Social-IST

19/11/2013

Research Challenges in SCI

Social-IST/UEDIN//wp2/d2.1

Social-IST Social Collective Intelligence Grant Agreement Nº 317681

D2.1 White Paper on Research Challenges in Social Collective Intelligence WP2 – Research Challenges and Strategic Analysis Version: Social-IST/UEDIN/wp2/1.0 Due Date: 31/10/2013 Delivery Date: 19/11/2013 Nature: Report Dissemination Level: Public Lead partner: UEDIN Authors: David Robertson, Stuart Anderson, Iacopo Carreras, Daniele Miorandi Internal reviewer: Daniele Miorandi (CN)

www.social-ist.eu The coordination and support action has received funding from the European Union's Seventh Framework Programme [FP7/2007-2013] under grant agreement n° 317681

Public

Page 1 of 24

19/11/2013



Executive Summary This report first situates and outlines the potential of social computation to provide the basis for Social Collective Intelligence (SCI) in future systems. This involves the close interaction of social groups and machines together with systems of incentives and social structures to perform tasks that would otherwise be difficult to achieve either using entirely human or entirely machine solutions. The deliverable considers the challenges both from a technical and from a social science standpoint, identifying the potential for aligning them in order to provide an interdisciplinary perspective on the development of SCI systems. The paper then describes some of the challenges in developing an engineering approach to the development of such systems. Finally the paper outlines some of the “big questions” that arise from the framework for SCI research developed in the white paper.

Document History Version History Version

Status

Date

Author(s)

0.1

Draft

20/10

David Robertson

0.2

Draft

31/10

Stuart Anderson

0.3

Draft

4/11

Daniele Miorandi

0.9

Draft

7/11

Stuart Anderson

1.0

Final

19/11

Daniele Miorandi

Summary of Changes Version

Section(s)

Synopsis of Change

0.1

All

Stable draft available

0.2

Sec. 6

Added big questions

0.3

All

Revised text; fixed document template

0.9

All

Revised according to comments

1.0

All

Proofread

Public

Page 2 of 24

19/11/2013



Contents 1 Introduction ........................................................................................................................ 4 2 Social Collective Intelligence ............................................................................................ 4 The need for foundational work ......................................................................................... 5 Engineering Social Collective Intelligence Systems ......................................................... 6 Shifting to the Social Level................................................................................................ 7 3 Bottom-Up Technical Research Challenges: from conventional to social computation ... 8 Data deployment: Locally used  Socially used ............................................................. 9 Data ownership: Corporate ownership and control  Personal ownership and control... 9 Programming model: Centralized+compositional  Distributed+aggregative ............... 9 Connective structure: Computer network  Social network ......................................... 10 Human engagement: Local engagement  Social incentives ....................................... 10 Knowledge context: Set in advance  Emergent from social group............................. 10 Trust: Through direct experience  Through recommendation .................................... 10 Security: Privacy in closed systems  Protection in open systems............................... 11 System properties: Software test and verification  Social computation analysis ....... 11 System requirements: Pre-deployment  Post-deployment emergence ...................... 11 4 Bottom-up Social Research Challenges: social engineering of social computational systems ................................................................................................................................... 13 Provenance: Rule based  Socially constructed ............................................................ 13 Data Curation: Centralized & automated  Distributed social sensemaking ................ 14 Governance: Legally-based, national  Polycentric, Internet scale............................... 14 Cooperation: Workflow  Boundary objects ................................................................. 14 Structure: Massive & distributed  Fractal .................................................................... 15 Expertise and skill: Training  Coproduction................................................................ 15 IPR/Ownership: Legal protection  Open social mechanisms ...................................... 15 Responsible behaviour: Legal regulation  Emergent mechanisms .............................. 15 Impact: Local  Multiscale ............................................................................................ 16 Performance/Compliance: Fixed framework  Emerging or evolving framework ....... 16 5 Top-down Research Challenges: towards engineering social computational systems .... 17 Candidate sub-classes of social computation................................................................... 17 Existing mechanisms ....................................................................................................... 18 Ideas from the Social Sciences ........................................................................................ 19 6 “Big Questions” in Social Collective intelligence ........................................................... 20 7 Conclusion ....................................................................................................................... 22 8 References ........................................................................................................................ 23

Public

Page 3 of 24

19/11/2013

1



Introduction

Across the world societies are seeing rapid change in modes of social interaction and organisation. These new interaction modes are predicated on emerging forms of information infrastructure together with rapidly evolving devices, systems and applications that are ever more deeply interwoven with our social fabric. This white paper explores the phenomenon. Our exploration: 1. Provides an overview of the phenomenon we are concerned to study, and 2. Considers the challenges we must face in order to make progress in leveraging this phenomenon for social and commercial benefit. We consider “bottom-up” and “topdown” analyses from both a technical and social perspective. We focus on the research challenges inherent in characterising and providing the infrastructure and environment to gain significant social and commercial benefit from social collective intelligence. The companion deliverable Roadmap for FET Initiatives in Social Collective Intelligence (D3.1) includes an overview of application areas that provide strong motivation for the study of Social Collective Intelligence.

2

Social Collective Intelligence

The phrase “social collective intelligence” (SCI) is multi-faceted because there are multiple perspectives from which we can view the phenomenon of the increasing sophistication and depth of embedding of computer technology into society and the deployment of that technology to create new computationally mediated behaviours. Three key perspectives are: 





Public

Technological: Looks at the devices that are deeply embedded in our environment and whose use (at least in part) is for social purposes. Over the past 10 years we have seen vast increases in the power and ubiquity of such devices. Device ownership has far surpassed what would have been considered market saturation only a few years ago. Behavioural: Looks at the emergence of new forms of social behaviour that would be impossible without modern information technologies. This can be in very specific areas of business (e.g. the structure of modern markets that include many different forms of computational mediation to create the total market [1]); or in the creation of new forms of social actor [2] whose decision-taking capacity differs significantly from any individual actor in the system; or in much larger scale phenomena where large numbers of people give some computer-mediated activity broader social meaning; for example: the notion of “friend” in Facebook; the tagging activity in the ESP game [10]; or the more explicit response to incentives in the DARPA balloon challenge [3]. This last category covers a wide spectrum that includes situations that we can interpret as society acting computationally. Such computational groups of people (mediated through computer or other means) fit their behaviours into some broader algorithm, for example when “crowdsourcing” the answer to some query or when undertaking a coordinated collection of small tasks allocated by a “mechanical turk” to achieve a larger socially significant goal. Configurational: Looks at the social processes involved in programming and configuring technical devices together with establishing the social environment that will allow the emergence of new social behaviours that are directed to achieving some activity or situation. This will involve machines equipped with software (or the means to create such software) that allows them to be orchestrated/choreographed in (semi-) automatic ways to afford new social behaviours that could not be achieved without the use of information technologies. Concurrently with the technical task of orchestration, configuration or programming there will also be the creation of social environment that will allow the adoption of the technical innovations together with any social mechanisms

