ArticlePDF Available

Abstract and Figures

See a video on the YouTube channel on Self-Modeling Networks here: https://www.youtube.com/channel/UCCO3i4_Fwi22cEqL8M_PgeA. In this paper, a multilevel cognitive architecture is introduced that can be used to model mental processes in clients of psychotherapeutic sessions. The architecture does not only cover base level mental processes but also mental processes involving self-referencing, self-awareness and self-interpretation. To this end, the cognitive architecture was designed according to four levels, where (part of) the structure of each level is represented by an explicit self-model of it at the next-higher level of the architecture. At that next-higher level, states reify part of the structure of the level below; these states have a referencing relation to it. In this way the overall architecture includes its own overall self-model. The cognitive architecture was evaluated for a case study of a realistic type of therapeutic session from clinical practice.
Content may be subject to copyright.
A multi-level cognitive architecture for self-referencing,
self-awareness and self-interpretation
Jan Treur
a,
, Gerrit Glas
b,c
a
Social AI Group, Department of Computer Science, Vrije Universiteit Amsterdam, the Netherlands
b
Epistemology and Metaphysics Group, Department of Philosophy, Vrije Universiteit Amsterdam, the Netherlands
c
Anatomy and Neurosciences Department, Amsterdam University Medical Center, the Netherlands
Received 1 September 2020; accepted 31 October 2020
Available online 23 November 2020
Abstract
In this paper, a multilevel cognitive architecture is introduced that can be used to model mental processes in clients of psychother-
apeutic sessions. The architecture does not only cover base level mental processes but also mental processes involving self-
referencing, self-awareness and self-interpretation. To this end, the cognitive architecture was designed according to four levels, where
(part of) the structure of each level is represented by an explicit self-model of it at the next-higher level of the architecture. At that next-
higher level, states reify part of the structure of the level below; these states have a referencing relation to it. In this way the overall archi-
tecture includes its own overall self-model. The cognitive architecture was evaluated for a case study of a realistic type of therapeutic
session from clinical practice.
Ó2020 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/
4.0/).
1. Introduction
Within the discipline of psychiatry there is a longstand-
ing tradition and much experience in supporting clients in
the path toward discovery of themselves. Many types of
therapy at least partly aim at enhancing the client’s knowl-
edge and awareness of her or himself and use this to his or
her benefit. This may involve elements such as:
Getting familiar with one’s own personal characteristics
Being aware of important aspects of oneself
Being able to interpret one’s behaviour in relation to
other aspects of oneself
Based on such insights being able to manage oneself
more effectively
Much of the knowledge about such therapeutic processes
has been acquired within a practical clinical context. This
can be considered a very valuable source of knowledge
which only very partially has been exploited and further
developed from a more academic research perspective; see
however (Montgomery, 2006; Nicolini, 2012; Gascoigne &
Thornton, 2014). So far, to our knowledge, no form of for-
malisation has been applied or any further computational
analysis, except for statistical analyses of gathered data.
The current paper does address formalisation and more
detailed computational analysis of the types of processes as
briefly sketched above. Based on a recent conceptual anal-
ysis of clinical practice (Glas, 2017; Glas, 2019), a cognitive
architecture has been designed involving the relevant
processes of self-referencing, self-awareness and self-
interpretation. This cognitive architecture has been
specified formally using the network-oriented modeling
language described in more detail in (Treur, 2020b). Based
on this specification, for a particular case study simulation
https://doi.org/10.1016/j.cogsys.2020.10.019
1389-0417/Ó2020 The Authors. Published by Elsevier B.V.
This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
Corresponding author.
E-mail addresses: j.treur@vu.nl (J. Treur), g.glas@vu.nl (G. Glas).
www.elsevier.com/locate/cogsys
Available online at www.sciencedirect.com
ScienceDirect
Cognitive Systems Research 68 (2021) 125–142
experiments have been conducted using the available dedi-
cated software environment.
