Edinburgh Research Explorer

Planning and Doing Things

Citation for published version: Tate, A 2007, 'Planning and Doing Things', AISB Quarterly, vol. 125, pp. 7-8.

Link: Link to publication record in Edinburgh Research Explorer

Document Version: Publisher's PDF, also known as Version of record

Published In: AISB Quarterly

General rights Copyright for the publications made accessible via the Edinburgh Research Explorer is retained by the author(s) and / or other copyright owners and it is a condition of accessing these publications that users recognise and abide by the legal requirements associated with these rights.

Take down policy The has made every reasonable effort to ensure that Edinburgh Research Explorer content complies with UK legislation. If you believe that the public display of this file breaches copyright please contact [email protected] providing details, and we will remove access to the work immediately and investigate your claim.

Download date: 07. Oct. 2021

Quarterly No.125

The Newsletter of the Society for the Study of Artificial Intelligence and Simulation of Behaviour

In this issue:

Features Three-Dimensional • Daniel Myatt and Slavomir Nasuto 1 Three Dimensional Recon- Reconstruction of Neurons struction of neurons

• Nicholas Jennings 3 with Neuromantic Decentralised Control of Complex Systems

• Austin Tate 7 Despite the tendency of traditional artificial neural nected cylinders of varying radius and length. The Planning and Doing Things networks to simplify the role of dendritic and axonal tree is thus stored as a series of points, where each morphology to simple connections between neurons, point is associated with a 3D position and radius, as Society in reality neurites have an important role to play in well as having a corresponding ‘parent point’. From • AISB Symposia 9 determining neuronal behaviour. For example, the such a model, the behaviour of the neuron may be AISB’08 and call for pro- posals for AISB’09 diameter significantly affects the oscillatory behaviour estimated using a compartmental simulator such as of a neurite. Similarly, signals take time to travel NEURON [2] or GENESIS [3]. Also, it is possible Reviews down the length of dendrites, adding a spatiotemporal to calculate a variety of useful statistics from such • Ariel Procaccia 9 Cooperative Information characteristic into neuronal operation. models to either classify them or differentiate between Agents a control and experimental group. Father Hacker There are two main motivations for the creation • Computational Theology 12 of models of neuronal morphology. Primarily, such The raw data required for a reconstruction is a models are required in order to investigate the role stack comprised of a series of images taken of the that the morphology of a neuron plays in determin- tissue sample at varying distances away from the ing function by allowing validation of computational microscope lens. In each image a different planar simulations with the results of in vitro electrophysi- section of the sample is in focus, and thus the image ological experiments. Additionally, it is thought that stack as a whole forms a pseudo-3D representation aberrant neuronal morphology may play a role in of the neuron. (continued on p. 2) brain diseases such as epilepsy[1], and thus being able to meaningfully characterise abnormalities may yield novel insight into such conditions. The standard way to model neuronal morphology is to represent the dendritic tree as a series of con-

Figure 1. The basic process of neuronal reconstruction: from the original image stack (with resolution 2862 by 1649 and 86 images) in the top left panel, points are marked out sequentially down the dendrites (top right panel). This process culminates in a full 3D model of the basal tree converted to metric coordinates (bottom panels).

