ABSTRACT. A 45-minute tutorial on analogy-making, analogical reasoning, and corresponding computational models thereof. Further topics addressed include concept blending and concept invention, as well as their relation to current discussions in cognitive science.

ABSTRACT. The notion of cognitive economics is being introduced in the work and the premises causing the appearance of this current in contemporary economic theories are given. There is proposed a generalization of cognitive economics conception on the basis of introducing to this domain some parts of intellectual systems in economics, models based on knowledge, soft calculations and knowledge management in economics. It is shown that creation of intellectual hybrid systems at the turn of these domains in cognitive economics leads to increase of cognitive potential of intellectual systems in economics, greater adaptability and a possibility of setting to mental user’s models.

ABSTRACT. This paper discusses how the study of a biologically inspired cognitive architecture based on conceptual spaces may offer new opportunities for cognitive theories of autism (Frith, 2003). Happé (1999) discusses the information processing style of autistic people; according to the review, this cognitive style is based on weak central coherence. In brief, autistic people are able to well perceive many detailed features of perceived objects but they lack the capability of perceiving global configurations; moreover they are unable to contextualize perception. Moreover, Frith and Happé (1999) hypothesize that people affected by autism may present a dysfunctional self-consciousness, lacking introspective capabilities. In brief, people affected by autism may be unable to develop a Theory of Other Mind and therefore they could be unable to develop a Theory of Own Mind (see also Happé 2003). Cohen (1994) proposed a computational models of weak central coherence by means of feed forwards neural networks with too many hidden units; McClelland (2000) propsed a model based on an excessively conjunctive form of neural coding. Gustafsson (1997), Gustafsson and Papliński (2004) modeled aspects of autism by means of SOM neural networks with abnormal inhibitory settings. Grossberg and Seidman (2006) proposed the iSTART neural model describing how different regions in the brain may interact in order to generate autistic symptoms. Kriete and Noelle (2015) proposed a computational model based on the Cross Talk Generalization model in the Leabra framework. Rosenberg, et al. (2015) proposed a model based on alterations in divisive normalization, i.e., in the balance of the excitation with the inhibition of the neurons of a neural network. Matessa (2008) proposed a model in the framework of ACT-R, modeling the underconnectivity between the declarative module and the other modules of the architecture. O'Laughlin and Thagard (2000) employ a theory of coherence based on constraint satisfaction in order to model weak central coherence in autism. Chella et al. (1997) developed a biologically inspired cognitive architecture aimed at modelling perceptive capabilities in a robot, organized in three computational areas. The subconceptual area is concerned with low level processing of perceptual data coming from the sensors. In the linguistic area, representation and processing are based on a knowledge representation system. In the conceptual area, data coming from the subconceptual area are organized in conceptual categories according to the notion of conceptual spaces (Gärdenfors, 2000). A conceptual space is a metric space in which each dimension is a perceived quality as space position and distance. A point represents a perceived entity, e.g., a chair, while the metric distance is a measure of perceived similarity of the corresponding entities. A Concept is represented by a region in which all the points considered as instances of that concept are located. The weak central coherence cognitive style may thus be modelled by means of a conceptual space with an excessive, redundant number of quality dimensions. A robot equipped with this abnormal space is able to represent every detail of perceived entities, but, due to the sparseness of the resulting space, the capability of representing concepts vanishes: e.g. it is able to store each instance of chairs seen in its operating life, while it is unable to represent the concept of “chair”. Chella et al. (2008) also developed an extended version of the cognitive architecture previously described aimed at modelling introspective capabilities in a robot. To model such introspective capabilities, the notion of higher-order conceptual space has been introduced: a point in this higher order space corresponds to a perceived agent together with its own first order conceptual space, i.e., the robot itself, a person, another robot with introspective capabilities. Therefore, the Theory of Other Mind and the Theory of Own Mind of the robot may be modeled by this extended formalism. A point corresponding to a perceived introspective robot or a person is linked with an estimate of the conceptual space of the robot, thus representing the Theory of Mind related with that robot or person. A point representing the robot itself is thus related with the Theory of Own Mind of the robot. In the case of abnormal conceptual space because of weak central coherence, the robot is unable of introspective capabilities: because of the excessive number of quality dimensions in the conceptual space, we can speculate that the higher order spaces would collapse to a first order space with an infinite number of dimensions. The robot is then unable to represent the Theory of Mind of the other introspective entities, including itself. In summary, a robot equipped with an abnormal, excessive conceptual space, may open new research directions in the study of cognitive theories of autism.

