[ CogSci Summaries home | UP | email ]

Vernon, D., Metta, G. & Sandini, G. (2007). A Survey of artificial cognitive systems: Implications for the autonomous development of mental capabilities in computational agents. IEEE Transactions on evolutionary computation, 11(2), 151-180.

  author = 	 {David Vernon and Giorgio Metta and Giulio Sandini},
  title = 	 {A Survey of artificial cognitive systems: Implications for the 
autonomous development of mental capabilities in computational agents. IEEE Transactions on evolutionary computation},
  journal = 	 {IEEE Transactions on evolutionary computation},
  year = 	 {2007},
  volume = 	 {11},
  pages = 	 {151--180}

Author of the summary: Kathy Van Benthem, 2009, kvbenthe@connect.carleton.ca

Cite this paper for:

The actual paper can be found in "IEEE Transactions on evolutionary computation, 11(2), 151-180."

Definition of Cognition [p151]

Cognition is an integrated process which "achieves robust adaptive, anticipatory, autonomous behavior, entailing embodied perception and action." [p151]
Embodiment is key to cognition, with strong ties found between perception, action and cognition. Intention, emotion and action are key grounding components of cognition.

Table 1, The Cognitivist vs. Emergent Paradigms of Cognition [p153]:

Each paradigm is compared according to key characteristics:

Description of cognitivist systems: Cognitivist systems can also be described as physical symbol information-processing systems. [p153]
See Figure 1: The Essence of a Physical Symbol System [p 154] Cognitivist systems describe behaviors as consequences of cognitive code operations. [p152]
Cognitivist systems are limited by human input such as bayesian networks (with predetermined probabilities) and by semantic structures. [p154-5]
Re: artificial intelligence- cognitivist systems utilize physical symbol processing and heuristic search methods for problem-solving. [p154]
A cognitive visual system which interprets traffic video input is an example of a cognitivist-type system.

Description of emergent systems: Emergent systems include the concept of co-determinism- whereby the agent acquires sensory information from the environment in order to construct its reality and to effect itself upon the environment using a reflexive and real-time process.[p155]
Three sub-types of emergent systems are:

  1. connectionist- networks based on parallel distributed processing [p157]
  2. dynamical- more theory than practice, whereby open systems use non-linear methods to construct real-time "realities" of itself and the enviroment- used more for system analysis then AI [p157-159] and,
  3. enactive models-autopoietic systems driven by self-maintenance. [p159].
    See page 159 for a full description of the three levels of autopoietic systems.
Bickhard's autonomy and stability theories outlined [p160]:
Two types of autonomy exist: 1) self-maintenance- where agents contribute to their persistance and,
2) recursive self-maintenance- where agents also contribute to the conditions of their persistance as per environmental shifts.
Two types of stability exist: 1) energy well stability (closed system with thermodynamic equilibrium) and 2) far from equilibrium stability (open system with no thermodynamic equilibrium)

Description of hybrid systems: Hybrid models or systems may use representation of percepts but operate without the constraints of a priori programmer knowledge systems.

Cognitive Architectures

Cognitivist architectures are the a priori "representational assumptions, the characteristics of its memories, and the processes that operate on those memories." [p162] Emergent architecture are structures for the "perception, action, adaptation, anticipation, and motivation that enable the ontogenetic development over the system’s lifetime." [p162]

Table 2, Cognitive Architectures Reviewed... [p163]:

Examples of Cognitivist Architectures [p163-6]

Examples of Emergent Architectures [p166-9]

Examples of Hybrid Architectures [p169-171]

Table 3, Cognitive Architectures vis a vis seven of the twelve characteristics... [p172]:

Table quickly demonstrates which architectures exhibit the desirable characteristics of autonomy. SASE And DARWIN exhibit the most with all characteristics represented either strongly or weakly.

The Developmental Stance [p172-5]:

See Figure 7: Maturana & Varela's ideograms to denote autopoietic and operationally closed systems.. (1987)
Architecture's phylogenetic structures must support motivation and structures for development. Similar to Kelsova(1995). Co-determinism and self-organization capabilities are important features of development. Learning (von Hofsten, 2004) [p173] and exploration are key elements of development. Metzoff & Moore (2002) suggest imitation is key to learning- with four phases: 1) body babbling, 2) imitation of body movements, 3) imitation of actions on objects and 4) imitation based on inferences re: other;s intentions.
Winograd & Flores (1986) suggest that learning is not the increased databank of representations from the environment, but is the continued refinement of behaviour based on satisfying internal or external demands. This leads to co-dependency of perception and actions...."mirror neurons" [p175].

Recommended Key features of autonomous cognitive systems [p175-6]:

Based on Kirchmar et al (2005) design principles for autonomous development of systems.

Summary author's notes:

Back to the Cognitive Science Summaries homepage
Cognitive Science Summaries Webmaster:
JimDavies (jim@jimdavies.org)