Artificial Models of Natural Intelligence

Bath AI  |  Intelligent Systems Group  |  Contact Us 

Older Research

This is a subset of our recent AmonI research projects, biased towards things we are interested in continuing working on if we find the right people and / or funding.  More former projects are available from the list of Joanna Bryson's former students, Joanna Bryson's former research, Alwyn Barry's research page and Daniel Richardson's home page. For more information on the below projects, see the author's publications pages. The sofware associated with many of these projects is available on the AmonI software page.

Making NI Modelling Easier

  • In addition to our current research on  Behavior Oriented Design (BOD), we maintain a BOD version of the agent-based modelling tool, MASON, as well as the jython/python version of BOD's action selection, jyPOSH.  We have used BOD/MASON both for modelling, and for training, though more could be done with this.  Jan Drugowitsch did a lot of this work, as did Tristan Caulfield and Andy Kwong.
  • Probably the best version of BOD/POSH for agent-based modelling is Michael Brooks' BODNetLogo.
  • The sort of planning we use for real-time AI (POSH dynamic plans) has also given us insight into what's wrong with a different kind of planning, project planning.  We have been working to develop tools & process around this. Meri Williams made a good start on this, and John Baillie Strong made a very good version.  This still needs to be productised and perhaps released with a management book.
  • Our affiliates Cyril Brom, Jacob Gemrot and their teams in Prague support a version of POSH (with an IDE) in their games-development platform Pogamut. They are also interested not only in making game AI easier to build, but also using it attract highschool students to programming.

Social Behaviour and Cooperation

  • Karolina Sylwester, Simon Powers, Daniel Taylor and Joanna Bryson (with advice from Benedikt Herrmann and Simon Gächter) worked on understanding geographic variation in the prevelance of an apparently maladaptive cultural norm, anti-social punishment.  They are using Behavioural Economics to gather the data and Agent Based Modelling to express and examine the theory. Some of this is ongoing work for Bryson and her collaborators.
  • Gideon Gluckman did a paper understanding the impact of primate social organisation on their genetic evolution.  This relates to understanding how cognition and individual learning affect operate as adaptive strategies, a common interest held with Joanna Bryson, Marios Richards and Daniel Taylor.  It also extends from recent work done by Hagen Lehmann on modelling macaque social behaviour.
  • Joanna Bryson and Julia Lehmann are trying to understand the adaptive strategy of fission-fusion.  Joanna is also working with Petra Kaczensky on this, looking at the specific case of Wild Mongolian Asses.  Rob Jenks did a recent model of chimpanzee-like fission-fusion which is not yet published.  Ando Yasushi did an MSc on the impact of predation on primate social organization which also still needs extension for publication.
  • Daniel Taylor is looking at how social structures in early hominids might influence cultural evolution (see below).
  • Michael Tremante explored extending the Behavioural Economics research paradigm into Second Life using VR avatars.
  • Hagen Lehmann completed a PhD on macaque social behaviour.  He and Joanna Bryson investigated the evolutionary origins of primate social organization, with help from Bernard Thierry.
  • Ando Yasushi did an exceptionally good MSc project on the impact of predation on primate social organization. 
  • Rob Jenks similarly did an award-winning undergraduate dissertation on emergent chimpanzee-like fission-fusion social organisation.  Julia Lehmann helped with this work.
  • Alwyn Barry and Dan Richardson has supervised a number of project students on flocking behaviour and group predation  Alex Hilton did an excellent undergraduate project on this.  See also the BOD Sheep & Dog Demo.
  • Paul Rauwolf and Dominic Mitchell worked with Gordon Ingram looking at understanding the impact of gossip on the evolution of communication. Dominic is also interested in language evolution and public language.
  • Daniel Taylor wrote a PhD on understanding the biological evolution of cultural evolution.
  • Joanna Bryson has been publising a good deal of work on how interacting differences in representational capacities may account for the exceptional size of human culture, see our evolving human-like culture web page.
  • We worked with Harvey Whitehouse at Oxford University on understanding the origins of religious behaviour.
  • See also our previous experience with real time imitation learning.

Naturalistic Emotions

  • Emmanuel Tanguy built a system, EE-FAS, which produces natural real-time models of emotions for facial animation using the Dynamic Emotion Representation.  This system is now being extended.  For example, Borui Zhang & Dr. Tanguy integrated EE-FAS with Elzware's chat bots; Jiang Yifan worked on using the emotional state to alter the speech production appropriately. Mary Estell incorporated the avatar into an intelligent assistive environment.
  • Joanna Bryson and Hagen Lehmann were investigating how emotions interact with the rest of action selection in primate social models, then they got Philipp Rohlfshagen to fix it.  This is now a part of the standard jyPOSH, BOD/MASON, and BODNetLogo distributions.
  • Paula Ellis demonstrated that people's ability to recognize emotions on VR characters depends heavily on the texture (picture) the character wears.  This work should be extended.
  • Visiting student Maria Shugrina and BAI member John Collomose made an intelligent system, Empathic Painting that adjusts art to the viewer's emotional state.
  • Swen Gaudl and Rob Wortham both created emotion systems improving on Philipp Rohlfshagen's as part of their PhDs

