Content uploaded by Antonio Lieto
Author content
All content in this area was uploaded by Antonio Lieto on Sep 25, 2018
Content may be subject to copyright.
This space is reserved for the Procedia header, do not use it
Higher-level Knowledge, Rational and Social Levels
Constraints of the Common Model of the Mind
Antonio Lieto1, William G. Kennedy2, Christian Lebiere3, Oscar J. Romero4,
Niels Taatgen5, and Robert L. West6
1University of Turin, Dept. of Computer Science, and ICAR-CNR (Palermo), Italy
lieto@di.unito.it
2Center for Social Complexity, George Mason University, USA
wkennedy@gmu.edu
3Carnegie Mellon University, Dept. of Psychology, USA
cl@cmu.edu
4Carnegie Mellon University, Dept. of Machine Learning, USA
oromero@cmu.edu
5AI and Cognitive Engineering Institute, University of Groningen, The Netherlands
n.a.taatgen@rug.nl
6Institute of Cognitive Science, Carleton University, Canada
robert.west@carleton.ca
Abstract
We present the input to the discussion about the computational framework known as Common
Model of Cognition (CMC) from the working group dealing with the knowledge/rational/social
levels. In particular, we present a list of the higher level constraints that should be addressed
within such a general framework.
Keywords: Unified Cognitive Architectures, Computational Models of Cognition, Common Model of
the Mind, Cognitive Constraints, Knowledge Level, Rational Level, Social Level
1 Introduction
In his famous 1982 paper, Allen Newell [23, 24] introduced the notion of knowledge level to
indicate a level of analysis, and prediction, of the rational behavior of a cognitive artificial agent.
This analysis concerns the investigation about the availability of the agent knowledge, in order
to pursue its own goals, and is based on the so-called Rationality Principle (an assumption
according to which “an agent will use the knowledge it has of its environment to achieve its
goals” [23, p. 17]. By using the Newell’s own words: “To treat a system at the knowledge level
is to treat it as having some knowledge, some goals, and believing it will do whatever is within
its power to attain its goals, in so far as its knowledge indicates” [23, p. 13].
1
Higher-level Constraints of the CMC Lieto, Kennnedy, Lebiere, Romero, Taatgen, West
In the last decades, the importance of the knowledge level has been historically and system-
atically downsized by the research area in cognitive architectures (CAs), whose interests have
been mainly focused on the analysis and the development of mechanisms and the processes
governing human and (artificial) cognition. The knowledge level in CAs, however, represents
a crucial level of analysis for the development of such artificial general systems and therefore
deserves greater research attention [18]. In the following, we will discuss areas of broad agree-
ment and outline the main problematic aspects that should be faced within a Common Model
of Cognition [13]. Such aspects, departing from an analysis at the knowledge level, also clearly
impact both lower (e.g. representational) and higher (e.g. social) levels.
2 Areas of Agreement
The analysis at the knowledge level is directly involved in, at least, three of the four dimensions
considered within the Standard Model of the Mind [13] (later renamed Common Model of Cog-
nition – CMC). In particular, it concerns the issues related to: i) the Structure and Processing
mechanisms, ii) the Memory and Content of the CMC, as well as iii) its Learning processes.
Concerning the first element, there is an agreement about the architectural necessity regarding
the distinction between a Long-Term Declarative Memory and a Procedural one, as well as the
necessity of a working memory module operating as a control interface between the Procedural
module and other modules such as Declarative Memory and the Perception/Motor modules.
Also, the cognitive cycle assumption [13] (with both serial and parallel information processing
mechanisms between/within modules), seems perfectly compatible with the above mentioned
Rationality Principle through which it is possible to evaluate the agent intelligent behavior.
For what concerns the Memory and Content issues, the integration of hybrid symbolic-
subsymbolic representations and processing - and the inclusion of relevant metadata like fre-
quency, recency, similarity, activation etc - represents the main element of difference with respect
to the classical early symbolic CAs. The fact that such integration is necessary, in order to
build integrated intelligent agents able to interact in the real world, is widely accepted. Some
issues concern the way in which such integration can be obtained and novel solutions to address
some representational problems of the Declarative Memories have been proposed and will be
discussed in the sections below.