No. 125, Spring 2007  Three-Dimensional Simulation of Neurons

The AISB Quarterly is published by the Society for the Study of Typically, neuronal reconstruction starts at the References Artificial Intelligence and soma and sequentially marks out points down Simulation of Behaviour. the dendritic tree and their corresponding 1. B.J. Whalley, M. Postlethwaite and A. Constanti, Further ISSN 0268-4179 radii, either manually or automatically. The characterization of muscarinic agonist-induced epileptiform bursting activity in immature rat piriform cortex, in vitro, most manual form of reconstruction requires Copyright © AISB 2007 Neuroscience 134(2), p 549-66. 2005. the user to explicitly identifying points along Articles may be reproduced as long as each dendrite, whereas more advanced meth- 2. N.T. Carnevale and M.L. Hines, The NEURON Book, the copyright notice is ods can take distant points on a dendrite and Cambridge University Press, 2006. included. The item should automatically trace between them. The image be attributed to the AISB 3. U.S. Bhalla, The Book of GENESIS: Exploring Realistic Quarterly and contact slice for which a given short section of den- information should be Neural Models with the GEneral NEural SImulation System, listed. drite is most in focus yields the correspond- second edition, New York: Springer-Verlag, 1998. ing Z-coordinate. Hence, the reconstruction is Quarterly articles do not 4. K.M. Brown, D.E. Donohue, G. D’Alessandro and G.A. necessarily reflect the discretised in the Z axis. official AISB position on Ascoli, A Cross-Platform Freeware Tool for Digital Recon- issues. struction of Neuronal Arborizations From Image Stacks, Neuromantic, a freeware tool that is currently Neuroinformatics 3(4), p 343-359, 2005. Editor under development at the University of Read- Colin G. Johnson. Computing Laboratory, ing, is designed to facilitate the reconstruc- 5. E. Meijering, M. Jacob, J.C.F. Sarria, P. Steiner, H. Hirling University of Kent at tion of neurons across a variety of microscopy and M. Unser, Design and Validation of a Tool for Neurite Canterbury. techniques. Figure 1 illustrates the basic Tracing and Analysis in Fluorescence Microscopy Images, [email protected] Cytometry 58A, p 167-176, 2004. http://www.aisb.org. process of reconstruction using an image stack uk/aisbq/index.shtml of the basal tree of a pyramidal neuron [4]. 6. W.T. Freeman and E.H. Adelson, The Design and Use of Advertising and Recent experiments with Neuromantic show Steerable Filters, IEEE Transations on Pattern Analysis and Administration Machine Intelligence 13(9), p 891-906, 1991. Therie Hendrey-Seabrook that it is more time efficient for semi-manual [email protected] reconstruction than other comparable applica- School of Science and Technology tions such as Neurolucida and Neuron_Mor- Darren Myatt and Slawomir Nasuto University of Sussex pho. Work is currently ongoing to complete School of Systems Engineering Falmer, Brighton UK, BN1 9QH the semi-automatic tracing capabilities of the The University of Reading Tel: (01273) 678448 program, which employs a 3D extension of Email: [email protected] Fax: (01273) 671320 the algorithm used by NeuronJ [5]: steerable http://www.rdg.ac.uk/neuromantic AISB PATRON gaussian filters [6] are employed to identify John Barnden University of Birmingham neurites and a modification of Djikstra’s algo- rithm is used to calculate optimal cost routing AISB FELLOWS Prof Harry Barrow in real-time. Schlumberger Prof Margaret Boden University of Sussex The laborious task of reconstructing neurons Prof Mike Brady has resulted in a significant bottleneck in University of Oxford Prof Alan Bundy computational neurobiology, as it is generally University of Edinburgh difficult to obtain the sample size necessary Prof Tony Cohn for large-scale statistical comparisons. It is Prof John Fox hoped that by providing a freeware tool that Imperial Cancer Research Fund significantly reduces the user effort required, Prof Jim Howe further research into this interesting area can University of Edinburgh Prof Nick Jennings be stimulated. University of Southhampton Prof Aaron Sloman University of Birmingham Dr Richard Young University of Hertfordshire Prof Austin Tate University of Edinburgh

 AISB Quarterly Decentralised Control of Complex Systems

Many modern computing systems have Jennings 2001). In particular, autonomous 2007), automated negotiation (Fatima et to operate in environments that are highly agents are a natural way of viewing flexible al., 2004; Fatima et al., 2006), coordination interconnected, highly unpredictable, in a service providers and consumers and the (Rogers et al., 2007b) and computational constant state of flux, have no centralized interactions between these autonomous mechanism design (Dash et al., 2003). In point of control, and have constituent com- components are naturally modeled as some the latter case, we have built applications ponents owned by a variety of stakeholders form of economic trading process that, if using these techniques in areas such as: that each have their own aims and objec- successful, results in a service contract virtual organizations (Norman et al., 2004), tives. Relevant exemplars include the Web, (or service level agreement) between the sensor networks (Padhy et al., 2006; Rog- Grid Computing, Peer-to-Peer systems, agents involved. ers et al., 2005; Rogers et al., 2006) and Sensor Networks, Pervasive Computing personalized recommendations (Wei et al., and many eCommerce applications. Now, With the team here at Southampton, 2005; Payne et al., 2006). I believe that all of these systems can be we have focused, in particular, on the viewed as operating under the same con- design of the agents and their interac- In what follows, however, I will fo- ceptual model: (i) entities offer a variety tions. Specifically, we have concentrated cus on just one of these applications to of services in some form of institutional both on the fundamental science involved provide more details of the fundamental setting; (ii) other entities connect to these in constructing such computational service methodology of this work. Specifically, I services (covering issues such as service economies and in how these techniques will focus on multi-sensor networks (MSN) discovery, service composition and service can be applied in a variety of real-world that are being deployed in a wide variety procurement); and (iii) entities enact ser- applications. In the former case, we have of application areas ranging from military vices, subject to service agreements, in made advances in the areas of game theory sensing to environmental monitoring and a flexible and context sensitive manner. (Gerding et al., 2007), auctions (Dash et traffic control. Such networks consist of a Moreover, I believe agent-based comput- al., 2007, David et al., 2007; Rogers et number of sensors connected via a com- ing is an appropriate computational model al., 2007a; Vetsikas et al., 2007), coali- munication network. for conceptualizing, designing and imple- tion formation (Dang and Jennings, 2006; menting such systems (Jennings, 2000; Dang et al., 2006; Rahwan and Jennings,

