Thursday, January 26, 2006

EVA version 3 - Who's that girl ?

This is the new 3D design for EVA (Evolutionary Virtual Agent). There are many improvements since the previous one (see last post on this subject). Now the student team is working on the screen setup and on the various face expressions: reference (no expression), happy, disapointed, annoyed, angry, etc. There are three main animation parts: left eye, right eye, and nose/mouth. Since all images are pre-calculated, we try to reduce the number of animations to a small set of primitives. The Java-based animation engine is quite simple since it is basically an image switching system. This is not the case for the AI Ghost which is far more complicated. I will try to describe it briefly in a future post. I have a question: could you recognize the (real) woman who inspired this virtual model ?

Monday, January 23, 2006

A 3D Hierarchical Multi-Agent System for studying Cosmological Complexity

We have designed a hierarchical 3D multi-agent system for studying complexity at the cosmological level (see previous post on M31). Each agent is characterized by a set of properties: its position, velocity and mass, but also its diameter, density, viscosity, spin, etc. There are three simulation levels: (1) inside each agent (internal gravity that can lead to star formation, supernova, etc.), (2) local interactions between agents (viscosity) and (3) global interactions between agents (gravity and expansion). These two pictures (two different views of the same configuration) show an elliptical galaxy with an angular momentum. This dynamical structure has emerged from a random spatial distribution of 10,000 agents in the 3D space. Current thinking is that an elliptical galaxy is the result of a long process where two galaxies of comparable mass, of any type, collide and merge. This was not the case here...

Thursday, January 19, 2006

Complexity and Social Networks

Do you know that anyone on earth can be connected to any other person through a chain of acquaintances that has no more than five intermediaries ? (Six degrees of separation theory). Do you know that the size of a genuine social network is limited to about 150 ? (Dunbar Number).
What is the network model of my organization ? What are the emerging patterns and dynamics ? Is my company well adapted to its market ? How can I locate key players ? …
Social network theory views humans relationships as a network of agents connected through various acquaintances. The analysis of these networks is crucial for a better understanding in sociology, anthropology, organizational studies, economy and finance, etc. Within these domains, Internet plays a central role since online social networks (forums, chats, blogs, wikis, online games, etc.) are becoming increasingly popular. We can model such networks of interacting entities using classical approaches in the science of complexity: cellular automata, random networks, multi-agent system, etc. However, it appears to me that a hierarchical modeling could constitute a better approach (more precisely a hierarchy of intricated small-world agent-based networks). Let me try to explain why. First, it seems obvious that only multiple levels can reflect the complexity of real cases. At the lower level there is always a communication medium. For online social networks, this physical network is a graph of connected computers and servers, portable devices, mobile phones, etc. On top of this physical network, there are many possible different levels: individuals but also organizations, companies, countries, nations, etc., and other ones like social, cultural communities or any other forms of communities of interest. My second argument is that "classical" models are simple and universalist, let say: easy to model, to code and to understand. However, even they provide very interesting results in terms of emerging global dynamics and trends, they are generally inadequate for modeling a specific case to obtain practical informations. I think that the study of hierarchical multi-agent systems for modeling (online) social networks represents a promising approach and could contribute to give useful results in many real-world applications.

Tuesday, January 17, 2006

EVA version 3 - Preliminary 3D model

Here is the first 3D model for EVA (Evolutionary Virtual Agent) based on the preliminary drawings done in December (see previous post on EVA). This is a draft design since the model is not yet completed and there is no texture, no light, etc., but it gives a good idea of what will be the look of the final rendered face. All the 3D design is done by a student team of the IIM school (International Institute of Multimedia) using the 3DS Max Sofware tool. At the same time, I have coded near 60% of the Java-based AI "ghost" that will animate the model. The next steps are: (1) complete the face model and (2) test a working ghost with reactive behaviors (no genetic programming). Then, we will add animations and more complex behaviors. More information later. Keep tuned.