Finally, for what concerns the learning part, the facts that: i) all types of long-term knowl-
edge are learnable, ii) learning is an incremental processes typically based on some form of
a backward flow of information through internal representations of past experiences and, iii)
learning over longer time scales is assumed to arise from the accumulation of learning over
short-term experiences, also seem to be accepted constraint elements. These elements are also
explicitly grounded in, and compliant with, the Anderson’s Decomposition Thesis [1]1based on
the schema between different time-scales, types of operations and bands of cognition proposed
by Allen Newell.
Figure 1 reproduces an extended version of the original four bands of cognition schema
proposed by Newell [24] including the Biological, the Cognitive, the Rational, and the Social
band. In the Newell framework, each band captures different types of human experience and
represents different types of information processing mechanisms required to describe the levels
within them. In particular: the neural band is described in terms of cellular biology, the cogni-
1According to such thesis learning at the highest Band (the social one) can be reduced to learning occurring at
lower bands. In general this thesis suggests that there is good evidence that high level tasks can be decomposed
and understood at the micro-cognitive level, and that improvements at the micro-cognitive level can create
improvements as measured at higher levels.
2
Higher-level Constraints of the CMC Lieto, Kennnedy, Lebiere, Romero, Taatgen, West
Figure 1: Extended version of the Newell’s Time Scales and the different Action Bands.(Green
is Newell’s Figure 3.3, blue is Newell’s Figure 3.14, black added by W.G. Kennedy).
tive band in terms of symbolic information processing, the rational band in terms of knowledge,
reasoning and goals, and the social band in terms of distributed, multi-agent processing. The
elements discussed so far in the CMC are, as for the entire enterprise of the cognitive archi-
tectures, mainly focused on the deliberate act level of the Cognitive band. In the following we
provide additional elements of discussion for what concern both the Rational and the Social
Band. Discussions about the lower band (e.g. the Biological one) are out of the scope of the
present contribution.
3 Rational Band
Newell equates the rational band with the knowledge level. The knowledge level refers to the
level at which knowledge becomes abstract and can be treated largely independently from the
physical systems that process it. Currently, the position according to which there is no need
of including, in the CMC, specialized architectural modules to perform activities belonging to
the rational band (e.g. planning, language processing and Theory of Mind) is ma joritarian.
The underlying assumption is that all such activities should arise based on the composition of
processes executed during different cognitive cycles according to specific computational mod-
els. Despite this position, however, few cognitive architectures have formulated computational
approaches to this effect, in particular to the Theory of Mind (ToM). Among these few works,
there is that one by [27] where the SIGMA cognitive architecture [33] is used to demonstrate
3
Higher-level Constraints of the CMC Lieto, Kennnedy, Lebiere, Romero, Taatgen, West
two distinct mechanisms (automatic processing vs. controlled reasoning) for ToM using as an
example several single-stage simultaneous-move games, in particular, the well-known Prisoners
Dilemma. Authors left open the possibility of using SIGMA’s learning capability to allow the
agents to learn models of each other in a repeated game setting. ACT-R [2] has also been used
to build several models of false belief and second-order false belief task (answering questions of
the kind ‘Where does Ayla think Murat will look for chocolate?’) to assess whether children
have a ToM [42]. Additionally, in [40] several scenarios were set up using ACT-R to show how
ToM ability can improve the quality of interaction between a robot and a human by predicting
what a person will do in different situations; e.g., that a person may forget something and may
need to be reminded or that a person cannot see everything the robot sees.
Despite such efforts, however, the modeling attempts of such rational aspects remains still
limited (while there are many more models developed for other phenomena concerning, for
example, planning, or natural language processing etc.). An alternative proposal [44] to the
view of creating specific computational models for cognitive phenomena occurring at the rational
band, suggests to adopt additional schema to organize the activity at the higher bands to
complement those proposed in Newell [24]. Such proposal will be discussed in more detail in
the next sections since it also affects the modeling of phenomena occurring not only at the
Rational band but also at the Social one and beyond. In particular: in the section entitled
“Knowledge Problems at the Rational Band”, we will focus on some current limitations of the
knowledge level of CAs and will show how this level of analysis also suggests some reflections for
the CMC concerning the underlying representational assumptions to adopt/integrate within the
Declarative Memory of cognitive agents. In the following section, instead, we introduce some
of the main issues concerning the Social Band.