No. 125, Spring 2007  Decentralised control of complex systems

Each sensor is able to sense their local ferent stakeholders. In such cases, the vehicles (see figure 1). Each sensor is environment, but can also make use of data sensors are operating in a competitive provided with an imprecise estimate of its transmitted from neighbouring sensors in rather than cooperative environment, and own location by the navigation system of order to improve the accuracy of their own thus, they may attempt to optimize their the aerial vehicles to which it is mounted, measurements (e.g. by combining their own own gain from the network, at a cost to and it is tasked with detecting and tracking noisy observation, with the observations the performance of the entire system. To multiple targets within a region of obser- from a number of other sensors, in order address this challenge we have explored vation immediately surrounding itself (see to reduce the final uncertainty). the application of computational mechanism figure 2). Within its region of observation, design and auctions within information the sensor is able to estimate the posi- To date, research in sensor networks fusion scenarios. tion of each target by making noisy or has mainly concentrated on using coop- imprecise measurements of the range and eration among distributed sensors. At its In this work, we consider a real world bearing of the target from itself. However, core, this approach involves determining aerial surveillance scenario such as that in order to better resolve the uncertainty the exchanges of observed data between posed in disaster relief contexts, where in these position estimates, the sensors the sensors that results in the maximum multiple emergency response agencies, must acquire target observations from gain in information across the whole sensor with aerial vehicles of different capabilities, neighbouring sensors and then fuse these network. However, this approach overlooks must interact in order to locate casualties. observations with their own. the fact that in some applications each The sensor network is formed from sensors sensor may be individually-owned by dif- that are mounted onboard these aerial

 AISB Quarterly Decentralised control of complex systems

The sensors are connected together via that allows us to assign a value to the are said to be strategy-proof or incentive a communication network that has a limited observations provided by each sensor. compatible (Dash et al., 2005). That is, bandwidth, and thus, there must be some In our work, we have derived a metric the sensors have a dominant strategy to coordination to determine which data should based on the Fisher information of these truthfully reveal the information content be sent to which sensors in order to make observations. This measure is related to of their observations to the auctioneer best use of this limited global resource. the precision of the observations, and (they thus have no incentive to engage However, since each sensor is owned by thus sensors that make range and bear- in complex strategic behaviour). Such an a different stakeholder, it is thus selfishly ing measurements with greater precision, outcome is achieved through the payments seeking to maximize the accuracy of its generate observations with greater informa- that the sensors must make (or collect) own target position estimates. As a result, tion content, and hence greater value. This for receiving (or transmitting) observations. each sensor has a positive disinclination to metric is particularly attractive as when These payments are calculated by the share its observations since in doing so, it independent observations are fused, the auctioneer in order to align the individual will occupy valuable bandwidth which may information content of the fused observa- goals of the sensors with the overall sys- be used to receive observations from other tion, is simply equal to the sum of the tem-wide goals of the system designer. agents. Thus, we use techniques from two individual observations. By incentivising truthful reporting of the mechanism design to engineer a protocol information content of observations from that incentivises the sharing of observa- Adopting this measure allows us to the sensors, the auctioneer is then able tions, whilst also ensuring that the global develop a protocol, whereby an independent to ensure that the bandwidth is allocated resource of communication bandwidth is auctioneer allocates the communication efficiently (i.e. to maximize the information used effectively. bandwidth based on the valuations of obser- gain of the entire network). vations supplied by the individual sensors. However, in order to apply the tools Thus far, we have focused our research of mechanism design we require a metric onto a class of direct mechanisms that

