AbstractCan we develop a generic socio-technical computing device that lets emergent human collectives determine the computational program by their real-time inputs? In this conceptual article we present the system architecture of a novel approach to facilitate socio-technical computation beyond what the state-of-the-art in human-agent collectives and social machines considered so far. The system responds to bursts of activity around a topic by spinning up tasks made up from the observed content and by collecting instructions on those tasks from the general public on the Web. It is designed in an open fashion (open standards, open access) to embrace the social in computing as in the theories of Max Weber, which opens a variety of new challenges and future research directions as laid-out in this article. Finally, we introduce a new metaphor for what we regard one of the current grand challenges for the Computer Science discipline resulting from this kind of work.IntroductionAn increasing amount of voices in the research community states that “what goes viral” is heavily impacted by the commercial use of social media, and that the modeling of virality based on retrospective observation of successful campaigns does not bring us any nearer to predict what will go viral in the future \citep{Cebrian_2016}. One can observe that the affordances of today’s social media systems and research in this context are largely focused on sharing of information on one or at most a set of proprietarily linked platforms. Tools or environments to empower human collectives to actively shape their action and to form procedural knowledge around an actual event or topic in real-time are largely missing to date; apart from very few examples like IFTTT (https://ifttt.com), that are mainly focused on linking the personal information environment of the individual and lack openness to be suited for emergent collective action. Consequently, the ultimate vision of autonomously operating human-agent collectives \citep{Jennings_2014} or emergent social machines  \citep*{Hendler_2010}, incorporating the general public on the Web, cannot be seen in practice yet.We argue that this is largely because research and development in the space of social media is currently in a retrospective trap. Our views to the interplay of the technical and the social on the Web remain highly descriptive and the constructive dimension is limited. To fill this gap, we demonstrated an early stage prototype of a system that adds the formation of procedural knowledge to any social media system it is connected with \citep{Luczak_Roesch_2016}. In this article, we give detailed account to the principled architecture of the next iteration of this system, which reacts upon activity bursts and lets human participants perform low-level actions on content that they regard meaning- and purposeful in the context of a real-world event. The human input that is captured by this Social Computer forms the formal program running on it, while the technical backend simply facilitates that information can flow across platform boundaries to reach further human participants. Being fully based on the principles of the Web architecture, the system allows open access to the procedural knowledge that is created by the input from the crowd that engages with the system via one of its many instruction interfaces. This also enables the development of custom views to the computer’s state and tailored instruction interfaces. In the remainder of this this article we will give an overview of the current state of the literature on emergent socio-technical systems on the Web, such as Social Machines as well as human-agent collectives. Then we present our principled architecture of a Social Computer and describe use cases to illustrate how an instantiation of it works. In the end we discuss a number of research challenges arising from the rigorous openness of this novel computing system to human input. We also introduce a metaphor for what we regard one of the grand challenges for Computer Science in the socio-technical age.Emergent Socio-technical systems on the WebThe Theory and Practice of Social MachinesWhile first mentioned around 2000 by \citet*{m2000},  a more formal account to Social Machines has not been given until recently when various researchers have begun to investigate the entire spectrum of what has been abstractly promised as a novel computing paradigm. The work on this project can be roughly divided into three main work areas: 1) observing socio-technical systems to understand the interplay as well as micro and macro effects of human and machine coexistence; 2) devising novel technologies for decentralised social Web applications; 3) mapping out social, moral and ethical issues as well as principles of the World Wide Web today and in the prospected future. The work presented in this article is heavily related to the observational work on Social Machines, which so far has been either large scale and quantitative or very small scale qualitative work and can be seen as the foundation for the few theories about Social Machines that have been established to date. The first theory that came out of this puts individual systems such as Twitter, facebook, reddit, Zooniverse or Mechanical Turk at the centre of the consideration \citep*{Smart_2014}. By