In the paper, first in Section 2 the psychiatric context
addressed is briefly outlined and describes the case study
addressed in later sections. In Section 3 the notions of reifica-
tion and reified network are briefly discussed. Section 4
shows how these notions are useful to model the relevant ele-
ments of the considered psychiatric context and discusses a
global overview of the designed multilevel cognitive architec-
ture. In Section 5 an example case study is used to illustrate
the use of the architecture, while Section 6 provides a detailed
specification of this. Section 7 presents an example simula-
tion for this case study. Finally, Section 8 is a discussion.
2. Perspectives from a psychiatric context
This section briefly explores the notions self-
referentiality, self-awareness and self-interpretation that
play an important role in clinical practice and are the basis
of the cognitive architecture introduced in next sections.
2.1. Self-referentiality
Many symptoms in the context of psychiatric illness are
not just expressions of an underlying molecular, physiolog-
ical or even psychological derailment but also implicitly
refer to the person having them. Strikingly enough, aca-
demic psychology and psychiatry have largely ignored this
obvious fact. They still lack a conceptual framework to
make sense of the self-referencing quality of the emotions,
gestures, verbal expressions, and interactions that occur in
the context of mental illness. The term self-referentiality is
meant to highlight the implicit signifying aspect of these
behaviors. These behaviours refer to a ‘self’, i.e., to an
aspect or to aspects of the person having them.
In what follows we will restrict the term to the implicit
signifying aspects of affective, communicative, verbal and
social behaviors. Self-referentiality is not adopting a stance
toward these behaviors. Adopting a stance is a form of self-
relating. Self-referentiality differs from self-relating in that
the behaviors just mentioned by themselves ‘say’ (signify,
indicate) something about the person having them. Their
referring quality is not a product of self-reflection and
interpretation; the phenomena themselves signify and
reveal what is going on with the person having them. This
self-referential signifying does not exclude self-awareness,
self-relating or self-interpretation. Emotions, for instance,
don’t lose their self-referential qualities when one becomes
aware of them. Self-referentiality and self-awareness may
go together. But it still makes sense to distinguish them
conceptually; see also (Glas, 2017), p. 146.
2.2. Self-awareness
By self-awareness we mean that there exists an aware-
ness of an aspect of oneself or of oneself as a whole. It is
a state of mind in which an aspect of oneself, or one’s self,
becomes the object of consciousness (or: reflection); e.g.,
(Glas, 2017, p. 146).
2.3. Self-interpretation
Self-interpretation refers to the way people understand
(perceive, value) themselves. Self-interpretation builds
forth on self-awareness and on self-referential aspects of
certain behaviors. It often leads to a re-evaluation of one’s
initial perceptions. Self-interpretation is needed when the
initial awareness of an expression, thought, or utterance
is unclear or ambiguous. I may feel hurt by someone with-
out initially knowing why. Upon further reflection, I may
discover why I felt hurt. I may realize how the other person
subtly awakened my latent feelings of inferiority or threat-
ened my feelings about someone I love. The anger again
reveals that something that (or: someone who) is important
for me, is threatened. Further reflection (or discussion with
others) makes me aware of the nature and severity of the
threat; e.g., (Glas, 2019), p. 29.
2.4. Other literature
We understand that the conceptual framework we are
describing looks novel and that it does not seem to neatly
fit within the current scene of cognitive science, psy-
chopathology and even philosophy. To be honest, the term
‘self’ has become important in current psychology and cog-
nitive neuroscience. There exist, for instance, a lot of
research on emotion-regulation and self-regulation (Gross
& Jazaieri, 2014; Gross & Thompson, 2007; Tracy,
Klonsky, & Proudfit, 2014). While the concepts of
emotion-regulation and self-regulation are evidently highly
relevant for psychiatry, they typically aim at what occurs
with emotions, i.e., post hoc. Koole (2009) defines
emotion-regulation as ‘‘the set of processes whereby people
seek to redirect the spontaneous flow of their emotions.
Gross (2008) states that emotion regulation is about
‘‘how we try to influence which emotions we have, when
we have them, and how we experience and express these
emotions; see also (Gross & Thompson, 2007) and for
an opposite view, (Kappas, 2011).
