See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/315134740
Strategic Opportunities (and Challenges) of Algorithmic Decision-Making: A
Call for Action on the Long-Term Societal Effects of ‘Datification’
Article in SSRN Electronic Journal · January 2015 DOI: 10.2139/ssrn.2644093
Copyright By PowCoder代写 加微信 powcoder
2 authors:
186 PUBLICATIONS 6,527 CITATIONS SEE PROFILE
Some of the authors of this publication are also working on these related projects:
big data View project
Information Systems Development View project
77 PUBLICATIONS 493 CITATIONS SEE PROFILE
All content following this page was uploaded by on 18 November 2018. The user has requested enhancement of the downloaded file.
Journal of Strategic Information Systems xxx (2015) xxx–xxx
Contents lists available at ScienceDirect
Journal of Strategic Information Systems journal homepage: www.elsevier.com/locate/jsis
Strategic opportunities (and challenges) of algorithmic decision-making: A call for action on the long-term societal effects of ‘datification’
a School of Business, Management and Economics, University of Sussex, Brighton BN1 9RH, UK
b IPM Department, , Waltham, MA 02452, USA
article info
Article history:
Accepted 6 February 2015 Available online xxxx
Algorithmic decision-making Digital traces
Social and ethical issues
Introduction
Today, digital data are captured through a variety of devices that have the ability to monitor the minutiae of an individual’s everyday life. These data are often processed by algorithms, which support (or drive) decisions (termed ‘algorithmic decision-making’ in this article). While the strategic value of these data (and subsequent analysis) for business- es is unquestionable, the implications for individuals and wider society are less clear. Therefore, in this Viewpoint article we aim to shed light on the tension between businesses – that increasingly profile customers and personalize products and services – and individuals, who, as McAfee and Brynjolfsson (2012, p. 5) suggest, are ‘walking data generators’ but are often unaware of how the data they produce are being used, and by whom and with what consequences. Issues associated with privacy, control and dependence arise, suggesting that social and ethical concerns related to the way business is strategically exploiting digitized technologies that increasingly support our everyday activities should be brought to the fore and thoughtfully discussed. In this article we aim to lay a foundation for this discussion in the IS community and beyond.
Ó 2015 Elsevier B.V. All rights reserved.
The last decade has witnessed the widespread diffusion of digitized devices that have the ability to monitor the minutiae of our everyday lives (Hedman et al., 2013). Nolan (2012, p. 91) argues that ‘Global IT has enabled information on most everything to flow most everywhere at stealth speed’. The data trail we leave is increasingly used by companies to manage employees and target and personalize products and services for clients and customers, based on developing algorithms that can make predictions about individuals by recognizing complex patterns in huge data sets compiled from multiple sources. In this article we consider some of the observed and potential consequences of this new type of data-driven, algorithmic decision-making, illustrating that while it can offer strategic opportunities for business and sometimes benefits for indi- viduals, there are also costs, hence raising societal issues: as Galliers et al. (2012) indicate, there can be a difference between how business is benefiting and how society is benefiting – or otherwise.
The IS literature has already raised social and ethical concerns associated with IT (Smith, 2002; Smith and Hasnas, 1999), and in particular those concerns are often associated with privacy issues (e.g., see Belanger and Crossler, 2011; Chan et al., 2005; Coll, 2014; Greenaway and Chan, 2005). However, few IS studies have linked these concerns with the digitization of
⇑ Corresponding author.
E-mail addresses: (S. Newell), (M. Marabelli).
http://dx.doi.org/10.1016/j.jsis.2015.02.001
0963-8687/Ó 2015 Elsevier B.V. All rights reserved.
