Intelligent and Cognitive Systems

by Rolf Pfeifer and Alois Knoll

The grand challenge is that of building intelligent and cognitive systems that are capable of acting in the real world and interacting with humans in natural ways.

A promising research area tackling this challenge is the development of so-called complete agents, that is, agents that are embodied and self-sufficient (ie, they can sustain themselves over extended periods of time), situated (ie, they can acquire information about the environment through their own sensory systems), and autonomous (ie, they function independently of external control). For 'intelligent and cognitive systems', participants in this thematic group recommend to focus research efforts within the framework of embodiment. In this framework, intelligence and cognition are properties that emerge as an agent interacts with its environment.

Task Group 5 workshop participants
Task Group 5 workshop participants Giulio Sandini and Olaf Spoorns learn about the artificial whisker system developed by Miriam Fend of the Artificial Intelligence Laboratory in Zurich in the IST-FET-funded Artificial Mouse Project (AMouse,

Proposed Research Themes
In order to meet the above long-term challenge, the following research themes need to be pursued:

1. Mind-Body Co-Development and Co-Evolution
In order to maximally exploit the design power of evolution and development, controllers and robot morphologies have to evolve simultaneously. The permanent interaction of the body of an agent with the environment during growth enables its 'mind' to develop. We need to understand and model developmental processes in nature and build embodied artificial systems that are inspired by them; embed them into evolutionary processes, and study interaction of physical and information processes during this development. These also require physical growth (see the following point).

2. Systems and Materials that can Physically Grow
Recent research strongly suggests that physical instantiation and materials play an essential role in system behaviour. Also, through growth, biological organisms can form highly complex morphological structures. In this respect there are promising starting points (eg, modular robotics, self-assembling materials, etc). Although it is not expected to have growable materials that match biological capacities available any time soon, some aspects of physical growth can be studied through advanced principles of autono-mous modular systems that optimally adapt to a task and the environment.

3. Morphological Computation
Shape and the materials (eg, muscles) perform important functions for an agent in real time, an idea that bypasses the concept of classical Turing computation architecture. Morphological computation here refers to processes based on shape (eg, molecules/DNA, modules of a modular robot, shape of a limb) and material properties (eg of the muscle-tendon system). The challenge will be to explicitly apply morphological concepts in theoretical and practical explorations of embodied artificial systems.

4. Design for Emergence
As behaviour is always the result of the interaction of an agent with the environment, behaviour is emergent, meaning that it cannot be understood (and designed) on the basis of the internal control program (or 'brain') only. The question then is: how can we design purposive (goal-directed) agents without destroying the emergent nature of their behaviour?

In each of these research areas real physical embodiment plays an essential role. Simulation work ('embodied agent simulations') could form an important part of the endeavour.

Participate in the online consultation of this report from 1 February to 31 March 2006 at

TG5 Coordinator:
Rolf Pfeifer, University of Zurich, Switzerland