ABSTRACT. This lecture builds on Ron Sun’s influential Desiderata for Cognitive Architectures by focussing on the desirable attributes of an biologically-inspired cognitive architecture for an agent that has a capacity for autonomous development. Ten desiderata are proposed, dealing with various aspects of embodiment, value systems & motives, sensorimotor contingencies, perception, attention, action, memory, prospection, learning, and self-organization. These desiderata are motivated by studies in developmental psychology, cognitive neuroscience, and enactive cognitive science. All ten focus on the ultimate aspects of cognitive development — why that feature is necessary and what it enables — rather on than the proximate mechanisms by which they can be realized. As such, the desiderata are for the most part neutral regarding the paradigm of cognitive science — cognitivist or emergent — that is adopted when designing a cognitive architecture. Where some aspect of a given feature is specific to a paradigm, this is noted.

ABSTRACT. In this lecture I will focus on the epistemological role of the computational explanation in biologically inspired systems and architectures. Such aspect has impact on both the design phase of the computational systems and on the phase concerning the results interpretation. I will provide an overview of the main methodological approaches used for the design of BICA and will outline which kind of computational models have an explanatory role  w.r.t the natural cognitive system considered as source of inspiration, and which ones cannot be considered explanatory at all. In particular I will show that purely "functionalist" models and systems (i.e. based on the methodological approach known as “functionalism” and proposing a weak equivalence between biological/cognitive processes and AI procedures) are not good candidates for providing advances in the science of BICA. On the other hand, since a realistic strong equivalence between artificial models/systems and target natural system is not currently achievable, my claim is that the only way to make progress is based on the development of plausible "structural" models of our cognition able to couple some advantages of the functional perspective with some of the structural design constraints proposed by the so called “structural” approach. I will show that only models and systems designed with the proposed “functional- structural” coupling can be considered good “proxyies" of a given target biological system and can play an explanatory role about it. I will also show that such systems can be useful both to:

i) advance the science of BICA in terms of technological achievements

ii) play the role of “computational experiments” able to provide insights and results useful for refining of rethinking theoretical aspects concerning the biological system used as source of inspiration.

ABSTRACT. The term restoration refers to the renewal of psychological, physical and social resources that individuals use to meet the demands of everyday life, such as the ability to concentrate (Hartig, 2004; Hartig et al., 2011). Restoration is highly reliable on the environment and inadequate restoration can diminish the individual‘s cognitive capacity and even lead to health problems if prolonged (Hartig, 2004; Hartig et al., 2011; Kaplan and Kaplan, 1989).

To date, studies in the field looking at how restorative an environment is have relied on static photographic images (Hartig & Staats, 2006; Lindal & Hartig, 2013a, 2013c) or videos (Ulrich, et al., 1991; Karmanov & Hamel, 2008) of existing sites, presented in laboratory settings where confounding and external factors could be controlled. The disadvantage of these techniques is the lack of interaction between the participants and the environment. Another option is to examine environmental impact on restoration using field studies (e.g., Hartig et al., 1991; Hartig et al., 2003; Johansson et al., 2011), allowing participants to interact with the environment but raising the issue of internal validity as controling for counfounding variables is difficult (Hartig, 2011). 

Powerful and rapidly developing computer technologies offer alternative approaches for restoration studies. They open up for the creation and presentation of highly realistic, three-dimensional virtual environments in which almost every visual aspect of the physical environment can be precisely manipulated and confounding factors strictly controlled (de Kort & IJsselstein, 2006; Depledge, Stone, & Bird, 2011; Lindal & Hartig, 2013; Rohrmann & Bishop, 2002). They also allow study participants to interact with the environment and move freely within it. 

