Streamlining Social Decision Making for Improved Internet Standards

Many decisions in today's world are made through a complex, dynamic process of interaction and communication between people and teams with different interests and priorities - so called "distributed decision-making" (DDM). For example, many businesses work across multiple geographically dispersed offices and timezones, with teams specialising in quite diverse areas. Each team may have its own goals and reward models, which do not necessarily coincide, and may be spread across multiple organisational units (e.g. different businesses or governments). Communication may happen via several different modalities with very different timescales and properties (e.g. email, instant messenger, and face-to-face meetings).

Unfortunately, although many organisations have started to document these processes and even make records available (particularly governmental organisations e.g. https://data.gov.uk/), we have no way to automatically analyse these records. If we did, we could produce tools to automatically summarise decisions, trace who made them, and why and how they were made (and why other decisions weren't made). From a societal standpoint this would help make these processes more accountable and transparent. We'd also be able to identify collaborative failures, biases and other problems, and thus help improve decision-making in future.

This project will develop these urgently required methods, using a combination of natural language processing and social network analysis. We will collate, annotate and publicly release the first multimodal dataset of real-world distributed decision-making. We will devise techniques to take natural language and semi-structured data to recognise the dialogue and interaction structures in decision making, and analyse those structures to produce summaries and evaluate the efficacy of the decision making process. We will then use the outputs to inform strategic interventions that can streamline and improve decision making.

Our methods will be suitably generic to span several domains. However, the project will focus on one particular global organisation as its main use case: the Internet Engineering Task Force (IETF). This is an international forum responsible for producing Internet protocol standards - formal documents which specify the languages by which software and hardware "speak" across the Internet. To produce these documents, extensive international collaboration is performed - this spans several modalities including email discussions, collaborative document editing, face-to-face meetings and teleconferencing. Importantly, all of these modalities are documented via transparency reports ranging from public email archives to minutes from meetings. This project has partnered with the IETF to help model and streamline their decision making process. We will borrow from their experience, and employ our methods to extract decision making bottlenecks. We will devise tooling which will provide advice and proposed interventions to relevant parties within the IETF. Amongst many other things, we directly benefit the IETF, and the global Internet standards community, by helping them to uncover biases and help make important decision processes accountable.

The project will generate ground-breaking advances in the automated processing of natural language and semi-structured data relating to social decision making. By partnering with the IETF, it will also apply these results to improve the process of developing Internet standards, and to streamline the IETF's decision making process. We envisage impacts from both of these sets of outcomes, and their combination across various themes, discussed below:

Novel research approaches: By bringing together researchers in NLP, social network analysis, computer networks and psychology we expect novel synergy-producing approaches with impact across academic disciplines (see Academic Beneficiaries).

Datasets and tools: By producing and publicly releasing the first large-scale multimodal dataset of real-world distributed decision-making data we will benefit researchers in many areas (see Academic Beneficiaries). We will ensure that all publications, documentation, data and software are made publicly available e.g. via GitHub and the UK Data Archive to encourage uptake and re-use.

Technology transfer: The new methods we will develop will lead to economic and societal impact via technology transfer. Methods for understanding multi-modal human-human interaction (not only in decision-making but also in many social and business environments) will be commercially valuable for building dialogue systems, e.g. chatbots, business and decision support systems (amongst other possibilities). Further, our impact plan will help transfer this to commercial reality via spinout formation and/or licensing to existing UK industry including companies working with our partner organisation, the IETF. This will be supported by QMUL's technology transfer company, QM Innovation Ltd and dedicated mobile app division QApps (www.qappsonline.com); and/or dedicated funding via InnovateUK or similar. The PI has a strong track record of spinout formation, licensing and InnovateUK transfer from NLP research (Chatterbox Labs, IESO Digital Health, Quality Health Ltd).

Impact on the IETF: Our key impact partner is the IETF itself, which will deploy our tools for streamlining and improving their activities. Our research agenda has been coordinated with direct inputs from the IETF, such that the impact can be maximised. Our key long-term impact will come from the improvements to the IETF procedures derived from the project's interventions. Other impacts will be to improve the quality of standards, by better supporting cross-area review, and to help improve community representation by highlighting bottlenecks and biases in the process. To further the impact and extend it beyond the life of this project, we will work towards the integration of our tools with the IETF's datatracker (a web portal, which exposes public data about standards activities).

Impact on industry and society via internet standards: this positive impact on internet standards, and particularly opening them up to inputs from more diverse groups, will feed into better-met industry needs. Since this will have a direct benefit on the technical decisions underpinning how the Internet operates, the impact will eventually extend to most members of society. The UK is a key player in IETF, and we will work with industry to demonstrate how our techniques can improve their effectiveness, ensuring future Internet standards continue to reflect UK interests. Our impact plan therefore focuses on reaching and influencing three main constituencies: (1) organisations/companies interested in standardisation and social decision making (Sky, Ericsson, JISC); (2) the IETF, which has an inherent interest in improving their activities (Eggert, Navarro, Ford, Oever); and (3) the research community interested in exploring social decision making in other domains, as well as those interested in the standardisation process and the interplay of the agents building the Internet (Lascarides, Doty).