Artificial Intelligence

Dr Joanna Bryson, Dr Marina De Vos, Dr Julian Padget

Artificial Intelligence is a hybrid between two disciplines: the engineering of smarter machines using natural intelligence as a model, and the science of understanding natural intelligence using machines as models. The Bath AI group has internationally-known researchers in both of these areas.

Although divisions do not mean much in this group, since by its very nature AI research tends to be inter-disciplinary, current research activities for this group can be placed under three broad headings:

Agents, including:

• Electronic commerce
• Intelligent agents
• Mobile agents
• Agent societies
• Institutional frameworks
• Agent-based modelling
• Logics for agent-based systems

Artificial models of natural intelligence, including:

• Spiking neural networks
• Realistic emotions for avatars and intelligent assistants
• Modelling social behaviour, the evolution of human behaviour, and task learning
• Machine learning and artificial life
• AI music systems

Knowledge representation and reasoning, including

• Semantic web services (mathematics and music)
• Semantic brokerage
• Mathematical reasoning
• Answer set programming
• Ontologies and ontological reasoning
• Game theory
• Reactive planning and goal arbitration

Agents research is driven by the exploration of the specification, validation and application of institutional frameworks that can be used to mediate agent-to-agent and agent-to-human interaction. The defining factors of institutions are the norms, rules and procedures that may be combined hierarchically to give the institution its intrinsic characteristics, making them an ideal framework with which to model, understand and exploit complex systems. Current applications of institutions are in the delivery of culture and tourism information to wireless devices (involving mobility, user modelling and information retrieval), discovery and orchestration of mathematical web services (involving semantic brokerage, mathematical reasoning and ontology development), both of which have significant overlap with knowledge representation and reasoning. A third application area is music, where the goal is that agents can interpret a score in real-time and interact with human performers, possibly in distant geographical locations, to create a unified musical performance (involving ontology development, web services and institutions). Of course, there is also basic research on institutions themselves, including how to represent the rules that institutional participants should observe and how software agents might reason about them.

Artificial Models of Natural Intelligence: This research area is dedicated to using the best available AI technology in order to advance the scientific understanding of animal and human behaviour. These goals are manifest both in our own scientific research, and in our engineering of accessible AI tools. We provide general-purpose tools over the Internet, whilst also working directly with collaborators at a number of universities. Current research includes building a complete model of a jellyfish nervous system; explaining the differences between different species of monkey (macaque) social organizations; modelling the origins of human language; explaining the role of the hippocampal system in learning associative knowledge; explaining the coordination of social predators such as lions, wolves and tuna; making smarter VR game opponents through observational learning; and providing intelligent assistants for dementia patients so that they can live longer in their own homes.

Knowledge Representation and Reasoning research is driven by the various ways knowledge describing a problem, environment or an entity’s state of mind, can be represented and reasoned with, in order to fulfil a certain goal. Just as in natural organisms, representation of knowledge has its effects on the way the type or complexity of reasoning can be performed. Current work uses decidable fractions of logic, non-monotonic reasoning structures, and action languages. Knowledge can be studied through logic, sociology, biology and psychology, but from whichever perspective, the goal is the same: to represent the knowledge effectively and to reason efficiently at as high a level as possible, whether the domain is a web-service for mathematics, a data-mining, a decision-problem, diagnostic application, a negotiating multi-agent framework, the semantic web or a reactive planner.

An alternative approach to this topic is Machine Learning, through which we can discover or infer knowledge by training or experience. We are interested in how to learn compact knowledge structures reliably that perform complex tasks accurately using simple rewards, whilst finding the optimal generalisations. We are also seeking to develop genetic-based machine learning techniques that autonomously discretize the environment to build high-level internal models that can be used to enhance future learning and performance.

Recent publications:

Bryson, JJ., Ando, Y. & Lehmann, H. (2007). Agent-based modelling as scientific method: a case study analysing primate social behaviour. Philosophical Transactions of the Royal Society, B -- Biology, 362(1485):1685-1698.

Wood, MA. & Bryson, JJ. (2007). Skill Acquisition Through Program-Level Imitation in a Real-Time Domain. IEEE Transactions on Systems, Man and Cybernetics Part B--Cybernetics, 37(2):272-285.

Julian Padget, Richard Vidgen, James Mitchell, Amy Marshall, and Rick Mellor. Sendero: an extended, agent-based implementation of Kauffmans NK and NKCS models. Journal of Artificial Societies and Social Simulation, October, 2009.  http://jasss.soc.surrey.ac.uk/

Owen Cliffe, Marina De Vos, and Julian Padget: Modelling Normative Frameworks using Answer Set Programing. 10th International Conference on Logic Programming and Nonmonotonic Reasoning (LPNMR 09) Potsdam, Germany, 14-18 September, 2009.

D. Cosker, R. Borkett, D. Marshall, and P. L. Rosin, "Towards Automatic Performance Driven Animation Between Multiple Types of Facial Model", IET Computer Vision,  2008.

D. Cosker, D. Marshall, P. Rosin, S. Paddock, S. Rushton, ``Towards Perceptually Realistic Talking Heads: Models, Metrics and McGurk'', ACM Transactions on Applied Perception, vol. 2, no. 3, 2005.