The concept of the self is an emerging theme in many
other respects, i.e., in the context of cognitive neuro-
science (Christoff, Cosmelli, Legrand, & Thompson,
2011; Damasio, 1999, 2010; Immordino-Yang, 2011;
Northoff, Qin, & Feinberg, 2011; Reddy, 2009), develop-
mental psychology (Fonagy et al., 2002; Hobson, 2010),
general psychology and personality theory (Leary &
Tangney, 2003), social psychology (Tracy & Robins,
2007), and philosophy (Gallagher, 2013; Metzinger,
2007, 2004). However, the focus of attention has so far
often been on so-called self-conscious and moral emotions
like shame, guilt, embarrassment, social anxiety, pride,
and ambivalence (Leary, 2007, Prinz, 2010, Rorty,
2010). The self-conscious emotions have the self as focus
or object, though usually indirectly, by making inferences
about other people’s evaluations of oneself. With self-
referentiality we mean something else than this kind of
self-concern, however.
Similar accounts are rare, indeed, and are to be found in
the phenomenological tradition (Solomon, 1983; Stern,
126 J. Treur, G. Glas / Cognitive Systems Research 68 (2021) 125–142
2004; Zahavi, 2005; Ratcliffe, 2005, 2008, 2010; Slaby &
Stephan, 2008; Stephan, 2012; Atkinson & Ratcliffe,
2012). The term self-referentiality (and: self-referencing)
was initially coined by Paul Ricoeur (1992). The term indi-
cated the implicit reflexivity (or the ‘ipse’) of symbolic
expressions. Our account on self-referentiality comes clos-
est to Ricoeur’s; however, he seems to include forms of
referring that are part of one’s conscious awareness and,
therefore, are ready to be interpreted. So, we stick to a
more restrictive definition of self-referencing, i.e, as an
implicit form of signification.
2.5. Point of departure for the case study used
In this article we use a simple example to illustrate the
architecture that is indicated with the terms self-
referentiality, self-awareness, and self-interpretation. We
depart with a slightly changed and extended variant of a
case description that was given in other work by one of
the authors (Glas, 2017, p. 151):
John is a 24 year old student, with a positive family history
for mood disorder. He has recently developed a moderately
severe depression. John has some physical complaints
related to the depression and visits his general practitioner
who assures him that there is nothing wrong physically.
The GP considers psychosocial stress factors as a possible
source of what he calls `the burn-out'. John tells that he
broke up with a girlfriend six months ago and that he has
difculty with nishing his master thesis. However, he has
broken up earlier with other girlfriends and he does not
seem very worried about the lack of progress with his thesis.
The GP suggests physical exercises and prescribes sleeping
pills for a period of two weeks, after which he wants to see
John again. John does not show-up, however, and his con-
dition worsens. Finishing his study doesn't make sense.
He doesn't have a future and the world would be better
off without him, he thinks. His friends try to motivate him
to look for professional help but John refuses. He doesn't
do his physical exercises, begins to drink large amounts of
alcohol and slips into a state of sheer passivity.
3. Reification and reified network modeling
In the previous section it may already have become
clear conceptually and linguistically that for their content
many of the concepts used have some reference relation to
other concepts. For example, self-awareness refers to the
content of the awareness which can be one of oneself’s
personal characteristics, and in turn such a characteristic
refers to its content which can be a relation between men-
tal states such as a memory and a belief. To formalise
such reference relations, within AI and other disciplines
often the notion of reification is used; e.g., (Galton,
2006). Similar, related concepts within AI and Computer
Science are reflective architectures, metalevel architec-
tures, metainterpreters, and metaprogramming; e.g.,
(Bowen & Kowalski, 1982; Sterling & Beer, 1989;
Weyhrauch, 1980). In this section the notion of reification
is discussed in some more detail and it is indicated how it
has been used in (Treur, 2020b) for network-oriented
modeling in an iterative manner, thus obtaining multilevel
reified network models. Such network models will turn
out to provide an adequate basis to formalise what was
described in Section 2.