4 Social Band
Events happening at the Social Band can take place over longer time scales (days, weeks,
and months). However, they are composed of social events happening at much shorter time
scales (seconds, minutes, and hours) and, as evidenced by Anderson [1] may be supported by
cognitive and rational processing in individuals at lower time scales. To discuss the potentially
controversial topic of motivations, some of our needs, such as physiological needs, come from
the bottom up and may explain why we developed high level cognition, i.e., the ability to solve
those problems. The social level can also result in goals that people are intrinsically motivated
to pursue and our cognition needs to be able to account for that. Our biological and social
needs provide goals and may have resulted in innate cognitive capabilities. For example, we
seem to have some innate capabilities associated with social cognition, such as perceiving other
people, processing direction of gaze, determining intentions of others, limitations on knowing
others as individuals (Dunbar’s Number), and the topic of Theory of Mind. Related areas such
as cooperation, trust, collective action are tasks or behaviors that arise at the social level. These
social interactions may influence the CMC at the lower levels, not simply above the rational
level.
Concerning the modeling of macro-scale or macro-cognition events there are, as mentioned,
two different perspectives currently debated in the literature. On the one hand such elements
are seen as too high level to be included within the minimal information processing mechanisms
of a general cognitive architecture and, as such, are left to specific computational models to
be developed on the top of such architectures. ACT-R, for example, has been used to study
social behaviors and distributed collective decision-making processes which must balance diverse
individual preferences with an expectation for collective unity. Romero and Lebiere [30, 31]
4
Higher-level Constraints of the CMC Lieto, Kennnedy, Lebiere, Romero, Taatgen, West
proposed a multi-agent approach where cognitive agents have to reach global consensus while
opposing tensions are generated by conflicting incentives, so agents have to decide whether
to follow the most influential agent, follow the majority, negotiate with others, come to an
agreement when conflicting interests are present, or keep a stubborn position.
Alternatively, the PolyScheme cognitive architecture [6] applies ToM to perspective taking in
the human-robot interaction scenario, that is, the robot can model the scene from the human’s
perspective and use this information to disambiguate the command when moving in a scenario
with multiple occluding elements [41]. However, although there are computational models
of social interactions in a practical sense, there are no current CAs supporting research in
social cognition, at best there are frameworks (as examples, BDI [28, 29] and PECS [43, 38]).
An alternative position with respect to the current majoritarian view is proposed in [44] and
[26]. These authors suggest the inclusion of macro-cognitive architectural elements that should
specify the information processing mechanisms allowing to determine complex behavior at both
the rational and social level bands and other constraints from the social band. These elements
will be discussed in the section ‘Problems for Macro-Cognition’.
5 Higher Bands
As Newell observed: “It is not clear that there actually are any higher bands”. We agree
that currently there is not evidence for a system level above the social level for cognition of
individuals or for the cognition of groups of people/agents. For additional discussion on this
point, we remind to the section 3.11 of Newell’s Unified Theories of Cognition [24].
6 Knowledge Problems at the Rational Band
From a knowledge processing perspective, one of the main problem concerning the knowledge
level of CAs is that, currently, the CAs are not able to deal with wide and complex knowledge
bases that can be, even slightly, comparable (for what concerns both the size and heterogeneity
of the handled knowledge2) to the amount of knowledge heuristically managed by humans.
The limited size of the knowledge bases processed by the cognitive architectures was already
acknowledged by Newell as a functional problem to address [24]. More recently, the content
limit has been newly pointed out in literature [25] and some solutions for filling this “knowledge
gap” have been proposed.
The ACT-R architecture, for example, has been semantically extended with an external
ontological content coming from three integrated semantic resources composed by the lexical
databases WordNet [21], FrameNet [23] and by a branch of the top level ontology DOLCE [20]
related to the event modeling. In this case (see [25]), the amount of semantic knowledge selected
to extend the ACT-R declarative memory only concerned the ontological knowledge about the
events. While this is a reasonable approach in an applied context, it still does not allow to
test the general cognitive mechanisms of a CA on general, multi faceted and multi-domain,
knowledge. Therefore it does not allow to evaluate, strictu sensu, to what extent the designed
heuristics allowing to retrieve and process, from a massive and composite knowledge base,
conceptual knowledge can be considered satisfactory with respect to the human performances.
2The heterogeneity issue concerns the problem of representing different types of conceptual knowledge in
a cognitive agent, including the common-sense one. Handling common-sense knowledge representation and
reasoning mechanisms, however, still represents an unsolved problem to deal with. This aspect is problematic not
only for the rational band processes but also for the information processing mechanisms occurring at Cognitive
Band (in particular those involving the operations and unit task levels, see Figure 1).
