Agoric computation: trust and cyber-physical systems

In the past two decades advances in miniaturisation and economies of scale have led to the emergence of billions of connected components that have provided both a spur and a blueprint for the development of smart products acting in specialised environments which are uniquely identifiable, localisable, and capable of autonomy. Adopting the computational perspective of multi-agent systems (MAS) as a technological abstraction married with the engineering perspective of cyber-physical systems (CPS) has provided fertile ground for designing, developing and deploying software applications in smart automated context such as manufacturing, power grids, avionics, healthcare and logistics, capable of being decentralised, intelligent, reconfigurable, modular, flexible, robust, adaptive and responsive. Current agent technologies are, however, ill suited for information-based environments, making it difficult to formalise and implement multiagent systems based on inherently dynamical functional concepts such as trust and reliability, which present special challenges when scaling from small to large systems of agents. To overcome such challenges, it is useful to adopt a unified approach which we term agoric computation, integrating logical, mathematical and programming concepts towards the development of agent-based solutions based on recursive, compositional principles, where smaller systems feed via directed information flows into larger hierarchical systems that define their global environment. Considering information as an integral part of the environment naturally defines a web of operations where components of a systems are wired in some way and each set of inputs and outputs are allowed to carry some value. These operations are stateless abstractions and procedures that act on some stateful cells that cumulate partial information, and it is possible to compose such abstractions into higher-level ones, using a publish-and-subscribe interaction model that keeps track of update messages between abstractions and values in the data. In this thesis we review the logical and mathematical basis of such abstractions and take steps towards the software implementation of agoric modelling as a framework for simulation and verification of the reliability of increasingly complex systems, and report on experimental results related to a few select applications, such as stigmergic interaction in mobile robotics, integrating raw data into agent perceptions, trust and trustworthiness in orchestrated open systems, computing the epistemic cost of trust when reasoning in networks of agents seeded with contradictory information, and trust models for distributed ledgers in the Internet of Things (IoT); and provide a roadmap
for future developments of our research.