No. 125, Spring 2007  Decentralised control of complex systems

Proc. 21st National Conference on AI (AAAI), 285-289. To demonstrate the application of this Boston, USA, 635-640. T. Rahwan and N. R. Jennings (2007) “An mechanism, the information fusion scenario, R. K. Dash, P. Vytelingum, A. Rogers, E. Da- algorithm for distributing coalitional value cal- the auction mechanism, and the individual vid and N. R. Jennings (2007) “Market-based culations among cooperating agents” Artificial task allocation mechanisms for limited capacity Intelligence Journal. sensors have been implemented within a suppliers” IEEE Trans on Systems, Man and A. Rogers, E. David and N. R. Jennings (2005) fully asynchronous Java simulation environ- Cybernetics (Part A). “Self organized routing for wireless micro-sensor ment (see figure 3 and visit http://www. R. K. Dash, D. C. Parkes and N. R. Jennings networks” IEEE Trans. on Systems, Man and ecs.soton.ac.uk/~acr/demonstrator/). The (2003) “Computational Mechanism Design: A Cybernetics (Part A) 35 (3) 349-359. main view shows the measurements and Call to Arms” IEEE Intelligent Systems 18 (6) A. Rogers, R. K. Dash, N. R. Jennings, S. Reece 40-47. and S. Roberts (2006) “Computational mecha- communication of the individual sensors R. K. Dash, A. Rogers, S. Reece, S. Roberts and nism design for information fusion within sensor that make up the sensor network. The N. R. Jennings (2005) Constrained Bandwidth networks” Proc. 9th Int. Conf. on Information graphs to the right record metrics of the Allocation in Multi-Sensor Information Fusion: Fusion, Florence, Italy. entire system such as the total information A Mechanism Design Approach. In Proceed- A. Rogers, E. David, J. Schiff and N. R. Jen- gain of the network, the utilization of the ings of The Eighth International Conference on nings (2007a) “The effects of proxy bidding and Information Fusion (Fusion 2005), Philadelphia, minimum bid increments within eBay auctions” available bandwidth, and the outputs of PA, USA. ACM Trans on the Web 1. the auction process. E. David, A. Rogers, N. R. Jennings, J. Schiff, S. A. Rogers, R. K. Dash, S. D. Ramchurn, P. Kraus and M. H. Rothkopf (2007) “Optimal design Vytelingum and N. R. Jennings (2007b) “Co- In this work, we have demonstrated of English auctions with discrete bid levels” ACM ordinating team players within a noisy iterated an effective and efficient control mecha- Trans on Internet Technology 7 (4). Prisoner’s Dilemma tournament” Theoretical S. S. Fatima, M. Wooldridge and N. R. Jennings Computer Science. nism for sensor networks, in which the (2006) “Multi-issue negotiation with deadlines” I. Vetsikas, N. R. Jennings and B. Selman constituent sensors are owned by dif- Journal of AI Research 27 381-417. (2007) “Generating Bayes-Nash equilibria to ferent stakeholders that have their own S. Fatima, M. Wooldridge and N. R. Jennings design autonomous trading agents” Proc 20th goals and objectives. Our system has a (2004) “An agenda based framework for multi- Int. Joint Conf. on AI (IJCAI), Hyderabad, India, principled valuation metric based on the issues negotiation” Artificial Intelligence Journal 1543-1550. 152 (1) 1-45. Y. Z. Wei, L. Moreau and N. R. Jennings (2005) Fisher information of a sensor’s observa- E. H. Gerding, R. K. Dash, D. C. K. Yuen and “A market-based approach to recommender tions and uses computational mechanism N. R. Jennings (2007) “Bidding optimally in systems” ACM Trans on Information Systems design to engineer a sensor network with concurrent second-price auctions of perfectly 23 (3) 227-266. a particular desirable system-wide property substitutable goods” Proc. 6th Int. J. Confer- (in this case, the efficient use of limited ence on Autonomous Agents and Multi-agent Nicholas Jennings Systems, Hawaii, USA. bandwidth capacity), despite the selfish University of Southampton N. R. Jennings (2000) “On Agent-Based Software goals and actions of the individual sen- Engineering” Artificial Intelligence Journal, 117 [email protected] sors. Whilst this research is ongoing, and (2) 277-296. is currently focusing on how the role of N. R. Jennings (2001) “An agent-based approach the auctioneer may be distributed amongst for building complex software systems” Comms. of the ACM, 44 (4) 35-41. the sensors within the network, it is clear T. J. Norman, A. Preece, S. Chalmers, N. R. that computational mechanism design offers Jennings, M. Luck, V. D. Dang, T. D. Nguyen, V. an invaluable and powerful set of tools Deora, J. Shao, A. Gray and N. Fiddian (2004) and techniques to address the challenges “Agent-based formation of virtual organizations” posed by these scenarios. Int. J. Knowledge Based Systems 17 (2-4) 103-111. P. Padhy, R. K. Dash, K. Martinez and N. R. References Jennings (2006) “A utility-based sensing and communication model for a glacial sensor V. D. Dang and N. R. Jennings (2006) “Coali- network” Proc. 5th Int. Conf. on Autonomous tion structure generation in task-based settings” Agents and Multi-Agent Systems, Hakodate, Proc. 17th European Conference on AI, Trento, Japan, 1353-1360. Italy, 210-214. T. R. Payne, E. David, N. R. Jennings and M. V. D. Dang, R. K. Dash, A. Rogers and N. R. Sharifi (2006) “Auction mechanisms for efficient Jennings (2006) “Overlapping coalition formation advertisement selection on public displays” Proc. for efficient data fusion in multi-sensor networks” 17th European Conference on AI, Trento, Italy,

