over the course of 28 days “Robin” accumulated hundreds of social media connections throughout the U.S. intelligence community, several of whom asked her to review papers by professionals with over 10 years of experience, offered her tickets to security conferences, and approached her about career opportunities [42]. B. Automated Content Creation In addition to providing content targeting, algorithms may have a role in producing content according to the needs outlined by the models and stated objectives, if developments in machine creativity are a guide. The field of machine creativity has been producing emotionally compelling content since at least the early 1990’s, when a skeptical Douglas Hofstadter found himself shocked and “truly shaken” by the apparent emotional depth expressed in the music composed by David Cope’s Emmy (EMI) program [43]. Expanding beyond music, David Cope has applied similar principles to also create algorithms that produce haikus [44] and visual art [45]. Food too has recently experienced a creative mechanical hand through IBM’s Watson. The developers of the program used a model built on analysis of the chemical properties of food compounds and their human-rated odor pleasantness to calculate the probable appeal of a dish, as well as novelty of ingredient combinations in comparison to a database of recipes from around the world [46]. A third instance of dynamically-generated content can also brag that its products are mass-consumed by the public. Narrative Science’s Quill algorithm has been composing

5|P age www.conference.thesai.org

Science and Information Conference 2015 July 28-30, 2015 | London, UK

hundreds of thousands of sports news stories for the Associated Press since as early as 2011 [47], has been regularly composing stories for Forbes (primarily in financial reporting) and other media outlets since at least as early as 2012 [48], as well as providing individualized narrative generation for used cars [49], in addition to offering automatically generated reports for a variety of industries and roles, including marketing analytics [48] [50]. The company’s clients also include the National Health Services of England, MasterCard Advisors, American Century Investments, and the U.S. intelligence community [51]. Indeed, members of the U.S. Intelligence Community have effectively endorsed the company’s technology since at least June 2013, when the company announced that it had received investment from In-Q-Tel, the non-profit strategic investment firm that works on behalf of the U.S. intelligence community [52]. Narrative Science’s executives and clients have stated that the tone of stories can be easily customized to their clients’ needs, from “a breathless financial reporter screaming from a trading floor to a dry sell-side researcher pedantically walking you through it” and “It’s no more difficult to write an irreverent story than it is to write a straightforward, AP-style story” [48]. It seems a small leap to imagine content generation in auditory, textual, or visual media algorithmically geared towards inspiring certain emotional, intellectual, or behavioral reactions as research continues. With automated algorithms working towards subtly influencing individuals and groups, wide-scale, in-depth manipulation is freed from labor requirements and thus conceivably becomes quite feasible to implement on a massive scale. V.

SOCIETAL OPTIMIZATION

A. Dynamic Optimization A vastly improved understanding of the dynamics by which societies operate would seem an excellent foundation from which to make adjustments to that operation. Social control through subtle adjustment of the environment has a long, well-documented history in architecture [53], urban planning [54], music [55], and media, but such methods have thus far been relatively inexact, dependent as they have been on pre-existing understandings of social and behavioral dynamics and are the former two fairly static in adapting to developing situations. To return to our previous illustration of John, the projected proto-dissident, for example, it might be just a few months or years before he blossoms into a revolutionary, so even if changing his local urban layout, musical scene, or mass media weren’t hideously expensive (barring automated generation for the latter two), it is unlikely to be feasible in time (again, barring automated generation.) Thus, having short and mid-term forecasting and dynamic automated adjustments could be something of a game changer .