Page 4 of 24

19/11/2013



necessary to complement the technical innovation e.g. rules, regulations, laws, or whatever is necessary for people to understand how they are expected to behave together with incentives, sanctions, and other mechanisms to shape the expected behaviour [4]. Of course, this “configurational” activity could fit under the “behavioural” heading above but it is distinguished by its reflexive character. Society already provides each of these perspectives with a rich and rapidly developing picture. The emergence of new mechanisms like crowdsourcing, crowdfunding and games with a purpose all point to the potential for much more systematic and large-scale interventions that greatly increase the capacity of society to develop social computations that unleash the power of our “cognitive surplus”[5]. Clay Shirky has popularised the notion of the “cognitive surplus” of human reasoning that we currently devote to passive uptake of advertising or to playing simple games like solitaire. His argument is that suitable mechanisms could capture this surplus and put it to better use without the contributors feeling they are being imposed upon. At the same time the emergence of wide-scale cyber-bullying, identity theft and issues around trust in these new mechanisms of social cohesion point to the downside of our rapidly evolving human/computer synergy. Consideration of these issues also point to the need for innovation in these areas where the power of social computation points to the potential for new hybrid human/computer mechanisms to control these negative effects of the development of social/computational systems. We are aware that purely computational approaches to issues like privacy, data protection, trust and reputation pose intractable problems, particularly around the balance between safety and ease of use that could be alleviated by pursuit of social computation approaches to these issues. These considerations point to the broader issue of the sustainability of these Social Computational structures and the need for deep analysis of the synergy between the infrastructure and human social structures that will allow us to maintain confidence and trust in Social Collective Intelligence systems in the long term.

The need for foundational work

This rapid development of social/computational systems points to the need for a programme to explore the foundations of social computation. Already we are reaching the point where devices are sufficiently socially used and each device is itself “smart” enough to be socially choreographed that we are seeing the emergence of new and productive forms of computation from the ensemble of devices, orchestration and their social embedding. Developing appropriate technologies, orchestration and the engineering of the appropriate social environment at Internet scale is hard to achieve, however, because there are natural tensions between the sorts of open systems that one might wish to promote for large scale social interaction and the sorts of closed and tightly constrained systems that it is often more convenient to build. These tensions are often economic and social as much (or more) than technical - companies build constraining social computation systems for commercial reward as well as engineering expediency. This requires us to understand the most fundamental aspects of computation in a different way – we can no longer define computer algorithms in isolation from socio-economic assumptions and the conventions, sanctions and incentives that will allow them to operate comprehensibly and acceptably to civilised society. In all of this we recognize that comprehensibility and acceptability are malleable and will change because societies are not static structures but can be highly dynamic in their structure and behaviour [6]. The term “performativity” encompasses the idea that some mechanisms that are a relatively poor fit to human behaviour can have features that cause society to “perform” the mechanism and thereby draw social behaviour closer to that anticipated by the mechanisms. Of many examples, the Black-Scholes equations for the pricing of certain financial products are particularly striking [7]. Initially only a small part of market trading behaved in the way predicted by the equations

Public

Page 5 of 24

19/11/2013



but as actors in the market saw success coming to those who traded according to the equations more and more of the market began to conform to the behaviour predicted by the equations because traders began to use the equations outside their domain of validity. More generally, one can see the move to algorithmic trading together with many other IT-driven market developments can be seen as a conscious attempt to move towards the behaviour predicted by the “efficient market hypothesis”. This requires us to change our perspective on computer science because the elementary problems to be addressed are at the intersection of computation and human social experience. Our goal in SCI systems is to harness real, physical, social interaction as a key resource for computation. In order to achieve this we need to build computational mechanisms, shape and form collectives that can make effective contributions together with the incentives, sanctions, governance, regulation, training and, if necessary, law that link human and computational resources to achieve Social Collective Intelligence. Included in these considerations are how such mechanisms can encourage adoption and migration from older systems in order to achieve the necessary “momentum” towards adoption; how the system would adapt to exceptional events that are likely to occur in the dynamics of the social context; the consequences of a disorderly failure of the system; and how to support transfer to the succeeding generation of systems. In this context, we know that learning, and other social mechanisms have a powerful effect in terms of aligning collective behaviour around particular expectations of how the system will perform. Notions like performativity can be powerful tools in the developing notion of social collective intelligence. The importance of the Social Science analysis of complex socio-technical arrangements compels us to seek new design and engineering methodologies that take full account of the social in creating and adapting the computational, human and governance resources necessary to realise SCI. This requires a fundamental revision of conventional Software Engineering to consider the human collective and governance elements in this class of systems.

Engineering Social Collective Intelligence Systems We cannot build such systems by imposing a monolithic “solution” on society, yet we would also like to require the new systems we produce to have predictable behaviours that can be improved incrementally. However we may need to be more modest here. To some extent, we are entering into a new world where the predictability of systems will depend on our understanding of the predictability of the social dynamics of the system. We know from our experience of the legal system that legislators find it very difficult to predict the effect of new laws and interpretation in different social contexts can lead to divergent effects. Equally, predicting how users will make use of new software is difficult. So, a key challenge will be the extent to which we can guarantee the “safety” of such systems – in the sense that we should always be able to reach a safe state from any state of the system (e.g. that we can turn it off in some disciplined way that will result in an orderly adoption of alternative ways of working). For systems already in operation this is unclear, for example phenomena like the “flash crash” [8], where the New York Stock exchange lost 5% of its value in less than 15 minutes and valuable shares (e.g. Accenture) were briefly available to trade at 1 cent, suggest that the stability of this social/computational system in extreme conditions is at least questionable and reaching a stable state from any state in the system may not be possible. “Safety” is a key part of predictability but our ambition would be to develop much more sophisticated understanding of the dynamics of these social/computational systems. This is a key challenge in this area. One key to tackling this challenge is the development of social network analysis and, more ambitiously, the development of sociology to embrace the deep embedding of computation in modes of social interaction. This sociology of “computationally enhanced societies” will take on a new character, attempting to predict the effects of new modes of social computation in different contexts. Social computation systems engineering (SCSE) will require true interdisciplinary working that co-develops computational and social environments taking account of safety, stability,

Public

Page 6 of 24

19/11/2013



predictability together with the need for privacy, data protection, and an ethical approach to the development of systems. SCSE will draw on traditional systems engineering disciplines but will also build upon ideas from systems thinking; game theory to predict the effects of incentives and sanctions; organizational theory to understand learning and reflection in the creation of “organizational intelligence”; together with psychology, cognitive science and sociology to understand the role of the individual in group dynamics; and finally legal and political science to develop new regulatory processes and how systems should be engineered to take account of the need for regulation while maintaining openness and transparency. The development of these techniques will depend on deepening our understanding of the foundations of social computation as well as the development of tools and techniques to support engineering practice. Initially SCSE will study the development of the better-understood sub-categories of social computational systems. For example, the development of systems whose primary aim is to create and share aggregated data in order to amplify or make current activity more efficient or effective will start from a basic understanding of the raw material available from devices at individual level; will anticipate how these could form aggregate behaviours via interaction driven by social incentives to behave in particular ways and share data arising from that behaviour; and will harness these to form large-scale computations that are unpredictable but have predictable limits to that unpredictability because of the alignment achieved to human social structure and the design of appropriate governance regimes.