No. 125, Spring 2007  Planning and Doing Things

I was interested in computers by internal goal structure recorded. Edinburgh some pioneering and deployed influential the age of 15 and gave talks on them is a fantastic environment with many visi- systems in everyday use - for yellow pages at school. I attended evening classes a tors and collaborators. Nonlin was created layout, spacecraft assembly, integration and couple of years later while still at school to be used on real applications with the test, industrial plant diagnosis, logistics travelling on the bus for an hour in the UK electricity generation utility for turbine support and so on. evening to a college in Leeds to learn overhaul procedure project planning. It was programming (in COBOL!). Computers at also used to drive the Freddy II robot in My own Planning and Activity Man- that time filled a room, you submitted your a project led by Robin Popplestone. Its agement Group within AIAI is exploring exercises on punched card and got the core algorithms are still at the heart of representations and reasoning mecha- results the following day. I built my first most Hierarchical Task Network (HTN) and nisms for inter-agent activity support. AI planner over 35 years ago. I’d already partial-ordered (POP) planners. The agents may be people or computer been on an early AI course at Lancaster systems working in a coordinated fashion. University where the language of choice I had a period in the second half of the The group explores and develops generic for teaching a range of topics was POP-2 1970s working on IBM and ICL mainframes approaches by engaging in specific ap- and wanted to do a Summer project to leading a team which developed commer- plied studies. Applications include crisis create a problem solver. With support from cial data base software for engineering action planning, command and control, Donald Michie and his team at Edinburgh applications, and the team supported that space systems, manufacturing, logistics, I tried to create a Graph Traverser along for business critical applications. Just after- construction, procedural assistance, help the lines they were working on. Boy, am wards I headed up Edinburgh University’s desks, emergency response, etc. Our I glad I got involved with Computers, AI microcomputer support and office systems long term aim is the creation and use of and planning technology! team, when these systems were novel and task-centric virtual organisations involving distributed computing was just starting to people, government and non-governmental Planning is a key area for creating take off. Most computing before that was on organisations, automated systems, grid and intelligent behaviour and a long term aim large centralised mainframe systems. This web services working alongside intelligent of the AI community. I have been doing work on software engineering and being robotic, vehicle, building and environmental my bit to advance the concepts, technol- involved in systems and user help desk systems to respond to very dynamic events ogy and applications we have in his area. support was something that has stayed on scales from local to global. And, in doing so, I have been able to with me through my career. bring in a number of my other interests Over the last decade, I have concen- in search and rescue teams (from a child- Edinburgh University used some of the trated on applying planning and collabo- hood TV programme - Supercar), space funds generated by its commercial activities, ration technology to emergency response travel and future habitats. This has involved in which I was engaged, to give a number tasks, and a number of my projects con- collaboration with scientists, systems de- of academic entrepreneurs a fellowship to tribute to this. In the UK, the Advanced velopers and creative people worldwide. allow them to explore collaborations with Knowledge Technologies (AKT) project I love collaborative projects and the joint industry and with international organisa- with 5 university groups and a number of demonstrations of the results. tions. The three of us that got these industrial and government has pushed the fellowships explored AI (me), Computer research agenda for the semantic web and I joined Donald Michie’s Department of Science (Malcolm Atkinson, now Director knowledge systems on the web and we Machine Intelligence in Edinburgh in the of the National e-Science Centre) and are applying these in a challenging accident early 1970s during which time my AI plan- Electronic Engineering (Peter Denyer, who scenario situated in central London. In the ners Interplan and Nonlin were created. I formed Vision group, a company success- US, we have had DARPA (Defense Advanced shared a flat with my fellow PhD student fully floated on the Stock Exchange). I Research Projects Agency) support for our Dave Warren, who worked on the early joined Jim Howe (Head of the School of work since the late 1980’s) and this has Prolog compilers, using remarkably similar AI at Edinburgh) as he was creating the been directed towards multi-national coali- technology to that I was interested in for Artificial Intelligence Applications Institute tion operations for peace-keeping, collab- planning. I shared an office with Aaron (AIAI) and we obtained commercial and orative operations for disaster relief, search Sloman who was visiting and thinking deep other funding to start this organisation. and rescue coordination, etc. in projects thoughts about the philosophical aspects of I took over as its Director after the first such as CoAX, CoSAR-TS and Co-OPR. In AI. Earl Sacerdoti, who did work on multi- year of operations. AIAI has been a pio- Europe, we are engaged in projects such level plan representation and hierarchical neer in using AI technology for a wide as OpenKnowledge which has emergency planning was a visitor and we worked to- range of applications. It has concentrated response interests in dealing with the gether on concepts for flexible hierarchical, on knowledge systems, planning systems, aftermath of floods in Italy. partial ordered task network planners with and adaptive systems. Its been involved in

 AISB Quarterly Planning and Doing Things

The FireGrid project, in the UK, involves See Austin Tate being interviewed about the use a large team from academia, industry and of artificial intelligence techniques and their use government agencies who are exploring in emergency response centres in March 2007 concepts for emergency response with ad- at http://www.ukfuturetv.com vance simulation in intelligent buildings. Austin Tate Our aim is to create helpful agents AIAI, University of Edinburgh and collaboration between organisations [email protected] and people which can use knowledge of other agents, tasks, procedures, services and the environment so as create a “helpful environment” which improves safety and the lives of everyone.

References

AIAI Planning and Activity Management Group - http://www.aiai.ed.ac.uk/project/plan/

Tate, A. (2006) The Helpful Environment: Geo- graphically Dispersed Intelligent Agents That Collaborate, Special Issue on “The Future of AI”, IEEE Intelligent Systems, May-June 2006, Vol. 27, No. 3, pp 57-61. IEEE Computer Society.

Books for Review

The AISB has a number of books for Thinking About Android Epistemology, Unifying Computing and Cognition review in future issues of the Quarterly. ed. Kenneth Ford et al., MIT Press 2006 Gerry Wolff pp454 If you would be interested in writing a review of one of the books below, please Computer Models of Musical Creativity, From Molecule to Metaphor: A neural contact the Editor. Only AISB members will David Cope, MIT Press, 2006. theory of language Jerome Feldman, pp be sent books for review. If you subse- 357 quently decide that you cannot review the Wired For Speech: How Voice Activates book, either because of time constraints or and Advances the Human-Computer re- Putting Linguistics into Speech Recog- because you don’t think it is (or you are) lationship, Clifford Nass and Scott Brave, nition Manny Rayner et al., University of suitable after reading the guidelines, then pp296 Chicago Press, pp 305. you must let the Editor know. Please see more details on the web at http://www. My Mother was a Computer: Digital aisb.org.uk/aisbq/qbooks.shtml Subjects and Literary texts, N. Katherine Hayles, pp 290 Introduction to Machine Learning, Ethem Alpaydin, pp. 416. Musical Creativity: Multidisciplinary Re- search in Theory and Practice, eds. Irene Visions of Mind, ed. Darryl N. Davis, Deliege and Geraint Wiggins Idea Group Press, 2004.