Please cite this article in press as: Newell, S., Marabelli, M. Strategic opportunities (and challenges) of algorithmic decision-making: A call for action on the long-term societal effects of ‘datification’. J. Strateg. Inform. Syst. (2015), http://dx.doi.org/10.1016/j.jsis.2015.02.001
2 S. Newell, M. Marabelli / Journal of Strategic Information Systems xxx (2015) xxx–xxx
our everyday life (exceptions include Abbas et al., 2014; Boyd and Crawford, 2014; Lyon, 2014; Slade and Prinsloo, 2013), and fewer still have discussed this phenomenon in relation to algorithmic decision-making (one exception being Schroeder and Cowls, 2014). Here, we focus on the consequences of ‘algorithmic decision-making’, which occurs when data are collected through digitized devices carried by individuals such as smartphones and technologies with inbuilt sensors – and subsequently processed by algorithms, which are then used to make (data-driven) decisions. That is, decisions are based on relationships identified in the data, and the decision maker often ignores why such relationships may be present (Mayer-Schonberger and Cukier, 2013). While these data-driven decisions made by businesses lead to personalized offerings to individuals, they also result in the narrowing of their choices (Newell and Marabelli, 2014).
Given the above, we argue that algorithmic decision-making has societal consequences that may not always be positive and, in this Viewpoint article, we aim to articulate such concerns. In so doing, we bring to the fore the issues related to algo- rithmic decision-making and highlight the interdisciplinary nature of this topic (Chen et al., 2012; Smith et al., 2011). As we have indicated, some work has been done to shed light on the social implications of the widespread diffusion of digital devices in the IS community, but also in other disciplines such as sociology – as in the work of Lyon (2001, 2003, 2014), Doyle et al. (2013), and Ball (2002, 2005) on impacts of monitoring and surveillance on society, and of Castells et al. (2009) and Campbell and Park (2008) on societal changes determined by the diffusion of digital devices. Here, we call for IS research that examines (and challenges) corporations (and governments) in terms of the strategic decisions that are being made based on data that we are now constantly providing them (see also MacCrory et al., 2014), whether we realize it or not. Next, we define some key concepts and set the boundaries of our analysis.
Big data, little data, and algorithmic decision-making
Data-driven or ‘algorithmic’ decision-making is based on collecting and analyzing large quantities of data that are then used to make strategic decisions. Algorithmic decision-making incorporates two main characteristics: firstly, decision- makers rely on information provided by algorithms that process huge amounts of data (often big data, as we will explain next); secondly, the reasons behind the ‘suggestions’ made by the algorithms are often ignored by decision-makers (Mayer-Schonberger and Cukier, 2013). We expand on both characteristics below.
Digitized technologies and data analytics
Data that originate from digitized devices are increasingly permeating our everyday lives. These digitized devices have the ability to keep track of and record what we do. As a result, somebody else may eventually be able to use the data thus produced – often with purposes different from those originally intended. Thus, we focus on ‘digital traces’ – all data provided by individuals (1) during ‘IT-related’ activities, captured from social networks, online shopping, blogs, but also ATM with- drawals, and other activities that will leave a ‘trace’ (Hedman et al. 2013; Wu and Brynjolfsson, 2009) and (2) that are cap- tured through technologies that we use that have in-built sensors. These technologies include LBS (Location Based Technologies) that are IT artifacts equipped with GPS systems and so have the ability to collect a user’s location such as a smartphone with GPS – see Abbas et al. (2014) and Michael and Michael (2011) for social implications – and other surveil- lance and monitoring devices – see the previously cited work of Lyon (2001, 2003, 2014) for privacy implications.
It is clear that the huge amount of digital trace data that are collected through the many digitized devices that we now use to support our daily activities fall into the ‘big data’ umbrella. The big data (analytics) concept is very similar to the more familiar (and less sexy) business intelligence that has been studied for the past decade or so (e.g., Negash, 2004; Power, 2002; Rouibah and Ould-ali, 2002; Thomsen, 2003). McAfee and Brynjolfsson (2012). Following Gartner’s (2001) definition, it is the three Vs of big data1 on which we focus: Volume (the amount of data determines value); Variety (data arise from dif- ferent sources/databases and are cross-matched to find relationships), and Velocity (data are generated quickly). Big data encompasses much more than this individually generated data trail (see Chen et al., 2012 for a broad discussion of big data analytics) but here we focus just on this everyday digital trail that we each leave. That is, we focus on those big data that are generated by individuals during their everyday lives (and are captured as digital traces). In other words, we focus on data that arise as a consequence of each of us now being a ‘walking data generator’ (McAfee and Brynjolfsson, 2012, p. 5). This attention to the digitization of our everyday life allows us to narrow the focus of our inquiry and to expand on concerns regarding the use (and abuse) of one aspect of big data analytics that concerns algorithm-driven decision-making and associated personalization – to which we now turn.