3.1. Reification
We often describe the dynamics of processes in any
domain of the world by causal relations, and certainly this
usually happens when a biological perspective is used. In
principle, the occurrences of these causal relations are
based on the configuration of the world; this configuration
by itself can be dynamic as well, which is often described as
adaptation. For example:
(a) Dynamics of neural or mental states based on the
causal relations as used to describe the brain; e.g.,
(Kim, 1996). Adaptation of these causal relations
occurs, for example, as changing synapses or excita-
tion thresholds within the brain; e.g., (Hebb, 1949;
Tse, 2013; Chandra & Barkai, 2018)
(b) Dynamics of social processes based on causal rela-
tions describing how individuals affect each other as
used within the social domain; e.g., (Levy & Nail,
1993; Iacoboni, 2008). Adaptation of these causal
relations within the social domain, occurs as chang-
ing connections by bonding and within the underly-
ing brain processes such as mirroring; e.g.,
(McPherson, Smith-Lovin, & Cook, 2001; Iacoboni,
2008; Keysers & Gazzola, 2014)
(c) Dynamics of biochemical processes based on the cau-
sal relations as used within the biochemical domain.
Adaptation of these causal relations occurs as
changes in such networks; e.g., (Westerhoff et al.,
2014a; Westerhoff et al., 2014b)
(d) Dynamics due to causal pathways within organisms
(Westerhoff et al., 2014a; Westerhoff et al., 2014b).
Adaptation within evolutionary biological processes
occurs as changing the causal pathways; e.g.,
(Fessler et al., 2005, 2015)
(e) Dynamics of physical processes based on causal rela-
tions as used within the physical domain. Adaptation
of such causal relations, occurs, for example, as add-
ing smooth roads in a landscape achieving lower
resistance when moving or adding digital electronic
networks so that humans can interact via social
media, or changing the positioning of the earth with
respect to the sun achieving hourly and seasonal
differences in meteorological dynamics; e.g.,
(Descartes, 1644; Leibniz, 1698; Newton, 1729;
Lorenz, 1963; Lorenz, 1993)
So, these examples illustrate that the causal relations
themselves have some form of embodiment or representa-
J. Treur, G. Glas / Cognitive Systems Research 68 (2021) 125–142 127
tion within the world and if that is changing, the causal
relations and their effects change accordingly. This can be
described by reification. A causal relation is by itself an
abstract concept which is made concrete, ‘is reified’, by
the world configuration on which it is based. This will be
worked out for the context of (temporal-causal) network
models that model such causal relationships.
3.2. Reified network modeling
The concept of reification has been introduced for net-
work modeling in (Treur, 2018a; Treur, 2020a) and is the
basis for modeling adaptive networks in (Treur, 2020b).
Distinction between network characteristics and network
states
The following is a crucial distinction for network
models:
Network characteristics (such as connection weights and
excitability thresholds) have values (their strengths) and
determine (e.g., mental) processes and behaviour in an
implicit, automatic manner. They can be considered to
provide an embodiment view on the network. In princi-
ple, these characteristics by themselves may not be
directly accessible nor observable for network states
(or a person: usually you don’t see or feel a specific con-
nection in your brain).
Network states (such as sensor states, sensory represen-
tation states, preparation states, emotion states) have
values (their activation levels) and are explicit represen-
tations that may be accesible for network states or a per-
son and can be handled or manipulated explicitly. They
can be considered to provide an informational view on
the network; usually the states are assumed to have a cer-
tain informational content. In principle, for the case of a
mental network, states may be accessible or observable
for a person: you may see (mental image), feel (emotion)
or note in some other way a specific state in your brain.
The type of network used here has been called a
temporal-causal network, as introduced in (Treur, 2016a,
2016b). Such a network represents states Xand connec-
tions X?Ybetween them for (causal) impacts; here sta-
tes Xhave values X(t) that usually change over time t.
