Papers by Olivier Georgeon
This paper introduces an original model to provide software agents and robots with the capacity o... more This paper introduces an original model to provide software agents and robots with the capacity of learning by interpreting regularities in their stream of sensorimotor experience rather than by exploiting data that would give them ontological information about a predefined domain. Specifically, this model pulls inspiration from: a) the movement of embodied cognition, b) the philosophy of knowledge, c) constructivist epistemology, and d) the theory of enaction. Respectively to these four influences: a) Our agents discover their environment through their body's active capacity of experimentation. b) They do not know their environment "as such" but only "as they can experience it". c) They construct knowledge from regularities of sensorimotor experience. d) They have some level of constitutive autonomy. Technically, this model differs from the traditional perception/cognition/action model in that it rests upon atomic sensorimotor experiences rather than separating percepts from actions. We present algorithms that implement this model, and we describe experiments to validate these algorithms. These experiments show that the agents exhibit a certain form of intelligence through their behaviors, as they construct proto-ontological knowledge of the phenomena that appear to them when they observe persistent possibilities of sensorimotor experiences in time and space. These results promote a theory of artificial intelligence without ontological data about a presupposed reality. An application includes a more robust way of creating robots capable of constructing their own knowledge and goals in the real world, which could be initially unknown to them and unmodeled by their designers.
We present a new method and tool for activity modeling through qualitative sequential data analys... more We present a new method and tool for activity modeling through qualitative sequential data analysis. In particular, we address the question of constructing a symbolic abstract representation of an activity from an activity trace. We use knowledge engineering techniques to help the analyst build an ontology of the activity, i.e., a set of symbols and hierarchical semantics that supports the construction of activity models. The ontology construction is pragmatic, evolutionist and driven by the analyst in accordance with their modeling goals and their research questions. Our tool helps the analyst define transformation rules to process the raw trace into abstract traces based on the ontology. The analyst visualizes the abstract traces and iteratively tests the ontology, the transformation rules, and the visualization format to confirm the models of activity. With this tool and this method, we found innovative ways to represent a car driving activity at different levels of abstraction f...
This presentation discusses a notion encountered across disciplines, and in different facets of h... more This presentation discusses a notion encountered across disciplines, and in different facets of human activity: autonomous activity. We engage it in an interdisciplinary way. We start by considering the reactions and behaviors of biological entities to biotechnological intervention. An attempt is made to characterize the degree of freedom of embryos & clones, which show openness to different outcomes when the epigenetic developmental landscape is factored in. We then consider the claim made in programming and artificial intelligence that automata could show self-directed behavior as to the determination of their step-wise decisions on courses of action. This question remains largely open and calls for some important qualifications. We try to make sense of the presence of claims of freedom in agency, first in common sense, then by ascribing developmental plasticity in biology and biotechnology, and in the mapping of programmed systems in the presence of environmental cues and self-re...
Frontiers in Artificial Intelligence, 2022
We discuss the influence upon the fields of robotics and AI of the manner one conceives the relat... more We discuss the influence upon the fields of robotics and AI of the manner one conceives the relationships between artificial agents' perception, cognition, and action. We shed some light upon a widespread paradigm we call the isolated perception paradigm that addresses perception as isolated from cognition and action. By mobilizing the resources of philosophy (phenomenology and epistemology) and cognitive sciences, and by drawing on recent approaches in AI, we explore what it could mean for robotics and AI to take distance from the isolated perception paradigm. We argue that such a renouncement opens interesting ways to explore the possibilities for designing artificial agents with intrinsic motivations and constitutive autonomy. We then propose Artificial Interactionism, our approach that escapes the isolated perception paradigm by drawing on the inversion of the interaction cycle. When the interaction cycle is inverted, input data are not percepts directly received from the en...
Terrorism studies have and continue to face conceptual and analytic challenges that stem from the... more Terrorism studies have and continue to face conceptual and analytic challenges that stem from the assumption that terrorism can be understood outside of its social and political context, as essentially a ‘state’ of being and/or set of personal qualities specific to the terrorist (Sageman, 2004; Taylor & Horgan, 2006). An under-explored alternative to this view is to see involvement in terrorism, at least in psychological terms, as a process rather than a state. One consequence of this is that we shift the focus away from individuals and their presumed psychological or moral qualities to an examination of process variables. These, by their nature, are more susceptible to change and thus form the basis of developing interventions. Interpreting these variables, such as changes in operational context or relationships between temporal events and individuals, requires tools capable of capturing time-sensitive semantic content. To date, there are few process-oriented tools and fewer analys...
Single Agents can Be constructivist too
Constructivist Foundations, 2013
We support Roesch and his co-authors’ theoretical stance on constructivist artificial agents, and... more We support Roesch and his co-authors’ theoretical stance on constructivist artificial agents, and wish to enrich their “exploration of the functional properties of interaction” with complementary results. By revisiting their experiments with an agent that we developed previously, we explore two issues that they deliberately left aside: autonomous intentionality and dynamic reutilization of knowledge by the agent. Our results reveal an alternative pathway to constructivism that addresses the central question of intentionality in a single agent from the very beginning of its design, suggesting that the property of distributed processing proposed by Roesch et al. is not essential to constructivism.