5
Higher-level Constraints of the CMC Lieto, Kennnedy, Lebiere, Romero, Taatgen, West
More recent works have tried to completely overcome the size problem of the knowledge level.
To this class of works belongs one proposed by Salvucci [34] aimed at enriching the knowledge
model of the Declarative Memory of ACT-R with a world-level knowledge base such as DBpe-
dia (i.e. the semantic version of Wikipedia represented in terms of ontological formalisms) and
a previous one proposed in [3] presenting an integration of the ACT-R Declarative and Pro-
cedural Memory with the Cyc ontology [15] (one of the widest ontological resources currently
available). Both the wide-coverage integrated ontological resources, however, represents concep-
tual information in terms of classical symbolic structures and encounter the standard problems
affecting this class of formalisms concerning the representation and reasoning on common-sense
knowledge. (see [18] for a detailed treatment of this aspect). With respect to the size problem,
the knowledge level is also problematic for the Soar [14] and the SIGMA [32, 33] CAs. Both
architectures, in fact, do not currently allow to endow agents with general knowledge. For Soar,
this problem is acknowledged in [14] but there is no available literature attesting progress in
this respect 3. A possible alternative solution that, in this perspective, is suitable to account for
both the size problem and typicality (or common-sense) effects in conceptualization has been
proposed in DUAL-PECCS [19]: a system that has been successfully employed to extend the
Declarative Memory of diverse CAs and that combine, on a large scale, both common-sense
representation and reasoning with standard ontological semantics. The main merit of such
proposal lies in the adoption of the representational component of Conceptual Spaces [9] inte-
grated with other neural-symbolic formalisms. The benefits coming from the integration of the
Conceptual Spaces framework as an intercommunication layer between different types of rep-
resentations in a general cognitive architecture has been recently pointed out in [17] and, with
respect to the CMC, it has been acknowledged both in [7] and [11]. Recently, within the rep-
resentation framework adopted in such system, it has been proposed a unifying categorization
algorithm able to reconcile all the different theories of typicality about conceptual reasoning
available in the psychological literature (i.e. prototypes, exemplars and the theory-theory, see
[16]). Another interesting approach that aims at synchronizing beliefs and truth values among
multiple domains to provide a unified treatment of these various forms of knowledge is described
in [35]. In general, going towards unified representations and reasoning procedures seems to be
a reasonable path to explore within the CMC research efforts.
Another important issue to consider in this dimension of analysis concerns whether the
CMC should make a distinction between the knowledge about the ‘self’ and knowledge about
‘others’. If so, a plausible solution would be that knowledge about the self is maintained by spe-
cialized memory systems, for instance, autobiographical declarative memories (both semantic
and episodic) while beliefs about others are maintained by separate declarative memories. Like-
wise, procedural memory would contain distinct knowledge about actions allowing the agent
to pursue both individual and collective goals, and those actions would be competing against
each other inside the cognitive agent’s mind when there would be conflicting interests. Another
concern to be addressed is the potential necessity of keeping a collective memory to store shared
knowledge about the world, social behaviors and norms developed by agents. This would be
a distributed knowledge system by nature, but also would reflect high levels of redundancy,
that is, agents would have a partial copy of the knowledge system (constructed as a result of
learning processes) which would allow them to communicate ideas efficiently with each other.
One potential issue with this approach is the representational incompatibility between agents
grounded in a different cognitive architecture and, therefore, in different representational and
reasoning assumptions. A possible way out, as suggested above, is that one of exploiting the
3There are, however, attempts to extend in a efficient way the Semantic Memory of Soar with external lexical
resources such as, for example, Wordnet [8]
6
Higher-level Constraints of the CMC Lieto, Kennnedy, Lebiere, Romero, Taatgen, West
potential of representational and reasoning systems `a la DUAL-PECCS that have shown a good
level of compatibility with diverse architectures. More generally, this is the kind of capability
that the CMC aims to enable.
