I'm attending AISB09 in Edinburgh, talking about "Adaptive and Emergent Behaviour and Complex Systems"
I started in the Social Networks and Multi Agent Systems symposium, with "Leon van der Torre and Serena Villata. Four Ways to Change Coalitions: Agents, Dependencies, Norms and Internal Dynamics". They're looking at the way coalitions change, in groups of agents represented using dependence networks. They give four types of change: agents, their dependencies, the addition of norms and internal coalition processes, and then talk about metrics for stability.
I moved across to the Swarm and Evolutionary Algorithms session, for a couple of quite different presentations. Azevedo F, Vale Z. "A Long-term Swarm Intelligence Hedging Tool Applied to Electricity Markets" talked about the difficulty of predicting price in the electricity market after it's decentralisation/deverticalisation. Prices are negotiated 24 times a day (for "spot" selling), with huge price peaks around unexpected demands (c.f. inelastic!). There is then a large market around selling contracts to supply at a certain price at a certain time (forward selling), and also financial trading (options). They're using a particle swarm to optimise a mean variance utility function, given a risk aversion parameter, and find that risk aversion drastically reduces the expected profit, and also reduces "spot" sales (preferring forward and options trades). (There is regulation in certain markets to guarantee a certain level of spot trading...).
Again in the Swarm Intelligence section, Hashizume Y, Nishii J. "Intermediate selection pressure bring in the emergence of altruism and common words" created a grid based world of agents, food and predators. Agents lost health by sharing cells with predators and metabolic processes, gained it by eating food, and could also use some to emit "words". Shouting is altruistic - it cost energy, and may benefit surrounding agents if they then either move towards to food or away from the predator - there is also an idea of learning, as agents learn to associate certain words (4bit binary) with situations, so the evolution of the use of language can be tracked. Theyu found that language was used most when there was an intermediate level of selection pressure - i.e. it's hard to survive, but there is still some energy left over.
There was some good discussion after this about the use of heterogenous swarms - using different kinds of particles in swarms, (like coloured ants!) in different ways. Forces between particles are traditionally attriactive, but what about using repulsive ones to avoid getting stuck, or even more divergent behaviours. This led on to wondering about more symbiotic or predator prey relationships, and how far the biological inspiration could be taken.
Back to SNAMAS, for another pair of quite different presentations... Ugo Pagallo talked about the small world network in the context of law (Ugo Pagallo and Giancarlo Ruffo. The Paradox of Elegance - A Very Short Introduction to the Topology of Complex Social Systems and The "Small World" - Paradigm in the Realm of Law). He started by saying (semi-joking) that Kandinsky (a lawyer) invented the concept of small world networks with his paintings of the same name. This led on to Milgram (6 degrees of separation) and Granovetter (strength of weak ties). Looking at law, we can model cases as nodes, with citations as links between them. If we do this for the US legal system (see Chandler) we get a low density network with a Weibull distribution. Looking at the "semantic hubs" in this, (see Fowler and Jeon: [] and other good stuff at http://jhfowler.ucsd.edu/) we see that Freedom of Speech is the core of the network with an internal density of links 500 times higher than the network average. This network density (hubs) is a better predictor of importance than expert judgement and statistics. We can also use the change in authority of the hubs over time to understand the way the legal system changes.
The last "serious" talk I went to was Patrice Claire and Leon van der Torre: "The Design of Convivial Multiagent Systems". The first issue tackled was the definition of conviviality, which stayed fuzzy, but included:
- Illich: "individual freedom realized in personal interdependence"
- Reciprocity (performinc actions in return, mutual dependence)
- Conviviality mask: "masks the power relationships and social structures that govern
societies." (Taylor 04 "Oh no it isn’t")
They then talked about conviviality measures in the context: i*/Tropos (Yu 95) --creates-> Dependence Networks (Conte/Schuman 02) --identifies-> Coalitions/Cycles (Sauro 05) --provides-> Conviviality Measures (current talk)
They discussed a succession of extensions to dependence networks:
- dynamic DN
- temporal DN
- epistemic DN
- normative DN (where normative links are added to improve conviviality)
As a fun end to the day, Aldebaran Robotics demoed their Nao robots. You can find the specs on their website, but in short they've got wireless, two cameras (but no stereo vision), four mics (for voice location etc) and joint precision to 0.1 degree. They're pretty small (about 45 cm) and weigh about 5kg. And their eyes light up.
They have a good sense of character, looking around, sitting down or stretching when bored...
My favourite points were:
- you can use one robot to contol others, by transmitting power wirelessly
- the tai-chi demo, where they shifted weight to one leg and slid the other over the floor
- the software (Choreographie) looks lovely
If they didn't cost €10k, I'd get one...
Finally, the day closed with Ken McCloud's talk. It was a good overview of AI in films, but stopped short of all the really interesting stuff, e.g. he mentioned the Singularity, but didn't really go into what it might mean for us.