B. Foresight-Driven Optimizations In addition to dynamically adapting to developing situations, these studies may increase insight into how to broadly increase desired outcomes in the long-term. This could allow for a firmer quantification of the return on investment of various social programs, as well as insights into the hidden costs and benefits of policies, potentially including suggesting entirely new policies. For example, John could be one of many in a general youth bulge that concerns the government due to the political instability often associated with having a large number of unemployed young men in one’s country. Perhaps some of the methods that can be applied in John’s case could be applied elsewhere to channel this demographic productively, the government might quantitatively isolate the factors that contribute to employment (or unemployment), and/or social unrest, and if possible alter the variables that contribute to those factors. Less hopefully, it might unfeasible to provide a high probability of employment to the majority of the young population, so the government might prioritize the employment of those most likely to be effective in a revolution, such as by creating fostering employment opportunities catering to particularly charismatic or passionate individuals, while crafting policies that reinforce submission in the rest of the populace. There is now no shortage of data for regimes to mine on the factors that contribute to civil unrest and the dynamics of successful and unsuccessful revolutions. Whether or not social media played a defining role in furthering the Arab Spring, social media certainly documented it in detail, mostly in a publicly accessible fashion [56]. C. Optimized for What? Some might find the prospect of using these technologies to pre-empt civil unrest concerning, as they potentially undermine the accountability of a government to its citizens. Instead of optimizing for civil stability (and/or the power/well-being of the ruling regime,) the society could be adjusted for other ends, but all of these ends imply one sort of choice or another. Encouraging traditionalism and diligence might increase productivity and a sense of unity, but what if by necessity it comes at the cost of creativity, innovation, and long-term economic growth? Fostering a belief in the value of families might cultivate a feeling of belonging and responsibility for the future of one’s children, but what if by necessity it also lays the seeds for cronyism and ethnic tensions? More clearly than ever before, we may see precisely how we could reach the utopias that each of us envisions—and the costs that would be required to achieve those utopias. The culture wars could become quite hot when the ground territory over which each side is fighting becomes thoroughly mapped out— at least until one group uses these techniques to eliminate the motive for the conflict altogether.

6|P age www.conference.thesai.org

Science and Information Conference 2015 July 28-30, 2015 | London, UK

Alternatively, the internal divisions driving the culture one way or another could be secondary to what each society must become when their global neighbors start to use these technologies. Maybe they will be forced to make adjustments to keep up economically. Or, perhaps they will need to alter themselves to defend against the new weapon of aggression that these technologies may permit. VI.

SOCIETAL HACKING

A. Two Bank Runs in Latvia On the weekend of December 11, 2011, unfounded rumors spread largely via Twitter incited a bank run in Latvia on two Swedish banks, despite repeated assurances by both banks and the government that the rumors were false [57] [58]. Both banks repeatedly refilled their ATMs, but large proportions of their machines were emptied entirely at some point during the weekend, with total cash withdrawals throughout the country at seven times higher than usual [57]. The Wall Street Journal quoted Latvian Prime Minister Valdis Dombrovskis’s spokesman as claiming that the rumor spreading was done “in an effort to destabilize the situation Latvia” [sic], paraphrasing Latvian officials as saying that the rumors were intentionally aimed at destabilizing the country and that a police investigation was underway [57]. As both banks were heavily capitalized elsewhere there was little chance of an actual collapse, but one speculates that if their business had been more heavily concentrated in Latvia the rumors of a collapse might have spiraled into a reality. Whether or not there was any truth the government’s assertions that the rumors had been intentionally spread as a campaign of economic destabilization, the incident illustrates how unaccountable individuals acting through social media can have an impact on the economic health of a country. B. War Continued by Other Means Like internal social control within countries, external social influence in foreign countries through subtle means has a long history, though examples from the Cold War may be among the clearest. Radio Free Europe and Radio Moscow both broadcast the desired viewpoints into the target regions, while the KGB covertly funded left-leaning political groups around the world [59] and the CIA secretly promoted American Abstract Expressionism internationally to demonstrate U.S. cultural vitality [60]. Though more recent examples would naturally be denied by their governments, several reputable news agencies have reported them as fact. The USAid-funded development of the social media network ZuneZuneo in Cuba was allegedly intended to covertly foment and organize unrest [61]. Meanwhile, the incredibly opaque St. Peterburg-based “Internet Research Agency” (whose Kremlin affiliations are widely alleged but unproven), not only spearheads sophisticated organized campaigns to anonymously disseminate pro-Kremlin, anti-Western en masse internationally across the internet through fake social media