7 Problems for Macro-Cognition
As mentioned above, there is a majoritarian consensus that no additional specialized architec-
tural modules are necessary for performing high-order capabilities of the social band. However,
some specific architectural primitives may be still necessary for supporting social cognition such
as visual imagery for visual-feature reasoning; pre-attentive and attentive vision sub-modules
for the recognition of complex non-verbal signals of attention and emotional state, social-gaze
following, face detection, etc. In the CMC, in fact, there is a real consensus over the ‘minimum’
number of high-level modules that are present in a cognitive architecture (e.g., perception and
motor, working memory, declarative memory, and procedural memory); however, it still remains
incomplete concerning other modules and mechanisms that can be useful for social cognition:
emotion, direct communication, language, attention, metacognition, ethical/moral reasoning,
among others [5, 36, 12, 37]. Thus, we envision that the main challenge is to identify which of
these additional modules and mechanisms are completely necessary (and therefore should be
included in the CMC) and what their level of involvement in social cognition is. For instance,
it seems that at the lowest level, simple social interactions (such as gaze following) require at
least the interplay of perception, attention and memory modules, whereas more complex social
interactions (such as building rapport, negotiation, etc.) may require the interplay of additional
modules such as emotion/motivations, metacognition, and language, just to name a few. So, it
is necessary to define a well-structured hierarchy of a representative set of social interactions
that help us identify which modules and mechanisms are strictly necessary at each level of the
hierarchy and, from there, establish the architectural constraints that should be added to the
CMC in order to allow social cognition to emerge.
Concerning the learning assumptions of the CMC connected the Social Band: the current
version of the model [13] states that learning occurs mainly in two modules, procedural and
declarative, where procedural learning involves at least reinforcement learning and procedural
composition, and that more complex forms of learning involve combinations of the fixed set
of simpler forms of learning. We know, however, from the Social Learning theory [4] that
learning takes place in a social context and can occur purely through observation/imitation
or direct instruction, even in the absence of motor reproduction or direct reinforcement. In
addition to this, the field of robotics and multi-agent systems have reported [10] that learning
by imitation can be supported not only by classical reinforcement learning but also by supervised
learning (e.g., behavioral cloning, learning by demonstration), inverse reinforcement learning
(e.g., apprenticeship learning) and transfer learning. Therefore, some questions to address at
this point are: what other kind of learning, other than reinforcement learning, can be considered
to model cognitively plausible social agents? What kind of architectural criteria should be taken
into consideration in order to determine the ‘minimum’ requirements for procedural learning
when modeling social cognition?
As mentioned earlier, an alternative proposal concerning the modeling of macro-cognitive
phenomena has been recently pointed out by [44]. The authors applied to macro-cognition the
same criticism raised by [22] as to the epistemological value of creating different models for
each cognitive phenomena. In Newell’s view this praxis is of limited use because it leads to
a multitude of unrelated micro models. Similarly, [44] noticed that the field studying macro-
cognition currently produces a vast array of ad hoc models and pointed out that this is true
7
Higher-level Constraints of the CMC Lieto, Kennnedy, Lebiere, Romero, Taatgen, West
even when well specified cognitive architectures are used to make the models. This is because
there are multiple ways to implement complex tasks in a cognitive architecture since it does
not restrict the knowledge content at the rational band (see above). The adoption of Macro-
cognitive architectures may offer some resolution to this problem as they are based on the
proposal that the knowledge level is constrained and tends to be organized in particular ways.
In terms of relating the common model architectural principles to macro-cognition, a prag-
matic approach is to specify best practices and common solutions for supporting macro-cognitive
functions, such as planning, dealing with unexpected interruptions, task switching, problem
solving, and dealing with large knowledge bases. This could be thought of as constituting
a proto common model of macro-architecture, but one doesn’t need to buy into the macro-
architecture concept in order to appreciate that this sort of approach would have practical
advantages for facilitating applications of the common model to real world modeling projects.
A starting point for this would be to gather all the models built in common model architectures
that are related to macro cognition and look for commonalities. That is, use the same method-
ology that was used to conceptualize the common model [13]. Likewise this same approach
could be used with knowledge level languages to see if there are any commonalities.
Alternatively, an approach can be chosen in which the higher level of abstraction is treated
as a more independent modeling platform with its own representations and mechanisms, but
one that can be reduced to the level of the Common Model (e.g., [39]). The possible interplay
between these two contrasting views is not yet entirely clear and represents, for sure, an impor-
tant element of discussion and elaboration to address within this working group and between
diverse CMC working groups.
8 Conclusion
We have proposed an overview of some of the main constraints and open problems that should
be addressed within the CMC concerning the knowledge/rational/social levels. In our analysis
we also have specified, when possible, some plausible directions to follow in order to overcome,
or partially address, such problems.