The present research tested restorative effect on participants‘ physiological reactivity (e.g. bloodpressure, heartrate etc.), subjective experience and cognitive abilities using virtual reality. The results and future work will be reported and discussed. In particular, certain limitations of the virtual reality such as mode of navigation will be discussed.

ABSTRACT. We discuss general principles of interaction in groups of robots based on models of social behavior. A number of the models and methods implementing various forms of social organization in groups of robots on the basis of community typification are offered. We consider such fundamental mechanisms as coalition formation, language communication and task distribution in collectives. Implementation of psychophysiological features on the basis of the mechanism of emotions and temperament lies at the heart of the model of an individual capable to socialization.

ABSTRACT. Nowadays various biologically inspired cognitive architectures lack a front-page level of human behavior modelling – models of human linguistic and social activity. Well-known problems in these research areas – problems of sign grounding and term formation – are avoided in BICA where central space is occupied by memory procession and decision making. In the lecture we propose sign based approach aimed to build such knowledge representation (the world model) that will consume these problems and put into operation the sign formation module in cognitive architectures. Elements in the world model –signs – are grounded on external environment signals within a neurophysiologically inspired model such as HTM and other corticomorphic models. We consider the psychological theory of activity to represent the world model and to construct the model of cognitive processes (planning, attention). Within this theory an agent’s behavior is considered to be carried out in the course of the so-called activity directed by a motive (significance of a needed object). This activity is comprised out of a set of actions. Each action is aimed at achieving specific goal and consists of automated operations. Combination of operations forming an action is dependent on the observed conditions. While modeling the cognitive process actions and operations are formalized with some formal procedures. We assume that the separation of the available procedures to action and operation sets is dynamic and can vary during the process. In any case we attribute the goal oriented actions to be symbolic procedures forming the symbolic level of the behavior and automated operations to be subsymbolic ones (subsymbolic level). Within our assumption at a certain stage of the cognitive process some operations can be raised to the symbolic level and contrariwise some actions can be automated.

ABSTRACT. Biologically inspired cognitive architectures can adopt distinct levels of
abstraction, from symbolic theories to neural implementations.  Despite or
perhaps because of those widely different approaches, they can constrain and
benefit from each other in multiple ways (Jilk et al, 2008).

The first type of synergy, called the reductionist approach, occurs when a
higher-level theory is implemented in terms of lower-level mechanisms,
bringing implementational constraints to bear on functional abstractions.
As an illustration, we describe how the ACT-RN neural network implementation
(Lebiere & Anderson, 1993) constrained future developments of the ACT-R
production system cognitive architecture in biologically plausible
directions (Lebiere & Anderson, 1998).  Future efforts in this direction
demonstrated how the functional constraints that guided the earlier effort
could be further to a detailed biology of neural structures such as the
basal ganglia (Stocco et al, 2010).

The second type of synergy, called the hybrid approach, occurs when
cognitive architectures at distinct levels are combined, leading to
capabilities that woudn¹t be readily available in either modeling paradigm
in isolation.  A prerequisite for this type of integration is a compatible
mapping between frameworks of different levels.  The ACT-R cognitive
architecture and the Leabra neural architecture have converged on such a
mapping despite their opposite origins through a combination of functional
and neural constraints.  The SAL hybrid architecture, a Synthesis of the
ACT-R cognitive architecture and the Leabra neural architecture, provides an
illustration of the benfits of the hybrid approach through its combination
of high-level control and low-level perception (Vinokurov et al, 2011, 2012;
Szabado et al, 2013).

The third type of synergy, called the constructivist approach, results when
the same task or phenomena are modeled at different levels, bringing
insights and constraints across levels.  A necessary condition for this
approach is a close mapping between mechanisms at the different levels.  The
ACT-R architecture has integrated abstractions of neural mechanisms in its
subsymbolic layer including partial matching (Lebiere & Anderson, 1994),
blending (Lebiere, 1999) and a new associative learning mechanism (Thomson &
Lebiere, 2013).   Models of sensemaking processes developed in both ACT-R
and Leabra illustrate the deep correspondence between mechanisms at the
symbolic, subsymbolic and neural levels.

Finally, we conclude with a brief reflection of further directions in
neurally-inspired cognitive architectures, including new representational
directions.