Shifting to the Social Level Even social computation is, at some level, still computation so we still have to worry about conventional computational properties when shifting to a social context. These properties, however, change their nature as we make the shift. Correctness of a program requires social measures: whether or not the algorithm gives the right result is determined by the aggregation of experience built into the algorithm itself (part of which resides in society). Completeness of a program requires social judgement: since social algorithms begin as incomplete problem statements and grow to cover more of the problem via interaction with the population, there is not necessarily any specific point at which we can insist that we have a complete specification of either the problem or the algorithm to solve it, particularly since the environment within which society operates is subject to change. However, it will be important to develop the means to measure the adequacy of a particular program and to ensure the observability of its effects both positive and negative. The scaleability of the algorithm implemented by a program requires social engagement. This will proceed in at least two dimensions: (a) the capacity of the algorithm to diversify into different social contexts while maintaining some capacity to communicate across contexts and (b) within a sufficiently constrained context there must be some selfreinforcing system of feedback such that the incentive for an individual to engage with the algorithm increases as more individuals participate. Social algorithms are still algorithms in the traditional sense but the art of building such algorithms lies in allowing society (in the physical world) to take much of the burden of dealing with complexity of problem solving. A starting point for development in the virtual world is the construction of lightweight but scaleable algorithms to pull data in from individuals; generate new information of higher utility to individuals based on the social interaction; and return the higher utility information to individuals in such a way as to reinforce their participation in the algorithm. The human issues involved in achieving this are challenging and broad, ranging from the individual level at which humans interact conveniently/transparently with “socially enabled” devices to the social level at which those involved in social computations are aware of (and enriched by) the social dimension of their interaction without being overwhelmed by it or by experiencing the consequences of the actions of deviant or malicious individuals or groups. However ambitious we are about social computation, progress towards it will be achieved primarily through extension of conventional systems, infrastructures and methods. This “bottomup” perspective has both social and technical facets and their interaction is essential for the development of SCI. This bottom-up work should also be shaped by top-down challenges for

Public

Page 7 of 24

19/11/2013



social computational systems. For example, in our discussion above we have outlined some of the issues around the predictability or safety of such systems. This is but one key property we have identified and others will emerge from the study of particular subclasses of systems and by considering the emerging literature on the sociology and cognitive science of social computational systems. These considerations motivate the structure of the remainder of this document where we consider: a) Bottom-up research challenges as directions of travel along key dimensions that extend outwards from conventional computing towards the development of social collective intelligence. This journey from conventional computation to social computational systems can be seen from either a technical or a social standpoint. We consider each perspective in the following two sections providing a diagram to summarise the challenges we have identified. Then, in each section we take the dimension as the starting point for a discussion of that challenge area. b) Top-down research challenges in terms of particular sub-classes of social computational systems that provide “training wheels” where we can refine the key properties we require of such systems in order to help direct bottom-up developments towards achieving effective social computational interventions. The final section of the white paper considers these challenges.

3

Bottom-Up Technical Research Challenges: from conventional to social computation

Each of the dimensions in the diagram above represents the idea of “lifting” a reasonably well-understood computational concept to the social level. This lifting process brings many

Public

Page 8 of 24

19/11/2013



conceptual and engineering challenges. In the sections below we briefly expand on the challenges of moving conventional computation along these axes. Data deployment: Locally used  Socially used The raw material of social computation is data acquired from the nodes of the computation network. The scaling effect of social computation is that data is not used locally but propagated through social groups. However, if we are successful in providing mechanisms for delivering data at scale into social networks then we need ways of harnessing this and feeding useful derived data back to the nodes. Some of this social analysis may be done automatically (e.g. in social signal processing) but many other forms of analytics may be developed (e.g. for streaming data from social network traffic that involve “social sensemaking”[9] where humans work together with algorithms accurately to combine data drawn from different contexts). This activity of structuring human effort to tag primary data was pioneered by systems such as the ESP game [10]. There is a huge potential for this technique to use much richer social interaction. This will involve social network analysis to understand the structure of the population coupled with determining the need for data and individual preferences in terms of visualization of the data. In addition there will be important representational issues in order to meet requirements for privacy and ethical acquisition and distribution of data. Data ownership: Corporate ownership and control  Personal ownership and control In many conventional social computation systems the data supplied by “clients” is held corporately by the owners of the system. This is a powerful business model for those corporations but it seals data within systems rather than allowing it to be deployed as those supplying it might personally wish. Personal ownership and control of data is desirable as a way of avoiding “lock in” of data to particular systems but this raises issues of maintaining ownership and provenance when sharing data. If personally held data is to be shared this raises considerable questions around governance, transparency and openness of organisations that gain access to personal data since control of non-tangible property is extremely difficult to control [11]. The exploration of effective means to provide individual control over data is a major challenge along both the technical and social dimension. This also raises important economic questions around what constitutes value in this context and how value is created and distributed in social computational systems. As people become more aware of the capacity to socially organize in order to produce value using computational mediation they will become less tolerant of the current settlement where some utility provided “for free” by an organisation is actually exchanged for personal data [12]. Programming model: Centralized+compositional  Distributed+aggregative Conventional program design is a centralized activity, with programs being built through the composition of modular program components to tackle the various elements of the programming task. Social computation, by contrast, requires systems for combining human effort that may be widely distributed and supplied opportunistically by the humans engaged in the computation. This is a much more aggregative style of programming, where the designers of complex programs must take into account the vagaries of collecting human effort across large distributed systems. This raises the issue of how to represent such programs, in particular, do we use models of the social components or are there other approaches [13]? How do we represent rules, incentives and sanctions? How adequate is the framework of game theory to represent incentives and sanctions? How do we take account of human aspects of rule following behaviour which is quite different from mechanical rule