Algorithmic decision-making
(Big) data captured through digitized devices are processed by algorithms aimed at predicting what a person will do, think and like on the basis of their current (or past) behaviors. These algorithms can predict particular outcomes, as with
1 The definition of big data was updated by Gartner in 2012 as they now describe the concept as ‘high volume, high velocity, and/or high variety information assets that require new forms of processing to enable enhanced decision making, insight discovery and process optimization (Gartner, 2012). Moreover, others have added ‘new Vs’ – e.g., veracity, variability, visualization, and value, viewing big data in terms of 5 or even 7 Vs. Here, where we stick with the original definition (Gartner, 2001) as this reflects the essence of big data for the purposes of this article.
Please cite this article in press as: Newell, S., Marabelli, M. Strategic opportunities (and challenges) of algorithmic decision-making: A call for action on the long-term societal effects of ‘datification’. J. Strateg. Inform. Syst. (2015), http://dx.doi.org/10.1016/j.jsis.2015.02.001
S. Newell, M. Marabelli / Journal of Strategic Information Systems xxx (2015) xxx–xxx 3
the numbers of ‘friends’ on Facebook being used to predict a person’s credit risk (http://www.google.com/patents/ US8560436) or an individual’s Facebook ‘likes’ on a college Facebook page, used to predict her/his willingness to become a donator (http://www.nytimes.com/2015/01/25/technology/your-college-may-be-banking-on-your-facebook-likes.html?_ r=0). Interestingly, these predictions often represent a black-box: while humans must decide what to measure and produce the algorithms to analyze the data being collected, these decisions do not necessarily involve understanding the causes and consequences of particular patterns of behavior that are identified (Mayer-Schonberger and Cukier, 2013). Rather, it is deemed sufficient that connections are discovered. Traditionally, making decisions has been a human-centered, knowl- edge-based activity with humans discriminating on the basis of an understanding of theory or context (Tsoukas and Vladimirou, 2001). By contrast, algorithmic decision-making means that discriminations are increasingly being made by an algorithm, with few individuals actually understanding what is included in the algorithm or even why. In other words, it is seen as being sufficient that an algorithm is successfully predictive, never mind if the reasons for the associations found in the data from different sources are unknown. We argue that this is likely to create problems when no one in a corporation really understands why some decisions are made. For example, one could argue that the last financial crisis was at least par- tially a product of this problem, with the algorithms that predicted the pricing for mortgage-backed securities clearly not taking into account all the risks while at the same time not being subject to question because the basis of the algorithm was neither clear nor easily accessible, either to the senior managers in the financial institutions where the algorithms were being used or to the credit rating agencies who were evaluating these products (Clark and Newell, 2013).
In sum, here we focus on data collected through digitized devices that we increasingly use to support our everyday activ- ities. This is ‘big data’, because the three (or more) Vs of Gartner’s (2001, 2012) definition apply. In fact, data coming from digitized technologies are high in volume because of the widespread diffusion of digital devices that allow access to social networks at any time, as well as all other types of technologies that record what we do even if we do not ‘own’ them (e.g., surveillance cameras, or an ATM card machine, where the usage information goes into our bank’s database). Thus, data come from different sources (variety). For instance, data used for making ‘algorithmic decisions’ may come from a combination of contributions on social networks and LBS systems (e.g., a ‘check in’), or spending capacity of consumers associated with per- sonal facts of individuals (e.g., the partner’s birthday). Data velocity is clearly another characteristic of the digitization of our everyday life, because we are ‘walking data generators’ 24/7 and ‘More data cross the Internet every second than were stored in the entire Internet just 20 years ago’ (McAfee and Brynjolfsson, 2012, p. 4). On this point, it is worth noting that most of the digitized devices that collect such individual level activity data fall under the Internet of Things (IoT) umbrella (Miorandi et al., 2012; Xi et al., 2012). However, we do not restrict our analysis to those digitized devices that are connected to the Internet because some devices remain (for now) independent of the Internet (e.g., some OBD devices). One such example is provided by Progressive Insurance in the USA (http://www.progressive.com), which provides a memory stick that is plugged into a car’s on-board computer and the data must be uploaded to the insurance company rather than automatically sent via the Internet.