accounts designed to look like people of other nationalities, but a New York Times reporter links them to anti-Obama rhetoric intended to look like it was coming from U.S. citizens as well as coordinated social media hoaxes of major disasters in the U.S. like a chemicals plant explosion and an Ebola outbreak, as well as the shooting of an unarmed black woman by police in the wake of the Ferguson protests [62]. Fine-grained models of social dynamics may multiply the breadth, effectiveness, and deniability of options in this arena. The foreign actor might not have the same depth of intelligence on the local population or as much freedom of movement as the ruling government, but this wouldn’t render attack impossible. Sufficient capabilities in forecasting and extrapolating data from other sources (perhaps based on initial data purchased through data brokers) could make up for the information imbalance. If the aggressor’s forecasting capabilities exceed that of the defender or if the defender does not monitor the social environment effectively, the attacks may be able to hide amongst the “white noise” of day-to-day interaction, such that the object of aggression never knows that it is being attacked, let alone it’s aggressor’s identity. To increase deniability, aggressive actions might best be handled at a distance, with the some activities performed through the relative anonymity of the Internet and local cells or social phenomena given a great deal of autonomy in order to reduce the aggressor’s footprint, though this would naturally come with trade-offs in fine control. The accuracy of forecasting models could offset this trade-off, allowing the aggressor to act with ever more indirection such that the link between its action setting up the causal chain and the end result is obscured; to return to the analogy, the line of dominoes that the aggressor sets off is so long that the defender can’t see how the aggressor’s knocking over the first one causes the final one to fall. Because of the inherent unpredictability of such systems, the aggressor would likely want to start or encourage multiple such causal chains (setting up multiple lines of dominoes,) but this must be weighed against the increased risk of discovery. The use of advanced autonomous algorithmic agents might also provide a layer of deniability, but could present the risk of complicating coordination. One possible objective for the aggressor might be political destabilization. Suppose, for example, that a foreign state actually wants to destabilize John’s country by encouraging civil unrest or even revolution. The foreign state might use the inverse of some of the techniques that his government intends to use to de-radicalize him. They would model various probable scenarios for destabilization, identify individuals like him as potentially valuable to a revolutionary movement, and then nudge him along the desired trajectory. Radicalization might be accelerated through introducing John to online social groups that mesh with John’s psychological profile and are projected to spur his rebellious drives. These social groups might then be used to influence John to cut himself off from moderating influences. With radicalization far advanced, the foreign state might introduce members of the group to information that would make them more effective revolutionaries, whether that

7|P age www.conference.thesai.org

Science and Information Conference 2015 July 28-30, 2015 | London, UK

be training in non-violent resistance, guerilla tactics, or contacts in weapons smuggling. Covert economic warfare might also be a possibility. Aside from encouraging bank runs to inflict general economic harm, states might target foreign industries that compete with major domestic industries or state-owned enterprises. Such attacks could include enflaming company public relations issues, influencing stock prices, causing runs on industry raw resources or necessary services to drive up prices, or any of a number of other angles for threatening their livelihoods. C. The Defender’s Woes There are a number of inherent difficulties in defending against this sort of aggression. Firstly, detecting it in the first place may require monitoring the environment in which it takes place that is intensive surveillance of interpersonal communication. Secondly, detection of the anomalies created by calculated influence may require models assessing what “normal” behavior is—precisely the technologies required for aggressive action. Finally, countering such activity may be difficult in countries that legally mandate freedom of speech, as the activities may or may not constitute a clear and present danger. This is a legal grey area that may need to be clarified. D. Not-so-Mutually Assured Destruction The Mutually Assured Destruction that has historically constrained the use of nuclear weapons and other weapons of mass destruction may not necessarily apply in this case, as MAD is contingent upon the attacked party being able to identify that it has been attacked and the identity of its attacker. If wide-scale cyber-espionage allegedly originating in China [63] is any indication, a sense of deniability would seem to increase the willingness to act aggressively when there is an apparent economic advantage to be gained. Considering the vast array of potential economic, diplomatic, and political advantages to be gained through the use of these technologies and their inherent deniability, they may see a great deal of covert use in the coming decades. VII. CONCLUSION Every organization on earth works to influence people. Autocratic governments strive to ensure the compliance of their citizens. Political parties work to maintain popular support, for-profit companies encourage customers to purchase their goods and services. Non-profit groups solicit donations of time and labor. Terrorists inspire fear, chaos, and respect in pursuit of their ideological ends. Every group with the slightest shred of power has a strong incentives to develop and implement the reliable means to forecast and influence behavior to better suit their ends. Thus, the drive towards these technologies is both universal and overwhelming. If these technologies are feasible, then they are inevitable.