Public

Page 9 of 24

19/11/2013



following? We might also consider the issue of the maintenance and evolution of systems within this context. Indeed, social setting evolves and hence we need to consider the stability of the system and how best to co-evolve the infrastructure and tools to maintain stability in an evolving context. Connective structure: Computer network  Social network Social computations need some form of connective structure to bind computational units together. In conventional systems this structure is supplied by networked computing infrastructure. This network infrastructure is also used by social computations. However, social computations also depend on social networks that overlay the computing infrastructure. This social layer needs to be developed. The social layer has quite different properties from conventional networks in terms of errors, recovery from failure and latency [14]. In particular we need to understand how social computational mechanisms can help build more open and transparent social networks, so to provide the capability of detecting systematically attempts to disrupt social communication. This will need the development of new social computational mechanisms to ensure reasonably reliable social communication. Human engagement: Local engagement  Social incentives In conventional computation human engagement is assumed but in social computation the purpose of part of the computation itself is to engage the right group of people to support the computation [15]. This means that incentives for interaction with the computation have to be built into each social program or there will need to be mechanisms that will allow the evolution of the system of rules, incentives and sanctions associated with the program. This is a particularly complex challenge since we will be dealing with heterogeneous notions of utility across different populations and this structure will also be an important element in avoiding competitors, malicious or deviant groups from disrupting the social program. Again the mechanisms for evolving rules, incentives and sanctions will be social programs themselves – ensuring these social programs result in acceptable structures of rules, incentives and sanctions are key to the success of the main social program. Knowledge context: Set in advance  Emergent from social group In conventional programming the context in which knowledge is shared between users of a program is assumed to be set in advance (and maintained by the program as it runs). Social computations are more open than this because the contexts in which different people engage with each computation cannot be assumed to be pre-determined. This raises the need for mechanisms that express key aspects of the social context as the computation spreads through a social group and allows prospective members of the group to be aware of and adhere to that emergent context [16]. We must assume that any social program will bridge many different contexts and thus the interpretation of many components of social programs, such as data visualizations, incentives, etc. will require to take account of heterogeneous contexts and the translation and adaptation across contexts. Trust: Through direct experience  Through recommendation Trust in conventional software is normally achieved through direct experience of use of that software or through the reputation of the company who produced it. This may also be true, in part, in social computation but in this case much more of the computation itself is done socially, with other people whom one may not know, so trust in the computation must somehow be derived through (personal or automated) recommendation of others within the social group, based on their commitments and behaviours. Of course, we already have recommendation agents in the form of Internet security providers like Norton who

Public

Page 10 of 24

19/11/2013



recommend on the basis of crowdsourcing experience of individuals. Nonetheless, in the social computation environment we face substantial new challenges because the network we utilize is a hybrid machine/human network and the programs we are using “run” on human as well as mechanical infrastructures, so understanding what it means to trust in this context is a new challenge. This challenge will involve new approaches to transparency and the development of the means to survey how social programs are evaluated on the human/machine infrastructure [17]. Security: Privacy in closed systems  Protection in open systems To be secure in a conventional system, an effective strategy is to close off as much as possible of one’s personal data from potential attackers [18]. The aim of social computations, however, is to engage people as problem solvers and data providers. To retain security while promoting this sort of openness requires trusted forms of protection for people in open environments, operating through social networks. A key part in developing protection mechanisms is to better understand the potential new threats in this environment. Social computation opens up a new vista of potential applications but simultaneously it opens up whole new categories of attack on social computations. Social computations are open so competitors will have the potential to disrupt the computation legitimately and illegitimately. That boundary between legitimate and illegitimate disruption is a complex legal matter and will require the development of new approaches to regulation of computational environments [19]. As we identify key areas of threat we can begin to develop defences to protect computations from disruption. System properties: Software test and verification  Social computation analysis We attempt to analyse, at various levels of detail, properties of conventional programs such as correctness or completeness. These are typically properties of the program plus its environment (normally the computer network) but social computations operate on social networks so we would expect analysis of conventional properties to have to take this into account. There may also be properties of social computations (e.g. those relating to incentives and gamification) that are core to social computations but peripheral to conventional programs. As the notions of computer network and social network converge we will see increasing need for social computation analysis as a core function that helps stratify social network populations to differentiate populations by concerns and capacities in order better to utilize their capacity and provide more personalized data and services while maintaining elements of social cohesion that are essential for the success of social programs. Works in Social Networks [20] and Complexity Science [21] have already made a good start on analysis tools. This work will challenge foundational notions of how we understand computation that will require a move to a more game theory-based modelling of computation so we have a framework for the analysis of competition and cooperation in social computation. System requirements: Pre-deployment  Post-deployment emergence In open social computations, where we may have little opportunity before attempting to run a computation to gain a deep understanding of how a program’s users will react to it, it is impossible to have a traditional phase of in-depth requirements analysis before running a computation. We must rely instead on emergence of acceptability and performance information derived post-deployment from social groups, developing methods for accumulating expertise based on this sort of emergence. We know from organizational science that there is always strong tension between the ostensive description of the operation of an organization and the performative behaviour of the organization in operation so, to

Public

Page 11 of 24

19/11/2013



some extent, we know there is a gap between formal requirements and the lived human experience of participating in an organization [22]. Social computations provide us with the impetus to revisit the whole notion of requirement since it provides us with the capacity to develop systems that can call on the lived experience of users to develop and change in the presence of problems in the social computational system. Providing effective means of managing these sorts of process will be a key challenge in animating the notion of social computation [23].

Public

Page 12 of 24

19/11/2013

4



Bottom-up Social Research Challenges: social engineering of social computational systems

This section, like the previous one, considers the challenges arising in moving from a conventional to a social computational world but here we consider this shift from a social perspective. Consequently, the dimensions differ somewhat from our technical dimensions. However there is considerable synergy between the technical and social dimensions and aligning the social and technical will be a highly productive area of work in social computation. As before, each of the short sections below develops the challenge associated with each dimension. Provenance: Rule based  Socially constructed Provenance information provides additional source and contextual information around data. Such information is critical to the correct interpretation of data. At the moment we see simple rule-based (but still quite ad hoc) approaches to managing provenance information [24]. Depending on how the data is being used, finding provenance information can be an open-ended, on-going task. This is probably best seen as a social computation where people embedded in the context can appropriately extend provenance information as necessary. Thus we see a move from machine-based provenance to social-machine-based provenance where people are involved in the construction, renewal and extension of provenance information. The design of such a social computation will need to take account of the potential for misleading or fraudulent provenance information being added. This might involve crowdsourcing or looking at agreement between people with different perspectives or Public