classifying the socio-technical properties of those systems (e.g. incentive mechanisms, information sharing capabilities or generally system goals) the approach seeks to enable system developers to imitate and adapt particular patterns of those systems in order to build new Web-based participatory systems most successfully. An alternative to this system oriented viewpoint is the work on narrative structures about purposeful collective processes that can range across the boundaries of individual systems \citep{Tarte_2015,Murray_Rust_2015}. Focusing on communities and the evolution of sociality within those communities, this work has leveraged archetypes as the fundamental theory of Social Machines. These two qualitative and small scale approaches are complemented by the information-centric view to Social Machines \citep{Luczak_Roesch_2015,Luczak_Roesch_2015a,Luczak_Roesch_2018}. In contrast to the classification work, but in-line with ideas of archetypal narratives, this approach assumes Social Machines being the emergent output of human activity rather than any engineered input. As retrospective approaches these three individual lines complement each other well to allow for a multiperspective classification of socio-technical processes on the Web. The quantitative approach can be used to sample relevant subsequences of user interactions to further investigate those qualitatively to give detailed account to narrative structures.Our approach presented in this article embeds the information-centric approach to Social Machines to facilitate system-agnostic detection of activity bursts as well as content filtering. However, we fundamentally change the focus of our theoretical consideration of Social Machines from the retrospective viewpoint to the constructive anticipation, planning and execution of purposeful collective action. Agents, Interactions and Social ProtocolsHuman-agent collectives are coming from a multi agent systems (MAS) angle to tackle the challenge of coordinating collective action in an open and decentralised environment. This angle heavily emphasises the role of economic principles in autonomous systems as well as dedicated interaction protocols that govern “how the agents’ actions translate into an outcome, the range of actions available to the participants, and whether the interactions occur over steps or are one-shot” \citep{Dash_2003}. Most recent work in this area widens the economic view to incentivisation slightly, to account for the diversity of motivations for different people in different situations \citep{Jennings_2014}. However, HACs still focus on solving fixed tasks with dedicated goals that are managed in dedicated applications (e.g. citizen science platforms or digital disaster response services). With our work, instead, we seek to let even the task design and goal setting arise from human activity only and to allow for composing multiple systems to contribute to the problem solving, an approach that has also been taken by other work that comes from a similar angle but still relies on predefined interaction models, social protocols or executable specifications \citep{Ahmad_2013,Giunchiglia2010,f2013,Murray_Rust_2014,Chopra_2016}.We seek to further expand this idea of task emergence and reduce even the interaction protocols down to a set of most fundamental atomic instructions that allow the formation of arbitrary process flows involving the interfacing systems and the reached human participants. This upgrades the role of flexible low-level interaction as opposed to fixed sets of algorithmic rules, a principle that has already been the foundation of our modern interactive computing \citep*{Wegner_1997} but now seems to get lost when we build on agents with fixed interaction protocols of a high level of abstraction. With the Social Computer human participants shall ultimately get the facility to formulate their own interaction protocols composed of sequences of atomic instructions \citep{Luczak_Roesch_2015b}.Collective Intelligence, Human Computation and CrowdsourcingOur approach differs from the typically coordinated approach in collective intelligence, human computation and crowdsourcing \citep{Malone_2009,Woolley_2010,Quinn_2011,Kittur_2013}. Research in these areas commonly calls for methods to engineer the way a human collective is going to perform a pre-defined task \citep{Minder_2012,Minder_2012a} and relies on dedicated crowdsourcing platforms \citep{r2015}. We, instead, want to expose the intelligence that lies in accumulated activities of human users on the Web, while minimising the presuppositions about the tasks to be performed as well as the communities or systems in which they take part.Such “loosely knit coordinated actions” \citep{Lee_2015} have recently been highlighted as an area of increasing importance for research on computer supported collaborative work (CSCW) as well as the role of activity sequences \citep{Keegan_2016}. We contribute to this line of research by introducing a system to capture and support emergent coordinated action that is also freely available for adaptation and further development by other researchers.Principled Architecture of a Social ComputerWe are now going to present the generic system architecture underlying the A1, our first prototype of a Social Computer as depicted in Figure 1.