No. 125, Spring 2007  Conference Report: Cooperative Information Agents

The Tenth International Workshop on rayanan also presented some empirical objections which are often raised against Cooperative Information Agents (CIA 2006) results which imply that the algorithm the semantic web. Most importantly, van was held in The University of Edinburgh often converges rapidly. In general, this Harmelen listed the four main questions eScience Institute, Edinburgh, , on work seems to be an important contribu- in semantic web research: September 11-13. Citing from the workshop tion to the field of multiagent negotiation, website: “An intelligent information agent and may have various applications, not 1. “Where does the meta-data come is a computational software entity that only for information agents, but also for from?” In short, from natural language is capable of accessing one or multiple, e-commerce agents, personal assistant processing, machine learning and social potentially heterogeneous and distributed agents, or indeed for any agent situated communities. information sources, proactively acquir- in a competitive environment. ing, mediating, and maintaining relevant 2. “Where do the ontologies come information or services on behalf of its Another prominent work (nominated for from?” Several problems in the area of human users, or other agents, preferably best paper), entitled “A Fuzzy Approach to ontology-learning remain difficult. just in time and anywhere.” The goal of Reasoning with Trust, Distrust and Insuf- the workshop was to provide a forum for ficient Trust”, was presented by Nathan 3. “What to do with many ontolo- researchers and managers to discuss agent- Griffiths [1]. Griffiths proposed using fuzzy gies?” Ontology mapping remains a very based cooperative information systems, logic to deal with the inherent uncertainty difficult problem; some even consider it as well as this field’s applications to the encountered when reasoning about agents’ the main weakness of the semantic web internet and the World Wide Web. Numer- trustworthiness. Existing approaches in approach. ous interesting projects were presented the field of reputation systems have con- in the workshop; three of them will be sidered fuzzy logic, but have used trust 4. “Where’s the web in the semantic sketched here. as a means of establishing reputation. web?” The most successful applications of Griffiths, on the other hand, decoupled the semantic web are currently in company One work (which won The Best Paper the concepts of trust and reputation (the intranets. However, in recent years the focus Award) with the name “Learning to Negoti- former is an individual assessment while of semantic web research is returning to ate Optimally in Non-stationary Environ- the latter is a social notion), and focused the web itself. ments” was presented by Vidya Narayanan on helping agents interact on the basis (joint work with Nicholas R. Jennings) of their own degree of trust. Griffiths References [2]. Since multiagent interactions can be discussed the three classic concepts of modeled as stochastic games, learning trust, distrust and untrust (where untrust [1] N. Griffiths. A fuzzy approach to reasoning this type of games in the face of partial occurs when an agent is positively trusted, with trust, distrust and insufficient trust. In M. Klusch, M. Rovastos, and T. R. Payne, editors, information has become a major trend in but not sufficiently for cooperation), and Cooperative Information Agents X, volume 4149 the past decade. The goal of the learning added a new notion of undistrust - nega- of Lecture Notes in Artificial Intelligence, pages process is a controversial issue, but most tive trust - which is insufficient for making 360–374. Springer, 2006. works attempt to achieve Nash Equilibrium, crisp decisions in the interaction process. using reinforcement learning techniques. Essentially, Griffiths defined fuzzy terms [2] V. Narayanan and N. R. Jennings. Learning to negotiate optimally in non-stationary environ- Narayanan mentioned that almost all for the abovementioned concepts, which, ments. In M. Klusch, M. Rovastos, and T. R. previous work on multiagent learning has together with fuzzy inference rules, specify Payne, editors, Cooperative Information Agents assumed that the underlying environment agents’ decisions in his framework. Griffiths X, volume 4149 of Lecture Notes in Artificial is stationary (and even when this is not presented experimental data, which vali- Intelligence, pages 288–300. Springer, the case, there are still strict assumptions). dates his approach: when agents use all 2006. Narayanan and Jennings, in contrast, have the fuzzy concepts to make decisions, the [3] F. van Harmelen. Semantic web research considered learning as a tool for negotia- results are superior to selection mechanisms anno 2006: Main streams, popular fallacies, tion; in this setting, assuming a station- in which agents do not make use of the current status and future challenges. In M. ary environment is unrealistic. Narayanan concept of distrust, and far superior when Klusch, M. Rovastos, and T. R. Payne, editors, discussed a model for negotiation based more restrictions are imposed. Another Cooperative Information Agents X, volume 4149 of Lecture Notes in Artificial Intelligence, pages on non-stationary Markov Chains, and interesting topic was presented by Frank 1–7. Springer, 2006. introduced a Bayesian learning algorithm. van Harmelen [3]. van Harmelen gave a The algorithm does not assume knowledge bird’s eye view of the state of semantic web of the opponent’s strategy profile, and fur- research. He gave two interpretations of the Ariel D. Procaccia ther, allows for changes in the opponent’s goals of the semantic web: as the web of Hebrew University of Jerusalem profile over time. Narayanan and Jennings data (integration of data sources over the [email protected] have shown that their algorithm converges web), and an enrichment of the current (in the limit) to an optimal strategy. Na- web. van Harmelen also discussed four