Friday, January 13, 2006

The New Furby

For Christmas, I bought the New Furby which was released in France at the end of the year. It is an advanced toy considering its technology and its relatively low price (near 50€ in France and less than 40$ in the US). The Furby looks like an armless Greemlins Mogwai with a beak. Evidence is that the New Furby cannot be considered as a living creature. However what is the opinion of my expert (Lara, 5 years old):
- Is Furby living ?
- Of course yes !
- Why ?
- Because it moves and it speaks to me...
Ok, well done Tiger (Furby is a product of Hasbro Inc.'s Tiger Electronic division). The New Furby is powered by the Sensory Inc's RSC-4128 ship which is a multi-purpose (14 Mhz max) microprocessor: voice recognition, text-to-speach, various input-output sensors and actuators, DTMF output, etc. Its documentation is easily accessible on the Web, but there is no information on the operating software called "Emoto-tronics" designed for the Furby. Even if the New Furby is a successfull toy, it lacks in many points from my point of view: (1) It reacts slowly to most events (probably needs a more powerful CPU and/or realtime programming). (2) It always repeats the same behaviors (needs more randomness/a larger set of behaviors). (3) The efficiency of learning is not so clear.... Some people like Benjamin Stove (blog at eecue.com) are working on "hacking" the New Furby. I think it could be very interesting to be able to reprogram this amazing toy using for example a subsumption architecture, genetic programming, etc. Avis aux amateurs...

Tuesday, January 10, 2006

Conway's Glider and Minimal Life

The Game of Life is certainly the most familiar Cellular Automata. It was introduced by John Conway in 1970 and popularized by the Scientific American journal. The Glider, a simple oscillating pattern that moves in the life universe, is generally presented as an emerging structure that gives an idea of what could be a minimal living creature. Is the Glider a useful modelof a living system ?
The Glider is a configuration of five "on" (black) cells that moves diagonally every four cycles. It represents a good example of a self-reproducing loop but it is far more simple than the self-reproducing automata proposed by John von Neumann in his theory. The Glider is a coherent pattern in space-time that continuously reconstitutes itself in the meaning of Maturana and Varela's notion of Autopoiesis. In the figure, I have colored in blue the "off" cells that contribute to its structure, forming a sort of membrane. However, the glider cannot be considered as a full autopoietic system since it is very sensible to any perturbation: most interactions with other structures are destructive. I personnaly think that the glider is too simple to represent a convincing model for minimal life, but it gives us a feeling of what it could be.

Saturday, January 07, 2006

Strasbourg Automata Clock

This week I was invited to give a talk on the evolution of complexity at the "Faculté de Géographie et d'aménagement" (Louis Pasteur University) in Strasbourg (Thanks to Christiane Rolland-May and her team). It was also an opportunityto take a picture of its famous astronomical clock. The Strasbourg cathedral clock has a long history. The first clock was built in around 1352. It was replaced in 1547 but due to the religious turmoil of the time, the work was not completed until some thirty years later. Some modifications and work took place again in the 19th century.
The clock includes a collection of automata: a cherub turningan hour glass, another hitting a gong and a figurative representationof the four ages of man. It represents an important step in the history of artificial creatures before the Lumière century and the automata designed by Jacques de Vaucanson in France and the Jaquet Droz in Switzerland. I am finishing to write a reference book on the history of artificial creatures: near 800 pages and more than 200 illustrations ! More information soon...

Wednesday, January 04, 2006

New year in the desert

The weather was so cold in Paris that we decided to go in the south Tunisian desert for the last days of 2005 :)
This travel gives me the opportunity to observe (and take pictures) many forms of complex natural landform patterns. Sand dunes represent a good example of rich emergent patterns produced by the nonlinear interactions between a very high number of simple agents : sand particules. Sand dunes formation has one basic cause: the transport of sand grains by the wind. They generally start small and disorganized due to the large number of degrees of freedom of sand grains. Once sand begins to pile up, ripples and dunes form. The wind moves individual grains along the inclined windward surface until they reach the crestand then cascade down the steep leeward side. Dunes formation relies upon the weight and size of the sand grains in relation to the wind force. Fine sand particles tend to produce a strong relief, the dunes having a short wave length, while coarse sand produces rounded dunes with a long wave length.

The sand accumulates into ridges which originally lie parallel to the direction of the prevailing winds. Over time these ridges create more irregular arrangement of hills and hollows. Global patterns emerge through interactions and mergers between these forms. Crescent shaped wind blown dunes spawn new dunes from their downwind pointing horns. This could be viewed as some sort of a simple self-reproduction mechanism.