Fabio Ruini’s blog

Dear Sir or Madam,

I’m writing this letter because I would like to apply for the PhD Studentship in Multi-Agent Modelling of Search and Rescue Behaviour, offered by the Centre for Interactive Intelligent Systems of the University of Plymouth.

I always been interested in all the issues related to computing and new technologies. And, like many other people, I’ve always been fascinated by the artificial intelligence field. Particularly because I can see the actual limitations of the technologies we currently use. Technologies capable to performs an incredible number of tasks, but frequently too much difficult to understand (and then to use) for non-expert people. In my opinion, a machine capable to perform in an “intelligent” way would be a machine easier to use, if we think in terms of human-machine interaction, with respect to the artifacts we’re used to interact with during our daily life, and also more efficient, due to the different philosophy on which it is based. This is why I believe we need to build “intelligent” machines: in order to achieve more from the technology we currently use, i.e. to extend the applicability’s range of the technologies surrounding us. And, important as well, I also believe that to realize a better understanding of the human intelligence become possible when we are able to recreate it inside a human-made (both physical or just virtual) artifact.

Apart from my general passion for the artificial intelligence field, my interest in Artificial Life and social modelling, more specifically, has born few years ago. Since its experimental character, the degree course I had in Italy – even though belonging to the economics field – has given me a strong background in the sciences of complexity, as well as it has really enforced my skills in computer sciences. The complexity-related topics studied made me discover the agent-based simulation’s world, which immediately captured my attention. Then, mixing the required classes with a lot of independently found readings, I started my journey into the neural networks and evolutionary computation’s world.

I absolutely agree with the point of view that an intelligent machine can only be built “starting from the bottom”. In fact, given its intrinsically difficult, I believe that the problem of building truly intelligent machines cannot be successfully tackled through the more widespread top-down approach. Every time we try to recreate a typical human behavior into an artificial artifact, we immediately notice the incredible complexity related to this task. The typical top-down approach requires, for some extents, a “superior knowledge” to be successfully carried on. But this superior knowledge, for matters related to “intelligence”, isn’t still available today. The “strong AI”’s supporters try to compensate for this problem, mainly using methodologies typical of the engineering science. What appears obvious, to me at least, is that an animal brain doesn’t works solving complicated systems of differential equations. When an animal makes a movement, surely it isn’t thinking to some strange geometrical configuration of its legs. The brain, simply, works in a very different way. So, the top-down approach can surely produce some interesting results, particularly from a “practical” point of view, but its potential is strictly limited.

A different kind of approach in building intelligent machines is needed. And the right approach, in my opinion, is the one typical of Artificial Life, which exploits the biological plausibility of neural networks, mixing it with the Darwinian metaphor of evolution. In fact, we don’t need to create something absolutely “new”. The nature already provides us with all the necessary insights. Our task, as scientists, is simply to carefully observe the nature’s products in order to identify the basic components of their behavior. Then, recreating these basic conditions in a simulation, we can be able to make the “complicated and intelligent behavior” evolve.

I can offer a relevant experience in this research field – the same as the Bursary I’m applying for – as I’m currently involved in a similar project with Professor Angelo Cangelosi. The work we’re carrying on aims to develop a neural network-based controller system for swarms of Micro-UAV (Unmanned Aerial Vehicles), providing them with the ability of autonomously navigate through a typical urban environment.

More in details, I’m developing a computer simulation (using C++ language as “engine” and the Qt framework for the graphical interface) that takes place into a two-dimensional scenario. The image that follows is a screenshot taken from the simulation.

[Hello, I'm a the attached screenshot... trust me!]

Swarms formed by four MAVs, each one of them starting from a different position, first of all have to learn how to move along the environment, developing obstacle-avoiding capability (the obstacles are constituted by walls, buildings, etc.). Then, the MAVs have to learn how to coordinate their behavior in order to simultaneously hit a particular target, placed somewhere in the simulated scenario, eventually exploiting the capability to use a self-developed language.

Each MAV’s controller system is represented by a feed-forward neural networks, that perceive sensorial information from the environment and in turn produce, as output, a certain behavior. At any moment, each MAV is aware of its position into the environment (perceived as its geographical coordinates), as well as the position of its teammates and of the target. The drones are also equipped with an ultrasonic sensor, which inform they about the distance from the nearest obstacles they’re facing, and a heat sensor, which inform they when the target is in the nearness. Again, MAVs have a given flight autonomy and they’re always informed about that, in order to trigger their speed (i.e., the fuel consumption) in the most appropriate way. The learning process is implemented through a genetic algorithm.

At last, I’d like to spend few more words about another work I carried out before to come here in Plymouth. In October 2006, when I was starting my degree thesis, I joined the LARAL (Laboratory of Autonomous Robotics and Artificial Life), at the ISTC-CNR in Rome, working with Professor Domenico Parisi. The work we carried out is focused on the role of “motivation” in the context of Artificial Life’s simulation. Our preliminary consideration was that, until now, in most of the scientific papers published in this field, no much attention has been paid in the interactions between the two kinds of levels in which behavior takes place: the “strategic level” (i.e., what to do in a given moment) and the “tactical level” (i.e., how to physically implement the chosen behavior). In fact, the typical Artificial Life’s simulation focus exclusively on the tactical level of behavior, following a metaphor that we could resume as: “one robot – one goal”.

In the simulation’s scenario we created, an organism usually lives simply looking for food, since the chosen fitness formulae provides a reproductive advantage to the organisms that eat more. Sometimes in the environment appears a predator, for which the only goal is to reach the organism. A capture provoke a great “fitness damage” to the organism, so when the predator is present the organism needs to ignore the sensorial perception related to food and to concentrate all of its cognitive resources in the escaping task.

We have obtained many interest results from our work. For example, analyzing the activation patterns of the neural networks’ hidden layers, we have been able to define a quantitative measure of “attention”, noticing that, the more a predator is near to the organism, the less the organisms pays attention to the food. A result expected, of course, but never studied in details before.

At now, an abstract of our work, entitled “Living in an environment containing both food and predators”, has been recently submitted to the referrers of WIVACE (4° Italian Workshop on Artificial Life and Evolutionary Computation) and will be hopefully presented at that workshop, which will takes place during the month of August. We’re currently writing a full paper too, in order to submit it to an international journal soon.

I conclude this letter underlining that I believe that the interdisciplinary nature of my studies could be very usefully exploited to do research in this field, allowing me to be not a merely computer programmer, but also giving me the opportunity for providing a little, but hopefully significant, contribution to this research field. I’ll be gratefully if you will give me the chance to pursuing this goal here in Plymouth.

Yours faithfully,