 AISB Quarterly AISB Annual Convention 2008

The AISB 2008 convention will be in convention also aims to provide a fertile a range of excellent restaurants and lively the University of Aberdeen from April 1-4, ground for new collaborations to be forged nightlife and music scenes. Aberdeen is also organised by Frank Guerin and Wamberto between complementary areas. Symposium an excellent base for tourism, whether it Vasconcelos. The theme for the 2008 con- proposals are currently being solicited. This be exploring the Deeside Woodlands and vention is “Communication, Interaction and will continue until 1st September 2007. Cairngorm Mountains or following the Social Intelligence”. This is a return to the Shortly after this the accepted symposia whiskey and castle trails. Other popular same broad area as the highly successful will be announced. Please see the web attractions include golf courses; horse AISB 2005 convention on “Social Intelli- site for more details. riding, windsurfing and skiing. gence and Interaction in Animals, Robots Aberdeen University is one of the an- and Agents”. The 2008 convention aims cient Universities of Scotland, dating back Convention web site: to bring together three areas of research: to the fifteenth century. The convention http://www.aisb.org.uk/convention/ Firstly agent-agent communication and will be hosted in the picturesque King’s aisb08/ interaction, for example the engineering College campus, which forms the heart of of agent societies and electronic institu- Old Aberdeen. Accommodation for conven- Frank Guerin tions; Secondly human-computer com- tion delegates will be provided on campus. University of Aberdeen munication and interaction, for example Aberdeen is very well connected by air, rail [email protected] embodied agents, natural language, and and road. The Airport has regular flights data interpretation and visualisation; Thirdly from Amsterdam, Paris, Copenhagen, Dub- supporting human-human communication lin, London and many other UK airports. and interaction, for example supporting Nestling on the coast between the rivers human teamwork, and social networks. Dee and Don, Aberdeen City is very compact In addition to providing a home for state- with most places accessible by foot. The of-the-art research in specialist areas, the city has a population of 216,000 and has

10 AISB Quarterly Call for 2009 Convention Proposals

AISB COMMITTEE MEMBERS The AISB would like to invite proposals to host Theme. The convention should have a theme the 2009 Convention. that should try to encompass a wide range of work Chair John Barnden in both Artificial Intelligence and the Simulation of University of Birmingham THE AISB CONVENTION Behaviour. A brief justification of your chosen theme [email protected] Vice Chair indicating why you believe it to be timely should Dimitar Kazakov, The AISB Convention is the major annual UK also be included. Our aim that over a period of University of York [email protected] Artificial Intelligence and Cognitive Science Event. years the Convention should represent the broad Secretary/Webmaster In the past the convention has been hosted by spread of UK research in Artificial Intelligence and Louise Dennis University of Liverpool a number of prestigious institutions including the Cognitive Science although this does not preclude [email protected] Universities of Abedeen (2008), Newcastle (2007), the possibility that the Convention may have similar Treasurer/Travel Awards (2006), Hertfordshire (2005), Leeds (2004), themes in two successive years. The committee Kris de Meyer Aberystwyth (2003), Imperial (2002), York (2001), will judge proposals in this context and may make King’s College London [email protected] Birmingham (2000), Edinburgh (1999). It aims to suggestions for adapting a proposed theme to fit [email protected] function as a enue for the presentation of recent and in with this aim. Membership Steve Croker emerging work in the fields of Artificial Intelligence University of Derby and Cognitive Science and as a productive environ- Name and Affliation of the Convention Organ- [email protected] ment for networking and the formation of collabora- iser - including both postal and email addresses and Publications tions. The convention exists as a series of themed telephone numbers. Mark Bishop Goldsmiths College London symposia and workshops together with some addi- [email protected] tional plenary talks. Previous themes have included: Case for Support - not more than 1000 words, Managing Editor AISB Journal Communication, Interaction and Social Intelligence arguing your case for hosting the Convention. You Geraint Wiggins (2008), Artificial and Ambient Behaviour (2007), may put observations about your own background Goldsmiths College London [email protected] Adaptation in Artificial and Biological Systems (2006), and suitability in the Additional Comments section Editor, AISB Quarterly Social Intelligence and Interaction in Animals, Robots below. This case for support should include sug- Colin Johnson University of Kent and Agents (2005), Motion, Emotion and Cognition gestions of individuals you intend to approach as [email protected] (2004), Cognition in Machines and Animals (2003), plenary speakers and symposia organisers. Public Understanding Logic, Language and Learning (2002), Agents and of AI Cognition (2001), Artificial Intelligence and Society Convention Location, Time And Length - Tim Blackwell Goldsmiths College London (2000), Creativity (1999). Typically an AISB convention runs for 3-4 days in [email protected] March/April. If you are proposing to host a conven- Publicity Nick Hawes The convention will have a convention organiser tion of unusual length or at an unusual time then University of Birmingham who has has overall responsibility for the convention you should also include a justification of this change. [email protected] Schools Liaison program, local arrangements and financial manage- The location should be in the UK. Fiona McNeill ment. Program detail is mostly delegated to individual University of Edinburgh [email protected] symposium organisers but the convention organiser Additional Comments - no more than 500 is responsible for arranging plenary talks. words, on, for example, the relevance of your AISB Convention 2007 Christian Kray background to the convention, and of the benefits University of Newcastle- Each convention has a broad theme suggested of your proposed location. upon-Tyne [email protected] by the convention organiser within which the plenary AISB Convention 2008 talks and the majority of the symposia should fit. Bibliography - any literature references cited Frank Guerin University of Aberdeen This gives organiser the opportunity to promote their above. [email protected] research area. Ordinary Member Proposals will be selected by the Committee of the Slawomir Nasuto A full description of the role of the convention AISB. Unless there are very special circumstances, University of Reading [email protected] organiser is available on request from the AISB please do not expect us to consider web pages or secretary ([email protected]) and also from the other documents referenced by the proposal AISB Web Site TIMETABLE MAKING A PROPOSAL Convention proposal submission deadline: 1st July 2007 Proposals should be made by emailing in plain Notification of Acceptance: 5th August 2007 text to Louise Dennis at [email protected], en- Suggested deadline for Call for Symposia closing the following information. (Prior informal email Proposals: 31st July 2008 enquiries from possible proposers are welcomed):