More precisely, the following notions form the defining
network characteristics of a temporal-causal network model
(in the form of mathematical relations and functions):
(a) Connectivity of the network
connection weights x
X,Y
2[1, 1] for each connec-
tion from a state Xto a state Y
(b) Aggregation of multiple impacts on a given state in the
network
basic combination functions c
j
(..),j= 1, .., mfor aggrega-
tion, selected for the whole network model from an
available combination function library; this is done by
specifying mcf =[k
1
, .., k
m
], where k
j
refers to the num-
ber combination function c
j
(..) has within the library.
for each state Y combination function weights c
j,Y
for the
basic combination functions c
j
(..),j= 1, .., mto indicate
by a weighted average of the functions c
j
(..),j= 1, .., m
the aggregation of incoming single causal impacts
xXi;YXiðtÞof the states X
1
, .., X
k
from which Ygets
incoming connections
for each state Yand combination function c
j
(..) combi-
nation function parameters p
i,j,Y
(c) Timing in the network
for each state Yaspeed factor g
Y
0
The above defined characteristics x
X,Y
,c
i,Y
,p
i,j,Y
,g
Y
define in a canonical manner an associated numerical rep-
resentation of the network model (Treur, 2016b), Ch. 2, in
difference or differential equation format which can be used
for simulation and mathematical analysis:
YðtþDtÞ¼YðtÞþgY½cYðxX1;YX1ðtÞ;;xXk;YXkðtÞÞ
YðtÞDtð1Þ
dYðtÞ=dt¼gY½cYðxX1;YX1ðtÞ;;xXk;YXkðtÞÞ  YðtÞ
Here the overall combination function c
Y
(..) for state Y
is the weighted average of the basic combination functions
c
j
(..) by the specified weights c
j,Y
for Y:
cYðV1;;VkÞ¼c1;Yc1ðV1;;VkÞþþcm;YcmðV1;;VkÞ
c1;Yþþcm;Y
ð2Þ
Such equations are hidden in the dedicated software
environment; see (Treur, 2020b), Ch 9. Making the param-
eters p1;j;Y,p2;j;Yof the basic combination functions c
j,Y
(..)
explicit, this becomes
cYðp1;1;Y;p2;1;Y;;p1;m;Y;p2;m;Y;V1;;VkÞ
¼c1;Yc1ðp1;1;Y;p2;1;Y;V1;;VkÞþþcm;Ycmðp1;m;Yp2;m;Y;V1;;VkÞ
c1;Yþþcm;Y
ð3Þ
There are many different approaches possible to address
the issue of aggregating multiple impacts by combination
functions. Therefore, for this aggregation a combination
function library with a number of basic combination func-
tions (currently >35) is available, while also own-defined
functions can be added. Examples of basic combination
functions from this library can be found in Table 1.
Network reification connecting network characteristics
and network states
As indicated above, ‘network characteristics’ and ‘net-
work states’ are two distinct concepts for a network. Net-
work reification is a way to relate these distinct concepts
to each other in an interesting and useful way:
Network reification is making the implicit network char-
acteristics (such as connection weights and excitability
thresholds) explicit by adding states for these character-
128 J. Treur, G. Glas / Cognitive Systems Research 68 (2021) 125–142
istics; thus the network gets an internal self-model of
(part of) the network itself. Such additional states are
called reification states.
In this way, different reification levels are created where
characteristics from one level relate to explicit states at a
next reification level. By iteration, an arbitrary number
of reification levels can be modeled, covering second-
order or higher-order effects.
Due to the above description, by network reification,
states at a certain level refer to characteristics of the next
lower level. This referencing relation directly relates to
the reification relation: state Xreifies characteristic Y
means that state X refers to characteristic Y, also
phrased as state X represents characteristic Y. As these
Yare characteristics of the network (or of a person),
the representations Xare also said to form an internal
self-model of the network (or of the person), and the ref-
erencing is also called self-referencing, as the referencing
concerns referring to the network (or person) itself.