We present a novel approach to state space discretization for constructivist and reinforcement le... more We present a novel approach to state space discretization for constructivist and reinforcement learning. Constructivist and reinforcement learning approaches are often characterized by simple grids. The manner in which the state space is discretized is the source of many problems for both constructivist and reinforcement learning approaches. The problems can roughly be divided into two categories: (1) wiring too much domain information into the solution, and (2) requiring massive storage to represent the state space (such as Q-tables. The problems relate to (1) the non generality arising from wiring domain information into the solution, and (2) non scalability of the approach to useful domains involving high dimensional state spaces. Another important limitation is that high dimensional state spaces require a massive number of learning trials. We present a new approach that builds upon ideas from place cells and cognitive maps.
During the initial phase of cognitive development, infants exhibit amazing abilities to generate ... more During the initial phase of cognitive development, infants exhibit amazing abilities to generate novel behaviors in unfamiliar situations, and explore actively to learn the best while lacking extrinsic rewards from the environment. These abilities set them apart from even the most advanced autonomous robots. This work seeks to contribute to understand and replicate some of these abilities. We propose the Bottom-up hiErarchical sequential Learning algorithm with Constructivist pAradigm (BEL-CA) to design agents capable of learning autonomously and continuously through interactions. The algorithm implements no assumption about the semantics of input and output data. It does not rely upon a model of the world given a priori in the form of a set of states and transitions as well. Besides, we propose a toolkit to analyze the learning process at run time called GAIT (Generating and Analyzing Interaction Traces). We use GAIT to report and explain the detailed learning process and the struc...
This collection of papers was presented at the first annual international workshop on self- super... more This collection of papers was presented at the first annual international workshop on self- supervised learning (IWSSL2020) held in Cambridge, Massachusetts, between February 27 and February 28, 2020. They represent the state of the art in an expanding field of research that attempts to build systems that can learn without human intervention with little or no hard-wired domain knowledge, as would a new-born child or animal.
We designed an autonomous agent that discovers, learns, and exploits basic spatial regularities o... more We designed an autonomous agent that discovers, learns, and exploits basic spatial regularities of interaction with its environment. To do so, we propose implementing a persistence memory system that records bundles of “possibilities of interaction” afforded by objects in the environment, coupled with a local space memory system that represents the agent’s surrounding local space (inspired by the vertebrate’s tectum). An experiment in a simple simulated environment demonstrates how the agent performs multimodal integration of sensory stimuli, and allocates the origin of such stimuli to “phenomena” located in the external spatial environment. Such mechanisms open the way to implementing agents with minimal preconception of their environment, and to modeling intrinsic motivation in autonomous agents.
We propose a learning mechanism that allows an artificial agent to construct and exploit a repres... more We propose a learning mechanism that allows an artificial agent to construct and exploit a representation of its surrounding space with minimal preconceptions about its environment. This representation is based on a data structure that encodes possibilities of behaviors afforded by the current context. The behaviors are modeled in the form of sequences of interactions. Over time, the agent learns to associate sequences of interactions with the presence of certain elements of the environment in certain locations in the agent's surrounding space. The agent uses this emergent relation between objects and possibilities of interactions to construct and maintain a representation of the surrounding space based on sequences of interactions. Experiments show that efficiently learning object and interaction associations requires implementing a form of curiosity as an additional motivational principle of the agent. These mechanisms open the way to implementing agents that learn to generate...
This work demonstrates a mechanism that autonomously organizes an agent’s sequential behavior. Th... more This work demonstrates a mechanism that autonomously organizes an agent’s sequential behavior. The behavior organization is driven by pre-defined values associated with primitive behavioral patterns. The agent learns increasingly elaborated behaviors through its interactions with its environment. These learned behaviors are gradually organized in a hierarchy that reflects how the agent exploits the hierarchical regularities afforded by the environment. To an observer, the agent thus appears to exhibit basic selfmotivated, sensible, and learning behavior to fulfill its inborn predilections. As such, this work illustrates Piaget’s theories of early-stage developmental learning.
Nous presentons une methodologie et un outil pour analyser l'activite d'un operateur huma... more Nous presentons une methodologie et un outil pour analyser l'activite d'un operateur humain en interaction avec un dispositif technique complexe. L'activite est observee pour etre modelisee sous la forme d'une trace ayant une structure de graphe. La trace collectee est constituee initialement d'une succession de descripteurs d'evenements, lies par une relation de sequentialite. Elle est ensuite enri-chie selon un modele d'utilisation pour construire une representation de l'activite a differents ni-veaux d'abstraction. Cela permet de retrouver des signatures de schemas mentaux mis en ½uvre par l'operateur. Cette approche est utilisee pour la modelisation cognitive du conducteur automobile.
Learning by experiencing versus learning by registering
Constructivist Foundations, 2014
Agents that learn from perturbations of closed control loops are considered constructivist by vir... more Agents that learn from perturbations of closed control loops are considered constructivist by virtue of the fact that their input (the perturbation) does not convey ontological information about the environment. That is, they learn by actively experiencing their environment through interaction, as opposed to learning by registering directly input data characterizing the environment. Generalizing this idea, the notion of learning by experiencing provides a broader conceptual framework than cybernetic control theory for studying the double contingency problem, and may yield more progress in constructivist agent design.
