Projects per year
A large amount of human interaction is prosecuted along the rambunctious pathways of social networks, producing vibrant and chaotic streams of communication. Utterances are channelled along ever more complex pathways to an increasingly fuzzily defined audience of humans and machines. These receivers are eager to find meaning and structure, to coordinate and support social endeavours. However, extracting interactional structures from these outpourings is a complex task, as we deal with overlapping conversations, between many actors, spread across multiple networks. A well developed method of coordinating activity exists in the form of Electronic Institutions (EI). These institutions provide an architecture which allows agents to carry out complex patterns of interaction, based on shared protocols, while assuming little knowledge about their compatriots, and providing guarantees about the outcomes of interactions. While EIs are a powerful tool for coordinating computational agents, they are less widely used to support human activity. A key reason for this is what one must give up in order to join an EI: one must first understand the language used to define the protocols, and then commit to carrying out interactions through the machinery of the EI, sacrificing control and leaving the openness of the interconnected online world. These barriers to entry have meant that traditional EIs have not become relevant to the vast surge of data, or the potential for interaction centred around socially driven systems such as Twitter. Here, we connect the power of EIs to describe and formalise interaction with the open social systems which are currently supporting such a wide range of human interaction. This allows for the modelling of behaviour, to extract patterns of interest from data for summarisation and exploration. It also provides a framework by which computational intelligence can be harnessed in support of informal human interaction. The cost of making this connection is the creation of an additional layer which binds freeform interaction streams into appropriate hooks and levers within EIs, matching loose social discourse with crisp institutional structures. While the general case matching utterances to formal semantics is extremely difficult, the presence of an interaction protocol allows us to concentrate on only the possible actions which would make sense for the institution in its current state, reducing the range of possibilities and simplifying the task of translation such that simple approaches give sufficient discriminatory power. This is not appropriate for every situation, but for well chosen models of interaction, a few matching rules can be enough to create a formal skeleton alongside free discourse. In this paper, we describe how this translation can be carried out, starting from an account of how institutions can still have power when they are no longer the gatekeepers of action. We detail the formal machinery necessary, and describe our implementation using a process calculus, with examples.
- Edinburgh Neuroscience
- College of Science and Engineering - Vice-Principal & Head of College of Science & Engineering
- Artificial Intelligence and its Applications Institute
- Data Science and Artificial Intelligence
Person: Academic: Research Active