Reification can be applied in relation to the physical
world by itself, but also for mental domains. For
example,
o in and by the world, in the brain information about
causal relations between brain states is represented
in physical states for connection weights, excitability
thresholds or other characteristics
o in the mental domain, a person can create mental
states in the form of representations of his or her
own (personal) characteristics, thus forming a subjec-
tive self-model (acquired by experiences)
In a wider context, reification means representing some-
thing abstract as a material or more real concrete thing
(Merriam-Webster and Oxford dictionaries). This concept
is used in different scientific areas in which it has been
shown to provide substantial advantages in expressivity
and in structuring and transparency of models, in particu-
lar, within AI; e.g., (Bowen & Kowalski, 1982; Galton,
2006; Sterling & Beer, 1989; Weyhrauch, 1980). Specific
cases of reification from a linguistic or logical perspective
are representing relations between objects by objects them-
selves, or representing more complex statements about
objects or numbers, themselves by objects or numbers,
for example, like Go
¨del used natural numbers to represent
logical statements to prove his famous incompleteness the-
orems for logic; e.g., (Nagel & Newman, 1965; Smorynski,
1977).
For network modeling, this notion turns out useful, not
only to describe adaptive networks in a suitable manner
(Treur, 2020b), but also to model the different levels for
self-referencing, self-awareness and self-interpretation as
discussed in Section 2. This will be explained in more detail
in the next sections. To model adaptive networks specifi-
cally, network reification for a temporal-causal network
is applied in the way that for each state Yof the base net-
work, for the adaptive ones among the network structure
characteristics x
X,Y
,c
i,Y
,p
i,j,Y
,g
Y
, additional network
states W
X,Y
,C
i,Y
,P
i,j,Y
,H
Y
(reification states) can be intro-
duced (see the blue upper plane in Fig. 1):
(a) Connectivity characteristics reification
reification states WXi;Yare added representing adaptive
connection weights xXi;Y
(b) Aggregation characteristics reification
reification states C
j,Y
are added representing adaptive
combination function weights c
i,Y
reification states P
i,j,Y
are added representing adaptive
combination function parameters p
i,j,Y
(c) Timing characteristics reification
reification states H
Y
are added representing adaptive
speed factors g
Y
The notations W
X,Y
,C
i,Y
,P
i,j,Y
,H
Y
for the reification
states indicate the referencing relation with respect to the
characteristics x
X,Y
,c
i,Y
,p
i,j,Y
,g
Y
:hereWrefers to x,C
refers to c(..),Prefers to p, and Hrefers to g, respectively.
For the processing, these reification states define the
dynamics of state Yin a canonical manner according to
equations (1) whereby x
X,Y
,c
i,Y
,p
i,j,Y
,g
Y
are replaced by
the state values of W
X,Y
,C
i,Y
,P
i,j,Y
,H
Y
, respectively.
An example of a representation P
i,j,Y
for combination
function parameter p
i,j,Y
is for the exitability threshold s
Y
Table 1
Examples of basic combination functions from the library.
Combination function Notation Formula Parameters
Identity id(V)V
Complemental identity compid(V)1V
Scaled sum ssum
k
(V
1
,...,V
k
)V1þþVk
kScaling factor k>0
Simple logistic slogistic
r,s
(V
1
,...,V
k
)1
1þerðV1þþVksÞSteepness r>0
Excitability threshold s
Advanced logistic alogistic
r,s
(V
1
,...,V
k
)½1
1þerðV1þþVksÞ1
1þersÞ(1 + e
rs
) Steepness r>0
Excitability threshold s
Scaled maximun smax
k
(V
1
,...,V
k
)minðV1;;VkÞ
kScaling factor k>0
Scaled minimum smin
k
(V
1
,...,V
k
)maxðV1;;VkÞ
kScaling factor k>0
Euclidean eucl
n,k
(V
1
,...,V
k
)ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
V1nþþVk
n
k
n
qOrder n>0Scaling factor k>0
Scaled geometric mean sgeomean
k
(V
1
,...,V
k
)ffiffiffiffiffiffiffiffiffiffiffiffiffiffi
V1Vk
k
k
qScaling factor k>0
J. Treur, G. Glas / Cognitive Systems Research 68 (2021) 125–142 129
of state Y; then P
i,j,Y
is usually indicated by T
Y
, where T
refers to s. Such reification states T
Y
will play a role in
the case study described below, as will reification states
W
X,Y
, referring to connection weights x
X,Y
. These two
types of reification states can be used to model adaptive
intrinsic neuronal excitability and adaptive connection
weights as described, for example, in (Chandra & Barkai,
2018); e.g.:
`Learning-related cellular changes can be divided into two
general groups: modications that occur at synapses and
modications in the intrinsic properties of the neurons.