Page 13 of 24

19/11/2013



some other regulatory mechanism. The overall perspective is that provenance information is an evolving reflection of the evidence that allows people to take decisions about the use of data. As a consequence, provenance should support different communities in interpreting data in a way that supports their practice and so should avoid imposing any particular view on provenance information. Data Curation: Centralized & automated  Distributed social sensemaking Data curation involves the preservation and “cleaning” of data sets to better reflect the situation they pertain to. Currently this activity is often centralized around database curators who are located close to the data but can be remote from the context of origin of the data. By distributing the curation activity out to the context where the data is collected we set the sensemaking activity the context of origination and so the context is more accessible and comprehensible. In concrete examples like telemonitoring for health this means that the patient or immediate carers can curate the monitoring time series in order to incorporate information about the patient’s physical activity, diet, environment in an open-ended fashion. Telemonitoring provides the data infrastructure for telehealth applications. Although this may initially seem like a relative small and simple matter, a working approach to distributed social sensemaking would revolutionize professional practice in many fields. In health, such a system would provide the locus for discussion and debate around the significance of monitoring data that could help transform healthcare practice. Governance: Legally-based, national  Polycentric, Internet scale As we move to large-scale Social Collective Intelligence systems we inevitably encounter governance difficulties. At present, at Internet scale, we have a dichotomy between nationstate-based monopolistic governance and the almost complete absence of governance, depending on what sphere of activity we are considering. As Social Collective Intelligence systems become more complex and widespread we will see a move towards an understanding of governance systems that take account of people’s complex motivational structures, the interaction between different sorts of SCI structures and how they interact within an ecosystem of SCI systems. The key to understanding this situation is Ostrom’s work on polycentric governance in complex economic settings [25]. In particular, there is a pressing need for governance that acknowledges the need to manage disagreement within and between SCI systems. Cooperation: Workflow  Boundary objects A key research area for SCI is the design and deployment of mechanisms that facilitate cooperation. Current systems use orchestration mechanisms based on workflows as task descriptions. However, we know that workflows capture human action and interaction rather poorly and evolve slowly in response to a changing environment [26]. In addition workflows have evolved in an environment where disagreement is taken to be an exceptional situation and this is unlikely to be the case in SCI systems. A key part of the SCI programme will be considering mechanisms that allow cooperation, the development of the social network, and expertise in the network without requiring agreement between participants. A good example of a class of such mechanisms is the notion of “boundary object” introduced by Star and Griesemer [27]. Boundary objects are shared between two or more interacting collectives and they mediate that interaction, for example, a classification of nursing interventions surfaces complex professional nursing activity in a way administrators can understand. However, boundary objects like the classification are always provisional, and subject to revision to take account of changing practice. Freezing, standardising or institutionalising them will just

Public

Page 14 of 24

19/11/2013



result in additional boundary objects springing to provide the necessary slack between potentially antagonistic collectives. Providing support for the creation of diverse collections of these sorts of mechanisms will enable us to capture monitoring information on their use. Use data on these mechanisms will enable better support of different classes of users and will enable us to identify work that is poorly supported by the SCI system. Of course this is an open ended process because there will always be some gap between the work the system is intended to support and the demands of the lived situation. Structure: Massive & distributed  Fractal Our current systems are massive and distributed but have relatively simple and uniform structure. As SCI systems evolve we will see the emergence of more structures that are selfsimilar at different scales as working modes, governance and programming models shape similar responses to complex tasks at different scales. Thus we will see the emergence of fractal structure over SCI systems with multiple levels emerging at different scales. Mandelbrot in his book [28] on financial markets provides a much deeper analysis of the structure of financial markets than is currently in use by mathematical finance. We believe that as SCI systems develop there will be a need for deep analysis of the structure of SCI systems both to predict the consequences of design decisions and to help explore the behaviour of SCI systems. Expertise and skill: Training  Coproduction A key element in any SCI system will be the capacity to leverage experience and knowledge in the collective both to enable SCI systems to fulfil their role correctly and to develop the expertise within the collective so its capacity and reach are fully developed. This development activity necessarily will be achieved by a coproduction mechanism that will see mechanisms like buddying, multi-skill clustering, etc. to improve understanding of the goal the SCI is intended to achieve and to optimize the means of achieving it by developing all members of the SCI to achieve their potential contribution to the operation of the SCI. Thus the training aspect in moving to SCI style systems will shift from formal training to coproduced training. Coproduction of training depends on establishing common purpose among potentially diverse groups in order that sufficient resource is devoted to developing training activities. The opportunities to stimulate expertise growth may vary considerably between environments [29]. IPR/Ownership: Legal protection  Open social mechanisms As governance moves from no or single jurisdiction systems to large scale polycentric governance we will see a corresponding shift in the approach to IPR protection since current approaches to IPR protection are almost impossible to implement in a polycentric governance environment. An important part of this shift is the recognition that there is no single common motivational structure that underpins participation in an SCI system. We also acknowledge that conventional approaches to protection of IP in software do not extend naturally to SCI systems. In SCI, indeed, the novelty and creativity of the human contributors is an important part of the definition of the system but in particular where we have heterogeneous expertise, so that developing appropriate IP arrangements within the current frameworks will be difficult. Responsible behaviour: Legal regulation  Emergent mechanisms Recent experience of social-network-mediated rioting, community recovery, and the formation of e-vigilante groups is evidence of the emergence of a range of mechanisms to enforce, patrol, or encourage responsible behaviour. Increasingly we will see developments

Public

Page 15 of 24

19/11/2013



of multiple, overlapping, approaches to identifying what is seen as responsible behaviour and to ensuring that, at least locally, participants behave responsibly. In the medium term there will be some need to consider approaches to regulating bottom-up initiatives and controlling potentially excessive behaviour around responsibility and responsible behaviour within SCI systems. In addition we will have considerably enhanced monitoring capability and working out how best to regulate this will be an important part of ensuring responsible behaviour and use of SCI systems. Ultimately, the dynamism of social networks poses a huge challenge for legal approaches to regulation that will always lag behind current practice. This suggests that we should expect to see the emergence of bottom-up mechanisms of control as social collective intelligence systems evolve to take account of changes in the underlying technologies and the development of bodies of practice in the different communities participating the system. Impact: Local  Multiscale SCI systems are intended to be deeply embedded within and across multiple large-scale complex public, private and voluntary organisations. This embedded aspect has the consequence that understanding the impact of SCI systems is no longer a local concern but is multiscale where coordinated operation across multiple small-scale contexts can have strong effects in larger scale components of the system. A key element in evaluating SCI systems will be the development of appropriate metrics and traceability in order to account of these “long-distance” effects of SCI systems across multiple organisations with very different levels of organization and measurement capability. Performance/Compliance: Fixed framework  Emerging or evolving framework Mechanisms to help monitor and control the performance of SCI systems and the need to ensure compliance to a range of regulations and/or policies are key areas for research. Achieving predictable effects is an important aspect of any system even though the precise mechanism for achieving the effect may not be entirely clear. This aspect is important because some element of predictability is important for motivation and engagement in SCIrelated tasks. However, we are considering large-scale, fractal, complex systems so there is a challenge to develop approaches to monitoring and compliance that take account of the emergent characteristics of Social Collective Intelligence Systems while also capturing some elements of predictability. In particular, one critical aspect of performance and compliance measurement and management is that the framework should adapt as the use and capability of the SCI system evolves. A good example of this is evident in High Frequency Trading (HFT) markets: it is becoming increasingly evident that leading a market move on any particular asset is an important feature because leading a change gives the leader scope for manoeuvre as the share movement takes place. Thus HFT systems increasingly take initiatives that lead to “flash crash” move in individual shares and these are going on all the time (10-20 per day). This is a new, formerly unheard of phenomenon but it is important to understand its emergence and importance. Static performance monitoring frameworks would not recognize the importance of evolving the monitoring framework to capture such new phenomena but capturing this sort of phenomenon is critical to ensure the threats associated with them can be assessed and mitigated if necessary. Thus we see that the performance and compliance of SCI systems will require incorporate hybrid human/machine monitoring frameworks that will adapt to the emergence of new phenomena using humans capacity to recognise new phenomena in open-ended contexts combined with the means rapidly to automatically classify and capture such phenomena.