Individual Learning and Modular Intelligence

  • One of the most interesting characteristics of intelligence is its modularity.  Joanna Bryson has since her PhD been working on modular models of a standard benchmark task from experimental psychology: Transitive `Inference' Performance.  Her published work was completed in collaboration with Mark Baxter and Jonathan C. S. Leong.  See also our current work on task learning and our primate learning web page.
  • Jack Kenyon extended this research as an undergraduate project by gathering data on implicit learning of TI in games environments.
  • Mark Wood modelled TI in ACT-R, which helped us better understand ACT-R and our own model. John Mann started this work final-year project. Emily Korvin worked on extending the TI results to transverse patterning in her 2004 final-year project.
  • Mark Wood for his PhD developed a common framework for individual and social learning.
  • Jan Drugowitsch did a PhD with Alwyn Barry on formalizing our understanding of Learning Classifier Systems, which combine genetic algorithms, reinforcement learning and production rule systems. Alwyn Barry also cosupervised Dr.  Jacob Hurst's PhD at UWE on LCS in robot action selection.
  • Bidan Huang did a PhD on modular representations for imitation learning of human behaviour for real-time control of robot grasping, working with our collaborator Aude Billard at EPFL.
  • Joanna Bryson has has a long-running interest on modularity in learning, for example why different regions of the brain learn at different rates and with different physical representational neural substrates. Her work is particularly focussed on modular models of a standard benchmark task from experimental psychology: Transitive `Inference' Performance.
  • Eugene Bann is working on determining the accuracy of models of emotions by looking at how they are implicitly described in language on twitter.

Social Learning, Memetics and Cultural Evolution

  • Mark Wood did a PhD on the computational tractability of imitation and social learning in task learning for his PhD.  Joanna Bryson has also done work on how culture and modularity help with this tractability issue, inspired by Dan Sperber.
  • Ivana Cace started our modelling of the evolutionary consequences of the altruistic communication of behaviour, when doing her diploma project for Utrecht with AmonI.  Avri Bilovich also helped fwith this research (funded by a Nuffield Bursary), and Christos Bechlivanidis extended the results to the evolution of prestige.
  • Joanna Bryson and Yifei Wang are working with Sam Brown on bacterial social behaviour and communication.
  • Joanna Bryson did sabbatical research from 2007-2009 in Vienna working on modelling the biological evolution of cultural evolution.
  • Joanna Bryson has also looked at how memetics affects semantics and language learning. With Will Lowe and Avri Bilovich  we have demonstrated these techniques can detect shifts in semantics over centuries, by comparing the Bible, Shakespeare and the British National Corpus. This work is now in Science with the help of collaborators Arvind Naraynan and Aylin Caliskan (thanks to another sabbatical, at Princeton.)

Biological Evolution, Development and Learning

  • Marios Richards worked on understanding the Baldwin Effect and how it interacts with development and individual variation.
  • Dan Taylor worked on understanding lineages, particularly in memetics.
  • Daniel Taylor was also very interested in cultural evolution specifically and evolutionary mechanisms more generally. He is working on Rogers Paradox and on modelling cultural drift and sexual selection.

Naturalistic Neural Networks

  • Dr. Richardson and two of his PhD students, Andrew Carnell and Carl O'Dwyer, are working on the dynamics of neural networks. This includes looking at spike timings and Liquid State Machines (e.g CSim, due to Maes). See Dan Richardson's home page for other papers on the topic.

Cognitive Systems:  Robots, intelligent spaces, VR Avatars and Computer Game Characters

Not all of our applications are scientific.
  • Human-like intelligence is useful in computer games too. Besides being fun, this is a good test-bed for the usability of BOD & helps us debug and improve it.  This work was most recently done by Francis Binns and Andrew Mansfield.  Sam Partington and Andy Kwong started this work at Bath. First Cyril Brom then Jakub Gemrot have been working with  a group of people working on it at Charles University, Prague to make creating game intelligence easier with a goal to using it as a platform for teaching programming.
  • Swen Gaudl did a PhD on improving game character AI, possibly through using genetic algorithms.
  • Bidan Huang, Jason Leake and Joanna Bryson extended BOD to work on the iCub robot, with help from collaborators and their groups at Imperial (Demiris), Plymouth (Belpaeme) and EPFL (Billard).
  • Mark Wood applied and demonstrated his PhD research on integrating individual and social learning using a single AI in UT.
  • Joanna Bryson has helped advise the Bath University Submarine Racing Team (BURST) for their annual competition effort, especially their autonomous robot team.
  • Joanna Bryson, Dylan Evans and Veronica Sundstedt put together a website to help promote and popularise cognitive systems like domestic robots, autonomous rescue robots and assistive environments.
  • Joanna Bryson originally became interested in intelligent environment through Michael Coen and Krzyzstof Gajos.  She was brought into  the medical end of it originally by Jim Edwardson of  Newcastle University's Institute of Aging and Health on this problem. Two of her project students, Mary Estell and Andrew Beggs have built prototype systems for this utilising BOD, the DER and Tanguy's facial avatar EE-FAS.
  • Bidan Huang, Jason Leake and Joanna Bryson are working with the Bath Institute for Medical Engineering on designing robots to assist nurses and help control infection.
  • Human-like intelligence is useful in computer games too. Besides being fun, this is a good test-bed for the usability of BOD & helps us debug and improve it.  The most recent published work on this is by John Grey, who created generative quests for non-player characters.  These give the characters individual history and social networks, making them generally more interesting to interact with.  This work needs extending. In 2012, Simon Davies made a BOD agent for StarCraft II: Brood War, and Tom Hyde made a BOD team for RoboCup 2D Simulation League.

Books and Previous Meetings