Acknowledgments
This paper reflects the current thoughts of the majority of the working group on the reported
issues. We also thank Phil C. Jackson and Jerald D. Kralik for their feedback. The first author
acknowledges the support by a MIUR RTD-A grant from the University of Turin, Department of
Computer Science. The second author is supported in part by the Center for Social Complexity,
George Mason University and the Defense Thread Reduction Agency (DTRA) grant. The third
author acknowledges the support of DARPA award FA8650-18-C-7826.
References
[1] John R Anderson. Spanning seven orders of magnitude: A challenge for cognitive modeling.
Cognitive Science, 26(1):85–112, 2002.
[2] John R Anderson, Daniel Bothell, Michael D Byrne, Scott Douglass, Christian Lebiere, and Yulin
Qin. An integrated theory of the mind. Psychological review, 111(4):1036, 2004.
[3] Jerry Ball, Stuart Rodgers, and Kevin Gluck. Integrating ACT-R and Cyc in a large-scale model
of language comprehension for use in intelligent agents. In AAAI workshop, pages 19–25, 2004.
[4] Albert Bandura. Social learning theory. General Learning Corporation, 1971.
8
Higher-level Constraints of the CMC Lieto, Kennnedy, Lebiere, Romero, Taatgen, West
[5] Paul F Bello and Will Bridewell. There is no agency without attention. AI Magazine, 38(4):27–33,
2017.
[6] Nicholas Louis Cassimatis. Polyscheme: a cognitive architecture for intergrating multiple repre-
sentation and inference schemes. PhD thesis, Massachusetts Institute of Technology, 2001.
[7] Antonio Chella, Marcello Frixione, and Antonio Lieto. Representational issues in the debate on
the standard model of the mind. In 2017 AAAI Fall Symposium, pages 302–307. AAAI Press,
2017.
[8] Nate Derbinsky, John E Laird, and Bryan Smith. Towards efficiently supporting large symbolic
declarative memories. 1001:48109–2121, 2010.
[9] Peter G¨ardenfors. Conceptual spaces: The geometry of thought. MIT press, 2000.
[10] Ahmed Hussein, Mohamed Medhat Gaber, Eyad Elyan, and Chrisina Jayne. Imitation learning:
A survey of learning methods. ACM Computing Surveys (CSUR), 50(2):21, 2017.
[11] Philip C Jackson. Toward human-level models of minds. In 2017 AAAI Fall Symposium Series,
pages 371–375, 2017.
[12] Iuliia Kotseruba and John K Tsotsos. A review of 40 years of cognitive architecture research: Core
cognitive abilities and practical applications. arXiv preprint, 2018.
[13] J. E. Laird, C. Lebiere, and P. S. Rosenbloom. A standard model of the mind: Toward a com-
mon computational framework across artificial intelligence, cognitive science, neuroscience, and
robotics. AI Magazine, 38:1–19, 2017.
[14] John Laird. The Soar cognitive architecture. MIT Press, 2012.
[15] Douglas B Lenat. Cyc: A large-scale investment in knowledge infrastructure. Communications of
the ACM, 38(11):33–38, 1995.
[16] Antonio Lieto. Heterogeneous proxytypes extended: Integrating theory-like representations and
mechanisms with prototypes and exemplars. In Biologically Inspired Cognitive Architectures 2018.
BICA 2018. Advances in Intelligent Systems and Computing, volume 848. Springer, Cham.
[17] Antonio Lieto, Antonio Chella, and Marcello Frixione. Conceptual spaces for cognitive archi-
tectures: A lingua franca for different levels of representation. Biologically Inspired Cognitive
Architectures, 19:1–9, 2017.
[18] Antonio Lieto, Christian Lebiere, and Alessandro Oltramari. The knowledge level in cognitive
architectures: Current limitations and possible developments. Cognitive Systems Research, 48:39–
55, 2018.
[19] Antonio Lieto, Daniele P Radicioni, and Valentina Rho. Dual PECCS: a cognitive system for
conceptual representation and categorization. Journal of Experimental & Theoretical Artificial
Intelligence, 29(2):433–452, 2017.
[20] Claudio Masolo, Stefano Borgo, Aldo Gangemi, Nicola Guarino, and Alessandro Oltramari. Won-
derweb deliverable d18, ontology library (final). ICT project, 33052, 2003.