Public

Page 16 of 24

19/11/2013

5

Top-down Research computational systems


Challenges:

towards


engineering

social

In this section we provide three top-down challenge areas that we believe could play an important role in shaping the development of the field of social collective intelligence. These challenge areas will help shape the bottom-up developments outlined in the previous section and will help identify how the bottom-up challenges interact and contribute to the development of deployed social computation systems. Each section provides some examples of the challenge and finishes up with a small number of high-level questions: 1. The first raises the issue of the existence of useful sub-classes of social computation that arise naturally and, on the face of it, appear to be less complex than the full notion. 2. The second stems from the observation that there are many “naturally occurring” social computation systems already in operation and we can challenge our conceptions of social computation using empirical means by identifying characteristic mechanisms in use in some of these social computational systems. 3. The third challenge identifies some key attributes of social computational systems and attempts to identify work in the social science literature that bears on these attributes. Candidate sub-classes of social computation One important challenge is to identify sub-classes of social computations that are in some sense “simpler” than the general notion. These should be linked to real applications but should be applicable across a range of situations. We have three (or four) sub-classes at the moment but we envisage this list could be extended both by considering particular domains of application but also using features such as architecture or scale to help control complexity: a) Computer-mediated social sensemaking of socially generated data: this involves developing social computational systems that allow people to contextualise, correct and interpret data gathered by sensors in the environment. This class is inspired by the problem of contextualising telehealth data gathered by patient in their own home. At the moment most systems generate many false alarms because users, their families and surrounding social context have no way of adding to the raw data and those interpreting the data are both remote and cautious. More generally social interpretation of data seems like a key component in many systems and an important simplified sub-class of social computation. b) Organisational routines [30,31,32] are notoriously difficult to capture because they involve very large numbers of exceptional cases and necessitate drawing on experience of past situations to help decide the best course of action in new situations but at the same time the situation is constrained since we are working with an identifiable data set and what we aim to deliver is fairly well understood. This provides a good context to consider how social computation can share experience and promote social learning in a relatively constrained and small-scale environment. c) Markets have already been very heavily studied and benefit from a single, financial, notion of value. This is an important sub-class with many well-developed social mechanisms in place. For example, hedge funds, are a good example of Social Collective Intelligence where the social group includes a range of skills, including the capacity to develop new trading algorithms and means of monitoring and visualising market dynamics. Hedge funds typically exist in a complex human/machine symbiosis with market-relevant information shaping the strategy and execution carried out using high frequency trading. Studies of hedge funds argue that decision taking in a hedge fund is markedly different from decision taking by individuals or

Public

Page 17 of 24

19/11/2013



groups that do not live in an information-rich environment [2]. Overall, markets comprise many of these new composite social/machine actors competing around maximising profit in a highly monitored risky environment. This context provides a good environment for studying the limits of a single notion of value and in exploring approaches to regulation and its circumvention. d) One could also consider a hybrid of (a) and (b) above to consider “transparent organisations” where data about the behaviour of the organisation is gathered continuously and made available to people working in the organisation as a means of helping develop novel lines of communication and action. There has been earlier work on transparent finance in organisations that we could use as a starting point [33]. These categories suggest the following questions: • What features of social computational systems does each sub-class highlight or eliminate from consideration? • Are there other such sub-classes that are interesting and might help us investigate social computation? • Are there refinements of the sub-classes that might be more useful in studying social computation? Existing mechanisms This challenge proposes a systematic study of some of the existing social coordination mechanisms such as: recommendation, trading, dating, gaming, friending etc. We should explore how they are deployed at the moment and what social structures they are capable of supporting. One particularly interesting area is to study how aggregated data is imbued with social meaning. For example: a) Friending or linking: how does this process differ between say LinkedIn and Facebook? In Facebook, what is the significance of your number of friends and how was that constructed? Does this construction imbue unfriending with increased significance? How do these mechanisms fit within our framework of social computation and can we account for the social element in the way these mechanisms operate? b) Recommendation: this works well in some contexts and less well in others. As we become more sophisticated users of recommendation systems they are becoming more complex and include features such as stratified populations and multidimensional recommendation measures. Can we characterise situations in which recommendation is likely to be effective? Can we also identify components and means of composition that can help in the design process of recommendation systems intended to support particular work practices? c) H-index: In academic circles h-index is increasing seen as an important measure of academic influence or significance. How has this been constructed? How have issues around the interpretation of h-index been dealt with? How do different communities interpret h-index? In the longer term will there be an attempt to create other competing impact measures and how will any discrepancies be reconciled or magnified, by which social groupings? This is a good example of performative notions where the social milieu “performs” the definition and it fills its intended social role (and may also have some unintended side-effects.) This is a deliberately short list of mechanisms intended to illustrate the idea. The ideas are apparently simple but they have a character we see in many social computation settings. Human mechanisms need to be conceptually graspable and apparently very simple

Public

Page 18 of 24

19/11/2013



mechanisms have the capacity to support complex behaviour. One important area will be the development of mechanisms that are easily socially graspable and are useful in a range of settings. This idea of humanly comprehensible or graspable mechanisms is interesting because some social mechanisms that do exist are very useful but remain difficult for people to comprehend. One example in this area is e-voting systems where the mechanisms have good properties but remain underused because people find them opaque and difficult to trust. In developing this area further we will consider the following questions: • Can we extend this list of mechanisms that are deployed in some social computational system? • Can we characterise contexts in which the deployment of these mechanisms is likely to be effective? Ideas from the Social Sciences We should also explore some notions from the social sciences that may give us some grip on barriers of promoters of the social computation concept. A short initial list would include the following: a) Governance: There is a substantial body of work in the social sciences on governance but of particular interest in Social Computation is the work of Elinor Ostrom [34] who has considered approaches to the governance of common pool resources or “commons”. b) Decision Making: There is a substantial body of work in social science on decision taking that takes account of limited resources and access to information. This will include game theory from an economic perspective, work like Giegerenzer’s [35,36] on the use of heuristic decision taking. In addition the work of Tversky and Kahneman’s [37] on choice and risk is potentially interesting. c) Trust: Trust is a key factor in the operation of social computations but there is a persistent tendency to attempt to eliminate trust in favour of some evidence-based measure of risk. The social science literature, in particular Mollering [38], avoids this shift and attempt an analysis of trust. This is important because it is a key contributor to smoothly operating human systems. d) Risk: Ullrich Beck’s [39] work on Risk Society is a key text in contributing to an understanding of the ethical foundations of large-scale systems and the role of science and technology in social transformation. e) Coordination: Star and Bowker [40] have identified boundary objects as elements in working contexts that allow groups to cooperate without requiring consensus. Boundary objects are ubiquitous in complex organisational settings and taking account of Star and Bowker’s work will be critical to the success of social computation. f) Evolution: Bowker, Edwards [41] and others have developed a framework to consider complex human/machine infrastructures that support complex social cooperation. For example, organised and citizen science is one area of study. Paul Edward’s study of the climate science infrastructure (“A Vast Machine”) points out many complex operations that are required for long-lived infrastructures that may not initially seem to be a requirement of an information infrastructure. In considering the social in Social Collective Intelligence we will need to take full account of work from the social sciences. This suggests we need to answer two related questions: 1. What other social aspects of Social Collective Intelligence do we need to take into account in developing SCI further?