Most of the research discussed in this paper is funded by companies, democratic governments, or academia. However, this reflects the relative transparency of these organizations and should in no way be taken as evidence that research in the field is dominated by them. Autocratic governments in particular have the resources, deep access to citizens’ personal data, and immunity to public scrutiny which give them the freest hand to develop and apply these technologies. Considering that these governments’ very survival depends upon their capability to manage dissent, they also have the strongest motivation of all to use them. Once these governments have developed these technologies for internal use, they may apply them to other societies, especially if the personal information of other countries’ citizens is easily acquired from third-party data brokers, either through purchase or theft. While widespread, malicious use of these techniques may be just as destructive to a country’s social and economic fabric as the use of a nuclear weapon, if executed effectively it may also leave no mushroom cloud—no sign that the devastation was caused by anything more than a “natural” social event. Thus, autocratic governments may have new leverage in reshaping global events in accordance with their interests. To detect and defend against this aggression, counties must have their own abilities to monitor and analyze their societies, so that anomalies indicative of calculated influence can be identified. Attempting to claim a moral high road of refusing to study these capabilities (or complicating their study) will only ensure that this new field is dominated by the less scrupulous, handing the reins of humanity’s future over to those with no qualms against oppression. Furthermore, even if it were possible to prevent the development of these technologies, it would mean forgoing the benefits that they might offer to society: identifying and correcting the root cause of social ills that have plagued humanity for millennia. In the best possible scenario, it could very well be the key to a world of unprecedented economic prosperity, social harmony, cultural richness, and scientific innovation. That said, these technologies are something of a Faustian bargain. By more efficiently controlling popular opinion and behavior they may undermine governments’ and political parties’ accountability to their constituents, companies’ accountability to their customers, and non-profits’ accountability to their stakeholders. Furthermore, the means to precisely, scientifically reengineer societies must be distributed with care. If the blueprints for this new society are written in coordinated fashion, those who currently hold power will presumably work to ensure that the changes favor their own interests. Not only could this fossilize the status quo, but it could consolidate social control in ways that threaten the balance of power. Whether democracy could survive such a consolidation is uncertain. On the other hand, if there is no coordination in the implementation of these techniques, it could cause highly

8|P age www.conference.thesai.org

Science and Information Conference 2015 July 28-30, 2015 | London, UK

unpredictable, perhaps fundamentally disruptive societal upsets, as thousands, if not millions of algorithms—each inhumanly effective at social manipulation—all work to covertly pull society into configurations that benefit their owners. Advanced automated social manipulation algorithms may exacerbate this issue even further, as the programs’ actions may be difficult to monitor and comprehend, such that even their nominal owners cannot effectively oversee them. These critical decisions must be considered in an open public dialogue, before the public dialogue itself is irrevocably bent into a weapon of oppression. VIII. REFERENCES [1]

[2]

[3]

[4]

[5]

[6]

[7]

[8]

[9]