Potential discriminations associated with the (ab)use of algorithmic decision-making: big and little data
The use of algorithmic decision-making associated with data coming from the digitization of our everyday lives improves the capacity of a business to make discriminations. Thus, businesses have always discriminated in terms of to whom they offer products and services, because products and services are targeted to different audiences (we cannot, unfortunately all afford to buy a Bentley car). With algorithmic decision-making they are simply taking this a step further. For example, they can now much more precisely target and personalize offerings to customers and potential customers – those predicted to buy particular products or services. As a more specific example, a car’s computer that monitors speed, usage of brakes, horn, lights, etc. (such as Progressive Insurance’s OnStar OBD technologies mentioned above) has the ability capture all these details that are then sent to data centers. Computers then analyze the (big) data and insurance companies are able to use the results to discriminate (e.g., by charging young men higher premiums because the data indicate that they – generally – drive less safely than other categories of drivers). Such data-driven decision-making has been questioned because it can go against the ethical principle of equal or fair treatment. This is exemplified in the recent case in the EU, where insurers are required to no longer use statistical evidence about gender differences to set premiums. Thus, despite the fact that gender differences are clear from the data (e.g., young male drivers are ten times more likely to be killed or injured than those – of both sexes – over the age of 35; women live, on average, longer than men), it is considered to be discriminatory (following an EU ruling that came into effect in December 2012) to use this trend evidence to differentiate between premiums (e.g., car insurance or actu- arial rates) for men and women. The point about this change in the law is that it was considered to be discriminatory because, for example, while young men in general may drive more recklessly and so be more prone to accidents, an indi- vidual young man may not and would therefore be discriminated against when insurers set premiums based on group trends observable in collective data.
While using big data and algorithmic decision-making to observe trends and so discriminate between groups of indi- viduals can have social consequences that are potentially unfair, this targeting can now be taken further when data are used not to predict group trends but to predict the behavior of a specific individual. This is sometimes described as ‘little’ data – although it should be noted that little data are actually based on big data but are simply used in a more targeted way. Thus, little data focuses on the everyday minutiae of specific individuals, using computing capacity to collect extremely granular data (Munford, 2014). Drawing on the previous example of a car’s OBD, little data can now allow us to concentrate on a
Please cite this article in press as: Newell, S., Marabelli, M. Strategic opportunities (and challenges) of algorithmic decision-making: A call for action on the long-term societal effects of ‘datification’. J. Strateg. Inform. Syst. (2015), http://dx.doi.org/10.1016/j.jsis.2015.02.001
4 S. Newell, M. Marabelli / Journal of Strategic Information Systems xxx (2015) xxx–xxx
specific driver, and we can decide whether an individual is a good or bad driver based on the sensor data from his/her car. Sensors have the ability to capture individual’s behaviors and are widespread. As an illustration, consider that approximately 85% of handsets now have a GPS system chipset installed (Abbas et al., 2014). By using sensor data, the insurer would not be setting premiums based on the general trends in accident rates between groups, but instead would base their calculations on the actual driving habits of an individual. However, if little data are more ‘objective’ in terms of discriminations made by corporations, it probably poses more issues for societies given the observed or potential social consequences; for instance, in terms of an individual’s privacy (Lyon, 2014) or in terms of the exploitation of the vulnerable – an issue that IS scholars seem not to have fully addressed as yet.
It is then clear that algorithmic decision-making poses two main concerns in terms of big and little data: first, (in terms of big data) this data trail provides the opportunity for organiza
程序代写 CS代考 加微信: powcoder QQ: 1823890830 Email: powcoder@163.com