Public

Page 19 of 24

19/11/2013



2. Can we refine our interest in particular topics (e.g. those listed above) to ask sharper questions of the social science literature?

6

“Big Questions” in Social Collective intelligence

Social Collective Intelligence has the potential to transform most aspects of human life if it is widely adopted. In particular it offers the potential for a radical transformation in the way information is gathered, integrated and used to support human activity. This document outlines some of the key dimensions of Social Collective Intelligence from an interdisciplinary perspective. In assessing the suitability of Social Collective Intelligence as a focus for further work we can reframe this work as a series of “big questions” that such a programme might take as important issues to resolve. Of course there is a balance to be struck between what we anticipate can be achieved by such a programme and setting ambitious “stretch goals”. Thus our big questions are not at the level of generality of a socalled “Hilbert question” (e.g. Is mathematics decidable?) but they are set to provide considerable challenge both at the level of individual disciplines and in interdisciplinarity where there is a need for insight that require multiple disciplinary perspectives. 1. How do we specify and verify Social Collective Intelligence systems? Borrowing from conventional verification perspectives we will be interested in both safety properties and liveness properties of SCI systems. Safety properties demonstrate the absence of particular behaviour, for example we might be interested to demonstrate an SCI system is incapable of identifying information about individual participants or to demonstrate that particular kinds of instability are absent (e.g. that some individual or group are persistently stressed beyond their capacity). Liveness properties might include demonstrating that an SCI system has the capacity to respond resiliently to some class of events or that the system admits the capacity for particular types of evolution depending on the context of use. Characterising such properties effectively is challenging and developing a verification framework where we can illustrate properties of a system to some level of confidence is also challenging and novel. 2. What is the data model for Social Collective Intelligence? There is considerable range of possibilities for data models. For example, we might want to have data that explicitly represents some measure over a population and particular sub populations based on characteristics of that measure. We might also want this to be represented as a time series to capture the dynamics of the measure over time. The general issue is how best to represent captured behavioural data and how to interpret that captured data to provide information on social action. 3. What is a programming model for Social Collective Intelligence? We know we need to specify human and computational resources combined by a structure of governance and incentives (or disincentives) associated with particular forms of action. The issue is the design of appropriate structuring mechanisms for the description of these resources and rules and how to test a system under development to observe the likely behaviour of the system in use. Since we envisage that at least some aspects of SCI system will develop after the system has been deployed we need to consider a programming model that includes how to describe the initial state of the system, how it evolves during its lifetime and how to bring things to an orderly close. We might also consider what a highly distributed “socialized” development environment might look like. The programming model would need to develop alongside the data model.

Public

Page 20 of 24

19/11/2013



An essential aspect of such models is the need to account for governance via policylike elements that take account of human rule-following characteristics. 4. What is the process of creating Social Collective Intelligence systems that are “fit for purpose”? This is the “engineering” process of coordinating a range of resources to bring about a change in the way social groups interact. This will involve identifying the needs of a complex, possibly highly structured, stakeholders group and working to meet needs effectively. This process will involve considering how to structure the human resources to gain the required effect. This may involve detailed simulations of social action and the negotiation of changes in governance and regulatory structure to enable changes in social action. 5. What is the network model for SCI? We will need to consider softer, more flexible ways of modelling networks that include a considerable element of entirely social networking. This will introduce different types of interaction into the graph that take account of human aspects of rule following, transgressions and the open-world structure of human experience. Any networking model will need to take account of locality, polycentrism and the need for the capacity to take coordinated action without requiring agreement between cooperating parties. 6. How do we manage key dimensions of Scale, Space and Time in Social Collective Systems? The stability of many governance systems depends on locality and limitation of scale arising from limits on locality combined with loss of information over time. Modern SCI breaks all of these limiting mechanisms. This has many immediate consequences that we can see arising today. For example, we are beginning to see the rise of global-scale online labour markets like elance that have the potential dramatically to transform the nature of work [42]. The emergence of such structures will require the development of stabilizing mechanisms that balance power relationships and permit more reliable prediction of the dynamics of these systems in new conditions. This understanding will pave the way for more effective governance of SCI systems that offer agility and flexibility in the presence of rapidly changing circumstances. 7. What is the model that supports the access and interpretation of data? Our goal will be to build on the developing approaches to Web data access and management mechanisms but to include much more of the social dimension in access and interpretation. We envisage the social curation of datasets and the development of a range of mechanisms that allow social sensemaking to play an important role in interpreting data. We will also consider the identification and development of expertise amongst users and the pivotal role played by locality in determining variation in the interpretation of data and understanding of context. 8. What is the impact of SCI on the nature of work? We have already touched on the capacity of SCI to transform the way labour and tasks are connected in the emerging new e-labour markets. More generally we envisage that the emergence of SCI as a key element in modern economies will see the development of a whole range of new employment that engages in supporting SCI operation and development. 9. What are good models to initiate SCI systems? Often SCI systems will only work effectively if a significant proportion of the population participate in using the SCI

Public

Page 21 of 24

19/11/2013



system. We need to consider generic mechanisms to launch SCI systems in a way that is likely to achieve levels of adoption that allow the system to function as intended. 10. SCI may transform what we see as intellectual work, how do we provide good governance of IPR in the context of new forms of intellectual production? Since SCI will transform intellectual work it is highly likely to undermine traditional views of Intellectual Property rights. Understanding better what we want to know about ownership of ideas will allow us to design systems that are adequately instrumented to allow the allocation of property rights in an appropriate manner. 11. Is it possible to regulate via SCI mechanisms? One possible role for SCI systems is the development of architectures that provide more agile and responsive regulatory regimes than is currently the case. Such a regulation controller would allow us to consider more dynamic approaches to regulation with very rapid decision-making. 12. We will want to place monitoring and transparency requirements on SCI, how is this to be achieved? We will need to provide the means to monitor SCI systems effectively to see that the system is behaving as we anticipate and to learn the dynamics of the system when it encounters radically different environments.

7

Conclusion

This white paper points to the fascinating interdisciplinary challenge in developing the field of Social Collective Intelligence. The companion document on the application of Social Collective Intelligence points to the considerable benefits that will arise from the development of SCI systems. In Europe we have a developing community of researchers with the capacity and interdisciplinary working style that will enable rapid advance in this important new field.