Executive Office of the President (of the United States), "Big Data: Seizing Opportunities, Preserving Values," May 2014. [Online]. Available: https://www.whitehouse.gov/sites/default/files/docs/big _data_privacy_report_may_1_2014.pdf. M. Kosinski, D. Stillwell and T. Graepel, "Private traits and attributes are predictable from digital records of human behavior," Proceedings of the National Academy of Sciences of the United States of America, pp. 58025805, 11 March 2013. B. Bi, M. Shokouhi, M. Kosinski and T. Graepel, "Inferring the Demographics of Search Users," in 22nd International World Wide Web Conference, Association for Computing Machinery, 2013. A. Stopczynski, V. Sekara, P. Sapiezynski, A. Cuttone, M. M. Madsen, J. E. Larsen and S. Lehman, "Measuring Large-Scale Social Networks with High Resolution," Plos One, 2014. J. Golbeck, C. Robles and K. Turner, "Predicting personality with social media," in CHI EA Extended Abstracts on Human Factors in Computing Systems, Vancouver, 2011. M. Kosinski, Bachrach, P. Kohli, D. Stillwell and T. Graepel, "Manifestations of user personality in website choice and behaviour on online social networks," Machine Learning, 2013. J. Golbeck, C. Robles, M. Edmondson and K. Turner, "Predicting Personalit from Twitter," in IEEE Third International Conference on Social Computing, Boston, 2011. C. e. a. Sumner, "Predicting Dark Triad Personality Traits from Twitter usage and a linguistic analysis of Tweets," in Machine Learning and Applications (ICMLA), 2012 11th International Conference on, 2012. J. e. a. Golbeck, "Predicting Personality from Twitter," in 2011 IEEEnternational Conference on Privacy, Security, Risk, and Trust, and IEEE International Conference on Social Computing, 2011.

[10] D. e. a. Quercia, "Our Twitter Profiles, Our Selves: Predicting Personality with Twitter," in Proceedings of 2011 3rd International Conference on Social Computing, 2011. [11] F. Mairesse, M. A. Walker, M. R. Mehl and R. K. Moore, "Using Linguistic Cues for the Automatic Recognition of Personality in Conversation and Text," J. Artif. Intell. Res.(JAIR), no. 30, pp. 457-500, 2007. [12] V. Daukas, "Cognitive Computing from IBM Watson – Text Analytics for Natural Language...and Beyond," IBM, 23 October 2014. [Online]. Available: https://www.google.com/url?sa=t&rct=j&q=&esrc=s&s ource=web&cd=5&cad=rja&uact=8&ved=0CDgQFjA E&url=https%3A%2F%2Fwww950.ibm.com%2Fevents%2Fwwe%2Fgrp%2Fgrp004.n sf%2FvLookupPDFs%2FVince%2520Daukas%2520Pr esentation%2F%24file%2FVince%2520Daukas%2520 Presentatio. [13] D. Takahashi, "How IBM's Michelle Zhou figured out my personality from 200 tweets (interview)," Venture Beat, 11 October` 2013. [Online]. Available: http://venturebeat.com/2013/10/11/how-ibms-michellezhou-figured-out-my-personality-from-200-tweetsinterview/. [14] A. Sadilek, H. Kautz and J. P. Bigham, "Modeling the interplay of people's location, interactions, and social ties," in Proceedings of the Twenty-Third international joint conference on Artificial Intelligence, 2013. [15] P. Dudas, "Visualization for Dark Network Analysis," From the Core: Common Operational Research Environment Quarterly Newsletter, no. 3, pp. 1, 7, 2013. [16] S. Lucente, G. Wilson, R. Schroeder and G. Freeman, "Crossing the Red Line: Examining Social Media and Social Network Analysis for Unconventional Campaign Planning," From the Core: Common Operational Research Environment Quarterly Newsletter, no. 3, pp. 2-5, 2013. [17] S. Everton, G. Wilson, D. Cunningham, D. Zimmerman and C. Padilla, "The Internationalization of the FARC: Brokers, Key Players and Emergent Leaders," From the Core: Common Operational Research Environment Quarterly Newsletter, no. 3, p. 6, 2013. [18] O. Tsur and A. Rappoport, "What's in a hashtag?: content based prediction of the spread of ideas in microblogging communities," in Proceedings of the fifth ACM international conference on Web search and data mining , New York, NY, 2012. [19] J. Yang and S. Counts, "Predicting the Speed, Scale, and Range of Information Diffusion on Twitter.," in 4th International AAAI Conference on Weblogs and Social Media, 2010. [20] R. O. G. Gavilanes, D. Quercia and A. Jaimes, "Cultural Dimensions in Twitter: Time, Individualism