[21] George A Miller. Wordnet: a lexical database for english. Communications of the ACM, 38(11):39–
41, 1995.
[22] Allen Newell. You can’t play 20 questions with nature and win: Projective comments on the
papers of this symposium. Visual Information Processing: Proceedings of the Eighth Annual
Carnegie Symposium on Cognition, 1973.
[23] Allen Newell. The knowledge level. Artificial intelligence, 18(1):87–127, 1982.
[24] Allen Newell. Unified theories of cognition. Harvard University Press, 1990.
[25] Alessandro Oltramari and Christian Lebiere. Pursuing artificial general intelligence by leveraging
the knowledge capabilities of ACT-R. In Artificial General Intelligence, pages 199–208. Springer,
2012.
[26] Mark G. Orr, Christian Lebiere, Andrea Stocco, Peter Pirolli, and William G. Kennedy. Multi-
scale resolution of cognitive architectures: A paradigm for simulating minds and society. In
SBP-BRiMS, pages 1–14, 2018.
9
Higher-level Constraints of the CMC Lieto, Kennnedy, Lebiere, Romero, Taatgen, West
[27] David V Pynadath, Paul S Rosenbloom, Stacy C Marsella, and Lingshan Li. Modeling two-player
games in the sigma graphical cognitive architecture. In Int. Conf. on AGI, pages 98–108, 2013.
[28] Anand S Rao and Michael P Georgeff. Modeling rational agents within a BDI-architecture. KR,
91:473–484, 1991.
[29] Anand S Rao, Michael P Georgeff, et al. BDI agents: from theory to practice. In ICMAS,
volume 95, pages 312–319, 1995.
[30] Oscar J Romero and Christian Lebiere. Cognitive modeling of behavioral experiments in network
science using ACT-R architecture. In MAS and Agent-Based Simulation, pages 239–251, 2014.
[31] Oscar J Romero and Christian Lebiere. Simulating network behavioral dynamics by using a multi-
agent approach driven by ACT-R cognitive architecture. 2015.
[32] Paul S Rosenbloom. Towards a new cognitive hourglass: Uniform implementation of cognitive
architecture via factor graphs. In 9th Int. Conf. on cognitive modeling, pages 116–121, 2009.
[33] Paul S Rosenbloom. Rethinking cognitive architecture via graphical models. Cognitive Systems
Research, 12(2):198–209, 2011.
[34] Dario D Salvucci. Endowing a cognitive architecture with world knowledge. In Procs. of the 36th
Annual Meeting of the Cognitive Science Society, 2014.
[35] Jonathan R Scally, Nicholas L Cassimatis, and Hiroyuki Uchida. Worlds as a unifying element of
knowledge representation. Biologically Inspired Cognitive Architectures, 1:14–22, 2012.
[36] Brian Scassellati. Theory of mind for a humanoid robot. Autonomous Robots, 12(1):13–24, 2002.
[37] Matthias Scheutz. The case for explicit ethical agents. AI Magazine, 38(4):57–64, 2017.
[38] Bernd Schmidt. The modelling of human behaviour, volume 132. Society for Computer Simulation
International, 2000.
[39] Niels A Taatgen. Cognitive architectures: Innate or learned? In 2017 AAAI Fall Symposium
Series, pages 476–480, 2017.
[40] J Gregory Trafton, Laura M Hiatt, Anthony M Harrison, Franklin P Tamborello II, Sangeet S
Khemlani, and Alan C Schultz. ACT-R/E: An embodied cognitive architecture for human-robot
interaction. Human-Robot Interaction, 2(1):30–55, 2013.
[41] J Gregory Trafton, Alan C Schultz, Magdalena Bugajska, and Farilee Mintz. Perspective-taking
with robots: experiments and models. In Robot and Human Interactive Comm., pages 580–584,
2005.
[42] Lara M Triona, Amy M Masnick, and Bradley J Morris. What does it take to pass the false belief
task? an ACT-R model. In Cognitive Science Society, volume 24, 2002.
[43] Christoph Urban. PECS: A reference model for the simulation of multi-agent systems. In Tools
and techniques for social science simulation, pages 83–114. Springer, 2000.
[44] Robert L West and Korey MacDougall. The macro-architecture hypothesis: Modifying newells
system levels to include macro-cognition. Biologically Inspired Cognitive Architectures, 8:140–149,
2014.
10