In 1920, the mathematician and philosopher Alfred North Whitehead argued: “If we are to look for ... more In 1920, the mathematician and philosopher Alfred North Whitehead argued: “If we are to look for substance anywhere, I should find it in events which are in some sense the ultimate substance of nature” (Whitehead 1920, ‘The concept of nature’, p19). Whitehead called process of abstraction the process by which cognitive beings infer the existence of objects from regularities of events. The Whiteheadian process of abstraction precedes the distinction between the subject and the object. It is not intellectual but instinctive and immediate: objects are abstracted but do not require judgment nor intellectual synthesis. We design algorithms for artificial agents to perform Whiteheadian abstraction. In addressing this issue, we investigate the components (sensorimotor schemes, hierarchical sequence learning, spatial memory, ontologies) that need to be implemented to realize this process. This led us to create agents capable of rudimentary self-programming (an important feature for achievin...
We present autonomous agents that are designed without encoding strategies or knowledge of the en... more We present autonomous agents that are designed without encoding strategies or knowledge of the environment in the agent. The design approach focuses on the notion of sensorimotor patterns of interaction between the agent and the environment rather than separating perception from action. The agent’s motivational system is also interaction-centered in that the agent has inborn proclivities to enact certain sensorimotor patterns and to avoid others. Such motivations result in the agent autonomously discovering, learning, and exploiting regularities of interaction afforded by the environment, and constructing operative knowledge of the environment. Because such agents have no predefined goals, we propose a set of behavioral criteria to both judge and demonstrate the agents’ capacities, rather than performance measurement. A design platform based on NetLogo is presented. Results show that these agents demonstrate interesting behavioral properties such as hedonistic temperance, active per...
We propose a method to investigate differences of driving at a tactical level, between different ... more We propose a method to investigate differences of driving at a tactical level, between different categories of drivers. We analysed the naturalistic driving performance of 19 French drivers using an instrumented vehicle. We assessed their sensation seeking scores with the Zuckerman questionnaire. We observed a significant correlation between their sensation seeking score and their mean speed on motorway. We set up a method to model their tactical behaviour and investigate possible correlation between their sensation seeking score and their tendency to perform certain types of behaviour. We applied it to the study of lane changes on motorways. We could model two categories of lane changes but we show that they were not correlated with the sensation seeking score. Despite this negative first result, we are proposing an innovative approach for this kind of study.
Modern design of human/machine interfaces requires a better understanding of how operators contro... more Modern design of human/machine interfaces requires a better understanding of how operators control their interaction with machines. To understand these interactions, cognitive ergonomists seek to construct cognitive models of operators. These models generally depict operator activity as a process of information- collecting, computing, decision-making, and action. While this symbolic approach effectively describes formal reasoning, it becomes ambiguous when con- sidering an activity in which operators are physically involved, such as driving a car. Here, operators’ cognitive process accompanies their actions and can be equally viewed as a cause or as a consequence of their activity. Perception, cognition, and action can hardly be separated, because expectations drive perception, and the feeling of comprehension relies on possibilities of action. Where interaction and perception are so tightly coupled, we take inspiration from psychologists like Piaget, who have proposed to keep perce...
We introduce Radical Interactionism (RI), which extends Franklin et al.’s (2013) Cognitive Cycles... more We introduce Radical Interactionism (RI), which extends Franklin et al.’s (2013) Cognitive Cycles as Cognitive Atoms (CCCA) proposal in their discussion on conceptual commitments in cognitive models. Similar to the CCCA commitment, the RI commitment acknowledges the indivisibility of the perception-action cycle. However, it also reifies the perception-action cycle as sensorimotor interaction and uses it to replace the traditional notions of observation and action. This complies with constructivist epistemology, which suggests that knowledge of reality is constructed from regularities observed in sensorimotor experience. We use the LIDA cognitive architecture as an example to examine the implications of RI on cognitive models. We argue that RI permits selfprogramming and constitutive autonomy, which have been acknowledged as desirable cognitive capabilities in artificial agents.
This study follows the Radial Interactionism (RI) cognitive modeling paradigm introduced previous... more This study follows the Radial Interactionism (RI) cognitive modeling paradigm introduced previously by Georgeon and Aha (2013). An RI cognitive model uses sensorimotor interactions as primitives-—instead of observations and actions-—to represent Piagetian (1955) sensorimotor schemes. Constructivist epistemology suggests that sensorimotor schemes precede perception and knowledge of the external world. Accordingly, this paper presents a learning algorithm for an RI agent to construct observations, actions, and knowledge of rudimentary entities, from spatio- sequential regularities observed in the stream of sensorimotor interactions. Results show that the agent learns to categorize entities on the basis of the interactions that they afford, and appropriately enact sequences of interactions adapted to categories of entities. This model explains rudimentary goal construction by the fact that entities that afford desirable interactions become desirable destinations to reach.
Uploads
Papers by Olivier Georgeon