9|P age www.conference.thesai.org

Science and Information Conference 2015 July 28-30, 2015 | London, UK

[21]

[22]

[23]

[24]

[25]

[26]

[27]

[28]

[29]

[30]

[31]

[32]

[33]

and Power," in International Conference on Weblogs and Social Media, 2013. D. M. Romero, B. Meeder and J. Kleinberg, "Differences in the mechanics of information diffusion across topics: idioms, political hashtags, and complex contagion on twitter," in Proceedings of the 20th international conference on World wide web, New York, NY, 2011. L. Weng, F. Menczer and Y.-Y. Ahn, "Virality Prediction and Community Structure in Social Networks," Scientific Reports, vol. 3, 2013. J. Berger and K. L. Milkman, "What Makes Online Content Viral?," Journal of Marketing Research, vol. 49, no. 2, pp. 192-205, 2012. C. Danescu-Niculescu-Mizil, J. Cheng, J. Kleinberg and L. Lee, "You had me at hello: How phrasing affects memorability," arXiv, vol. 2, 2012. J. Leskovec, L. Backstrom and J. Kleinberg, "Memetracking and the dynamics of the news cycle," in Proceedings of the 15th ACM SIGKDD international conference on Knowledge discovery and data mining, 2009. M. G. Rodriguez, J. Leskovec and A. Krause, "Inferring networks of diffusion and influence," in Proceedings of the 16th ACM SIGKDD international conference on Knowledge discovery and data mining, New York, NY, 2010. C. Song, Z. Qu, N. Blumm and A.-L. Barabasi, "Limits of predictability in human mobility," Science, vol. 327, no. 5968, pp. 1018-1021, 2010. S. Goel and D. G. Goldstein, "Predicting Individual Behavior with Social Networks," Marketing Science, vol. 33, no. 1, pp. 82-93, 2013. N. Ramakrishnan and e. al., "'Beating the news' with EMBERS: Forecasting Civil Unrest using Open Source Indicators," arXiv, February 2014. P. Shakarian, J. Salmento, W. Pulleyblank and J. Bertetto, "Reducing gang violence through network influence based targeting of social programs," in 20th ACM SIGKDD international conference on Knowledge discovery and data mining , New York, NY, 2014. D. Castilho, P. O. Vaz de Melo, D. Quercia and F. Benevenuto, "Working with Friends: Unveiling Working Affinity Features from Facebook Data (Working Paper)," Association for the Advancement of Artificial Intelligence, 2014. [Online]. Available: http://www.aaai.org/ocs/index.php/ICWSM/ICWSM14/ paper/download/8084/8103. UN Global Pulse, "Unemployment Through the Lens of Social Media," 8 December 2011. [Online]. Available: http://www.unglobalpulse.org/projects/can-socialmedia-mining-add-depth-unemployment-statistics. S. Asur and B. A. Huberman, "Predicting the Future with Social Media," Proceedings of the 2010

[34]

[35]

[36]

[37]

[38]

[39]

[40]

[41]

[42] [43]

[44] [45] [46]