Public

Page 22 of 24

19/11/2013

8



References

[1] Donald MacKenzie. An engine, not a camera: How financial models shape markets. The MIT Press, London, 2008. [2] Iain Hardie, Donald MacKenzie. “Assembling an Economic Actor: The Agencement of a Hedge Fund,” Sociological Review 55/1 (2007). [3]John C. Tang, Manuel Cebrian, Nicklaus A. Giacobe, Hyun-Woo Kim, Taemie Kim, Douglas "Beaker" Wicker, Reflecting on the DARPA Red Balloon Challenge. Communications of the ACM, 2011. [4] A. Ferscha, N. Davies, A. Schmidt, N. Streitz, “Pervasive Socio-Technical Fabric”, Procedia Computer Science, Volume 7, 2011, Pages 88-91, ISSN 1877-0509, 10.1016/j.procs.2011.12.027. [5] Clay Shirky. Cognitive Surplus: Creativity and Generosity in a Connected Age, Penguin, 2010. [6] Donald MacKenzie, Fabian Muniesa & Lucia Siu, Do Economists Make Markets?On the Performativity of Economics, Princeton University Press, 2010. [7] Black, Fischer; Scholes, Myron. The Pricing of Options and Corporate Liabilities. Journal of Political Economy 81 (3): 637–654 [8] Menkveld, Albert J. and Yueshen, Bart Z., Anatomy of the Flash Crash (April 10, 2013). Available at SSRN: http://ssrn.com/abstract=2243520 or http://dx.doi.org/10.2139/ssrn.2243520. [9] Buchanan, M. (2007) “The Science of Subtle Signals”. Strategy+Business, issue 48 [10] von Ahn, L. (2006) “Games with a Purpose”. IEEE Computer Magazine, 39(6) [11] Weitzner, D., Abelson, H., Berners-Lee, T., Feigenbaum, E., Hendler, J. & Sussman, G (2008) “Information Accountability”. Communications of the ACM, 51(6) [12] Pentland, A. (2009) “Reality Mining of Mobile Communications: Towards a New Deal on Data”. World Economic Forum Global IT Report 2008-2009 [13] Zhang, H., Horovitz, E. & Parkes, D. (2013) “Automated Workflow Synthesis” 27th AAAI Conference on Artificial Intelligence [14] Easley, D. & Kleinberg, J. (2010) “Networks, Crowds, and Markets: Reasoning About a Highly Connected World”. Cambridge University Press [15] Bernstein, M., Brandt, J., Miller, R. & Karger, D. (2011) “Crowds in Two Seconds: Enabling Realtime Crowd-Powered Interfaces”. ACM Symposium on User Interface Software Technology [16] For example, d'Inverno, M., Luck, M., Noriega, P., Rodríguez-Aguilar, J. & Sierra, C. (2012)"Communicating Open Systems". Artificial Intelligence, vol. 186 [17] A collection of recent approaches appears in Goldbeck, J. (ed) (2009) “Computing with Social Trust”. Springer Verlag Human Computer Interaction series [18] Halpern, J. & O’Neill, K. (2008) “Secrecy in Multi-Agent Systems”. ACM Transactions on Information and System Security, 12(1) [19] Harkins, M. (2013) “Managing Risk and Information Security: Protect to Enable”, Apress. [20] Lazer, D., Pentland, A., Adamic, L., Aral, S., Barabási, A., Brewer, D., Christakis, N., Contractor, N., Fowler, J., Gutmann, M., Jebara, A., King, G., Macy, M., Roy, D. & Van Alstyne, M. (2009) “Computational Social Science”. Science 323 [21] Flack, J. & Krakauer, D. (2011) “Challenges for Complexity Measures: A Perspective from Social Dynamics and Collective Social Computation”. Chaos: An Interdisciplinary Journal of Nonlinear Science, American Institute of Physics press, 21 Public

Page 23 of 24

19/11/2013



[22] Proceedings of 1st International Workshop on Requirements Engineering for Social Computing 2011 [23] Harel, D (2008) “Can Programming be Liberated, Period”. IEEE Computer 41:1 [24] Golbeck, J. (2006) “Combining Provenance with Trust in Social Networks for Semantic Web Content Filtering”. International Provenance and AnnotationWorkshop, Springer Verlag Lecture Notes in Computer Science 4145 [25]Ostrom, Elinor, Beyond Markets and States: Polycentric Governance of Complex Economic Systems, Nobel Prize in Economics documents 2009-4, Nobel Prize Committee, 2009. [26] Cain C, Haque S. Organizational Workflow and Its Impact on Work Quality. In: Hughes RG, editor. Patient Safety and Quality: An Evidence-Based Handbook for Nurses. Rockville (MD): Agency for Healthcare Research and Quality (US); 2008 Apr. Chapter 31. Available from: http://www.ncbi.nlm.nih.gov/books/NBK2638/ [27] Star, Susan Leigh; Griesemer, James (1989). "Institutional Ecology, 'Translations' and Boundary Objects: Amateurs and Professionals in Berkeley's Museum of Vertebrate Zoology, 1907-39". Social Studies of Science 19 (3): 387–420. doi:10.1177/030631289019003001. [28] Benoit B. Mandelbrot, Richard L. Hudson, The (Mis)Behaviour of Markets: A Fractal View of Risk, Ruin and Reward, Profile Books 2010. [29] Kearns, M. (2012) “Experiments in social Computation”. Communications of the ACM 55. [30] Pentland, B. T. and Feldman, M. S.. (2008). Designing Routines: On the folly of designing artifacts, while hoping for patterns of action. Information and Organization. 18: 4. 235-250. [31] Pentland, B. T. and Feldman, M. S.. (2007). Narrative Networks: Patterns of technology and organization. Organization Science. 18: 5. 781-795. [32] Pentland, B. T. and Feldman, M. S.. (2005). Organizational routines as a unit of analysis. Industrial and Corporate Change. [33] Wehmeier, Stefan, and Oliver Raaz. "Transparency matters: The concept of organizational transparency in the academic discourse." Public Relations Inquiry 1.3 (2012): 337-366. [34] Ostrom, Elinor, Polycentric Systems as One Approach for Solving Collective-Action Problems (2008). Available at SSRN: http://ssrn.com/abstract=1304697 or http://dx.doi.org/10.2139/ssrn.1304697 [35] Todd, P. M., & Gigerenzer, G., & the ABC Research Group. (2012). Ecological Rationality: Intelligence in the world. New York: Oxford University Press. [36] Gigerenzer, G., Hertwig, R., & Pachur, T. (Eds.). (2011). Heuristics: The foundations of adaptive behavior. New York: Oxford University Press. [37] Kahneman, D., & Tversky, A. (Eds.). (2000). Choices, values and frames. New York: Cambridge University Press. [38] Möllering, G. 2006. Trust: Reason, Routine, Reflexivity. Oxford: Elsevier. [39] Beck, Ulrich (1992) Risk Society: Towards a New Modernity. London: Sage [40] Bowker, G. C.; & Star, S. L. (1999). Sorting Things Out: Classification and Its Consequences, Cambridge, MA: MIT Press. ISBN 978-0-262-02461-7 [41] Paul N. Edwards, (2010) A Vast Machine: Computer Models, Climate Data and The Politics of Global Warming. Cambridge, MA: MIT Press. [42] Horton, John J. Online labor markets. Springer Berlin Heidelberg, 2010.

Public

Page 24 of 24