E. Tzafestas Compromising algorithmicity and plasticity in autonomous agent control architectures.: The autonomous cell Malgré leur efficacité, les architectures algorithmiques de con­ trôle comportemental présentent une rigidité structurale qui se manifeste comme susceptibilité aux pannes. À l'opposé, la redon­ dance inhérente dans la plupart des architectures connexionnistes permet une auto-organisation continue et une récupération des pannes de petite échelle. Dans cet article, nous explorons la re­ lation entre algorithmique et plasticité et nous tentons un com­ promis en élargissant un modèle cellulaire algorithmique de base. En premier lieu, nous introduisons la notion de motivation.: la cellule devient un système de plusieurs pulsions indépendantes capables de reconnaître et de consommer différents types de mes­ sages. Ces pulsions sont en compétition pour la consommation des ressources cellulaires. Les messages peuvent être consommables ou catalytiques et possèdent des templates d'identification qui ex­ citent les pulsions correspondantes. Les templates sont exprimées par des variables numériques continues, c'est-à-dire par des fréquences propres. À la différence des modèles connexionnistes habituels, il n'y a pas de connexions entre cellules, mais des buffers de messages partagés par les cellules d'un même niveau.; les cellules flottent alors dans un milieu commun d'interactions. Ensuite vient la socialité. Pour assurer l'opérationalité algo­ rithmique de l'agent cellulaire, les cellules doivent suivre les 'standards sociaux' de la population. Pour ce faire, nous avons introduit un mécanisme développemental d'adaptation des fréquences propres et un système immunitaire parallèle, c'est-à-dire une deuxième population de cellules chargées de re­ connaître et d'éliminer les messages nuisant potentiellement à l'intégrité de l'agent cellulaire. Les deux populations de cel­ lules ont des fréquences propres complémentaires. Le potentiel auto-organisationnel du modèle est illustré dans le cas d'un système de navigation. Nos résultats de simulation montrent que, d'une part, le réseau cellulaire peut 'découvrir'des voies alter­ natives de flot de messages et donc tolérer divers types de pannes imprévues et, d'autre part, que plusieurs pannes ralentis­ sent la réactivité du système aux événements extérieurs. Nous discutons également des questions telles que la sélectivité et le rôle de la diversité. While algorithmic autonomous agent control architectures demon­ strate high efficiency, they suffer from network structure rigid­ ity that shows in the liability to crucial errors. On the other hand, the redundancy inherent in most connectionist architectures allows for continuous self-organization that compensates for lim­ ited scale local neuron failures. In this article, we are inves­ tigating the relation between algorithmicity and plasticity and attempt a compromise by extending a basic algorithmic cell model. At first, we introduce motivation.: the cell becomes a system of multiple independent drives capable of recognizing and consuming different kinds of messages and in competition with one another for the use of the cell resources. Messages may be consumable or catalytic and have identification templates that excite the cor­ responding cell drives. Those templates are expressed as continu­ ous numeric variables, thus as eigenfrequencies. Unlike usual connectionist models, there are no connections between cells, but message buffers shared by all cells of a level.; this way, cells can be viewed as floating in a common interaction medium. Next to motivation, comes sociality. To ensure the cellular agent's algo­ rithmic operationality, cells should follow the 'social stan­ dards' of the population. This is achieved via a developmental mechanism of cell eigenfrequency adaptation and the introduction of a parallel immune system, i.e. a population of cells that rec­ ognize and eliminate the messages that might be detrimental to the integrity of the cellular agent. The two populations of cells have complementary eigenfrequencies. The model's self- organizational potential is illustrated on the example case of a navigation system. Our simulation results show that the cellular network is recover from various types of failures by 'discover­ ing'alternative message flow pathways and that multiple failures slow down the system's responsiveness to external events. Issues such as selectivity and the role of diversity are also discussed. While algorithmic autonomous agent control architectures demon­ strate high efficiency, they suffer from network structure rigid­ ity that shows in the liability to crucial errors. On the other hand, the redundancy inherent in most connectionist architectures allows for continuous self-organization that compensates for lim­ ited scale local neuron failures. In this article, we are inves­ tigating the relation between algorithmicity and plasticity and attempt a compromise by extending a basic algorithmic cell model. At first, we introduce motivation.: the cell becomes a system of multiple independent drives capable of recognizing and consuming different kinds of messages and in competition with one another for the use of the cell resources. Messages may be consumable or catalytic and have identification templates that excite the cor­ responding cell drives. Those templates are expressed as continu­ ous numeric variables, thus as eigenfrequencies. Unlike usual connectionist models, there are no connections between cells, but message buffers shared by all cells of a level.; this way, cells can be viewed as floating in a common interaction medium. Next to motivation, comes sociality. To ensure the cellular agent's algo­ rithmic operationality, cells should follow the 'social stan­ dards' of the population. This is achieved via a developmental mechanism of cell eigenfrequency adaptation and the introduction of a parallel immune system, i.e. a population of cells that rec­ ognize and eliminate the messages that might be detrimental to the integrity of the cellular agent. The two populations of cells have complementary eigenfrequencies. The model's self- organizational potential is illustrated on the example case of a navigation system. Our simulation results show that the cellular network is recover from various types of failures by 'discover­ ing' alternative message flow pathways and that multiple failures slow down the system's responsiveness to external events. Issues such as selectivity and the role of diversity are also discussed.