IEEE/WIC/ACM International Conference on Web Intelligence and Intelligent Agent Technology, vol. 1, pp. 492-499, 2010. I. Zheludev, R. Smith and T. Aste, "When Can Social Media Lead Financial Markets?," Scientific Reports, 27 February 2014. A. D. I. Kramer, J. E. Guillory and J. T. Hancock, "Experimental evidence of massive-scale emotional contagion through social networks," Proceedings of the National Academy of Sciences of the United States of America, vol. 111, no. 24, pp. 8788-8790, 2014. J. Shukla, "Seamless Prediction of Weather and Climate:," in Climate Test Bed Joint Seminar Series, Camp Springs, Maryland, 2009. United States National Oceanic and Atmospheric Administration, "Time Series of Monthly Mean Anamoly Corrections for 5-Day Forecasts," 2015. [Online]. Available: http://www.emc.ncep.noaa.gov/gmb/STATS/html/aczhi st.html. [Accessed 2015]. J. Lin, "Game Developers Conference Vault: The Science Behind Shaping Player Behavior in Online Games," 2013. [Online]. Available: http://gdcvault.com/play/1017940/The-Science-BehindShaping-Player. R. Epstein and R. E. Robertson, "Democracy at risk: Manipulating search rankings can shift voters’ preferences substantially without their awareness," in 25 Annual Meeting of the Association for Psychological Science, 2013. P. Sundsøy, J. Bjelland, A. M. Iqbal, A. S. Pentland and Y.-A. de Montjoye, "Big Data-Driven Marketing: How machine learning outperforms marketers’ gut-feeling," Social Computing, Behavioral-Cultural Modeling & Prediction Lecture Notes in Computer Science, vol. 8393, pp. 367-374, 2014. R. Wald, T. M. Khoshgoftaar, A. Napolitano and C. Sumner, "Predicting susceptibility to social bots on twitter," in IEEE 14th International Conference on Information Reuse and Integration, San Francisco, CA, 2013. T. Ryan and G. Mauch, "Getting In Bed with Robin Sage," in Black Hat Conference, 2010. D. Hofstadter, "Staring Emmy Straight in the Eye--And Doing My Best Not to Flinch," in Virtual Music: Computer Synthesis of Musical Style, Cambridge, MA, MIT Press, 2004, pp. 37-39. D. Cope, Comes the Fiery Night: 2,000 Haiku by Man and Machine, Santa Cruz, CA: CreateSpace, 2011. D. Cope, Ars Ingenero, Santa Cruz, CA: CreateSpace, 2012. L. R. Varshney and e. al., "A Big Data Approach to Computational Creativity," arX, vol. 1, no. 1213, 2013.

10 | P a g e www.conference.thesai.org

Science and Information Conference 2015 July 28-30, 2015 | London, UK

[47] The Economist Intelligence Unit, "Humans and Machines: The role of people in technology-driven organizations," The Economist, 2013. [Online]. Available: http://www.google.com/url?sa=t&rct=j&q=&esrc=s&so urce=web&cd=1&ved=0CCAQFjAA&url=http%3A%2 F%2Fspecials.nrc.nl%2Fricoh%2Fdocumens%2Ffiles% 2Fwhitepapers%2FWhitepaper%2520EIU_Humans%2 520%26%2520machines%2520PDF.pdf&ei=5pqnVM HaG8SdNouChOAI&usg=AFQjCNF1prWVHdC. [48] S. Levy, "Can an Algorithm Write a Better News Story than a Human Reporter?," Wired, 24 April 2012. [Online]. Available: http://www.wired.com/2012/04/can-an-algorithm-writea-better-news-story-than-a-human-reporter/. [49] Narrative Science, "Case Study: Dominion Dealer Solutions," Narrative Science, [Online]. Available: http://www.narrativescience.com/filebin/pdf/NS_Domi nion-Dealer-Solutions_CaseStudy.pdf. [50] Narrative Science, "Narrative Science Solutions," [Online]. Available: http://www.narrativescience.com/solutions. [51] Business Wire, "Narrative Science Announces New Customer Agreement and $10 Million Investment Round led by USAA," Business Wire, 28 November 2014. [Online]. Available: http://www.businesswire.com/news/home/2014112800 5028/en/Narrative-Science-Announces-CustomerAgreement-10-Million#.VKehwMnG5z8. [52] Business Wire, "Narrative Science Announces Strategic Investment and Technology Development Agreement with IQT," Business Wire, 5 June 2013. [Online]. Available: http://www.businesswire.com/news/home/2013060500 5246/en/Narrative-Science-Announces-StrategicInvestment-Technology-Development#.VKee08nG5z-. [53] D. Lockton, "Architecture, urbanism, design, and behavior: a brief review," Architecture, November 2011. [54] O. Yiftachel, "Planning and Social Control: Exploring the "Dark Side"," Journal of Planning Literature, vol. 12, no. 2, pp. 395-406, 1998. [55] S. Brown, U. Volgsten and eds., Music and manipulation: On the social uses and social control of music, Berghahn Books, 2005. [56] A. Bruns, T. Highfield and J. Burgess, "The Arab Spring and Social Media Audiences: English and Arabic Twitter Users and Their Networks," American Behavioral Scientist, vol. 57, no. 7, pp. 871-898, 2013. [57] G. Sandstrom, "Latvia Probes Rumors Behind Cash Withdrawals," The Wall Street Journal, 12 December 2011. [Online]. Available: http://www.wsj.com/news/articles/SB10001424052970 203430404577093872426329082.