No. 125, Spring 2007 11 Cognitive Divinity The Life of A. Hacker Programme Institute of Applied by Fr. Aloysius Hacker Epistemology

About the Society Episode 4: Computational Theology The Society for the Study of Artificial Intelligence and Simulation of Behaviour (AISB) is the UK’s largest and foremost The UK I returned to, at the start of the than John Lemon. The downside of this, otherwise Artificial Intelligence Swinging Sixties, was very different from the one welcome, publicity was to draw the attention of society. It is also one of the oldest-established I had left. It was alive to new thinking, especially an obscure but ancient, Italian organisation with a such organisations in the to the combination of “white hot” technological similar name to ours. To avoid expensive litigation, world. innovation with new forms of spirituality and the we were reluctantly forced to change our name to The Society has an spending of large sums of money. My latest en- the Church of God the Programmer. international membership of hundreds drawn from terprise, CATHOLIC™ (Church of Aloysius Theobald academia and industry. Hacker for Ordinations, Liturgy, Inquisitions and The central tenet of our new Church is that the Membership of AISB is open to anyone with Christenings), could hardly have come at a more Universe is a simulation, programmed by God, in interests in artificial opportune moment. The newly minted pop idols which Jesus was his avatar. Members of the flock intelligence and cognitive and computing sciences. provided me with both a source of high profile can communicate with God with PRAYER (Please acolytes and financial security. Thus freed from Reboot After Yet more Editing and Revision); God can AISB membership includes the following benefits: academic constraints, such as the need to publish intervene in the simulation with MIRACLES (Mysteri- and attract funding, I was able to give full rein ous Interventions in Reality, Altering Continuity by • Quarterly newsletter • The AISB Journal both to advancing a new Computational Theology Little Edits in the Simulation). These revolutionary • Student travel grants to and to perfecting the practical products proceeding ideas have underpinned our new Computational attend conferences • Discounted rates at from my new percipience. Theology. We have even used to them to make AISB events and advances in the sciences, such as cosmology. For conventions • Discounted rates on Despite the abstract ambitions of the ambas- instance, the apparent acceleration of the Universe’s various publications sadors of a new religion, the flock require concrete expansion can be explained without the need for • A weekly e-mail bulletin and web search engine realisations of their faith. Meeting this need resulted exotic dark energy, but merely by God’s coffee cup for AI-related events in my most successful technology transfer from the nudging the fast-forward button. and opportunities Cognitive Divinity Programme: the FETISH™ (Faith You can join the AISB Expounded, Theology Interpreted and Spirituality online via: Helped). The FETISH™ employed the latest Artificial http://www.aisb.org.uk Intelligence techniques to understand and answer all the user’s religious questions via a speech pro- cessing unit. It delivered uplifting sermons, listened to prayers, inspired virtuous living and encouraged generous giving, especially to CATHOLIC™. From the latest fashion accessory of the novo riche, it became the must-have Christmas gift of 1964, and soon took pride of place in an alcove of every lounge in AISBQ EDITORIAL the land --- it was, for a moment, more popular BOARD Joanna Bryson University of Bath Lola Canamero Want more say? University of Hertfordshire Simon Colton Imperial College London If you have lots of ideas about what we should have in the Quarterly, contact the Editor Yiannis Demiris about becoming an Editorial Board Member. Imperial College London Louise Dennis University of Nottingham Want to advertise? Colin Johnson U. Kent at Canterbury The AISBQ reaches hundreds of Artificial Intelligence and Cognitive Science researchers in Gary Jones the UK, Europe, and beyond. Advertising and general information is available via: University of Derby http://www.aisb.org.uk/aisbq/index.shtml Natalio Krasnogor University of Nottingham Kia Ng Want to write something? University of Leeds Patrick Olivier Univ. Newcastle upon Tyne The link above gives access to full guidelines for the submission of reviews and technical Frank Ritter articles. Books available for AISB members to review are also listed. Pennylvania State U. Barbara Webb The deadline for the next issue is: University of Edinburgh Geraint Wiggins Goldsmiths, U. London Friday 27th July 2007

12 AISB Quarterly