[58] A. Tapinsh and S. Johnson, "Update 1-Latvians drain ATMs, banks refute rumors," Reuters, 11 December 2011. [Online]. Available: http://www.reuters.com/article/2011/12/11/swedbanklatvia-idUSL6E7NB0MI20111211. [59] V. J. Schodolski, "Moscow Sheds Light On Some Dark Secrets: Communist Records Opened-in Part," Chicago Tribune, 9 March 1992. [Online]. Available: http://articles.chicagotribune.com/1992-0309/news/9201220451_1_soviet-communist-party-partyarchive-state-secrets. [60] F. S. Saunders, "Modern art was CIA 'weapon'; Revealed: how the spy agency used unwitting artists such as Pollock and de Kooning in a cultural Cold War," The Independent, 22 October 1995. [Online]. Available: http://www.independent.co.uk/news/world/modern-artwas-cia-weapon-1578808.html. [61] Associated Press, "US secretly created 'Cuban Twitter' to stir unrest and undermine government," 3 April 2014. [Online]. Available: http://www.theguardian.com/world/2014/apr/03/uscuban-twitter-zunzuneo-stir-unrest. [62] A. Chen, "The Agency," 2 June 2015. [Online]. Available: http://www.nytimes.com/2015/06/07/magazine/theagency.html. [63] United States Department of Defense, "Annual Report to Congress: Military and Security Developments Involving the People’s Republic of China 2013," United States Department of Defense, 2013. [64] Business Wire, "NHS Choices Selects Narrative Science to Assist with Delivery of Healthcare Information," Business Wire, 28 October 2014. [Online]. Available: http://www.businesswire.com/news/home/2014102800 5223/en/NHS-Choices-Selects-Narrative-ScienceAssist-Delivery#.VKeiG8nG5z8. [65] J. Brill, "Privacy in the Age of Omniscience: Approaches in the United States and Europe – Address," in Mentor Group Vienna Forum, Vienna, 2014. [66] Pew Research Center, "Cell Phone and Smartphone Ownership Demographics," January 2014. [Online]. Available: http://www.pewinternet.org/datatrend/mobile/cell-phone-and-smartphone-ownershipdemographics/. [67] Pew Research Center, "Emerging Nations Embrace Internet, Mobile Technology," 13 February 2014. [Online]. Available: http://www.pewglobal.org/2014/02/13/emergingnations-embrace-internet-mobile-technology/.

11 | P a g e www.conference.thesai.org

Science and Information Conference 2015 July 28-30, 2015 | London, UK

[68] M. Komorowski, "A History of Storage Costs," 9 March 2014. [Online]. Available: http://www.mkomo.com/cost-per-gigabyte-update.

12 | P a g e www.conference.thesai.org