While it is commonly agreed that changes in strength of
connections between neurons in the relevant networks
underlie memory storage, ample evidence suggests that
modications in intrinsic neuronal properties may also
account for learning related behavioral changes. Long-
lasting modications in intrinsic excitability are manifested
in changes in the neuron's response to a given extrinsic cur-
rent (generated by synaptic activity or applied via the
recording electrode).' (Chandra & Barkai, 2018, p. 30)
In addition to such specific reification states W
X,Y
,C
i,Y
,
P
i,j,Y
,H
Y
that directly relate to the actual values of the
related characteristics as described, also other reification
states can be considered that do not have such a direct rela-
tion, but still can be considered a form of reification. For
example, subjective reification states can be introduced that
indicate what an invidual thinks that is the connection from
one state to another one. Such types of self-referencing reifi-
cation states will be discussed as well in Section 4, in the con-
text of the designed cognitive architecture.
4. The overall cognitive architecture
The global structure of the overall architecture is shown in
Fig. 1. In addition to the base level it displays at different
levels Self-Referencing, Self-Awareness, and Self-
Interpretation, as also discussed in Sections 2.1, 2.2, and
2.3, respectively.
In some more detail this architecture can be described as
follows.
4.1. Base level
At the base level the mental processes of a person are con-
sidered, in the case study the client during some type of ther-
apeutic session. These mental processes are modeled
through causal connections between mental states that drive
their dynamics. Using the concept of temporal-causal net-
work for that, a number of network characteristics are spec-
ified that determine the person’s mental functioning; by Eq.
(1) these characteristics define the base level’s dynamics for
the mental processes. The characteristics such as connection
weights and excitability thresholds may cover, for example:
o tendencies to do something or avoid something in terms
of actions
o connections associating emotions to events
o the sensitivity for responding
For the case study, the base network has the following
network connections (see the case description in Section 2.5
used as a point of departure):
srs
Complaints
?ps
GotoGP
complaints make preparing to go
to GP
srs
Complaints
,ms
Stigma
?bs
Stigma
complaints and memory
about stigma lead to the
belief state that confirming a mental problem provides a
stigma
bs
Stigma
?ds
AvoidConfirmation
belief bs
Stigma
leads to desire
state ds
AvoidConfirmation
to
Fig. 1. The overall four-level cognitive architecture for Self-referencing, Self-awareness and Self-interpretation.
130 J. Treur, G. Glas / Cognitive Systems Research 68 (2021) 125–142
avoid confirming mental problems
srs
Complaints
?bs
Irrelevant
complaints lead to the belief
bs
Irrelevant
that after a
while mental problems go away by themselves
bs
Irrelevant
,ds
AvoidConfirmation
?ps
AvoidMentalHelp
the desire
to avoid confirming mental
problems, and belief bs
Irrelevant
make preparing to avoid
mental help ps
AvoidMentalHelp
In Fig. 2 a graphical representation of these causal rela-
tions for the case study is depicted. For a brief explanation
of all states for the case study, see Table 3.
4.2. First reification level: Self-Referencing
At the first reification level a number of representation
states for a first-order self-model of the own base level
characteristics are included. These representations can be
for the physical world or (subjectively) relate to the per-
son’s mental processes.
Representation states W
X,Y
For the physical world, W-states are used as already
mentioned in Section 3.2 above, where W
X,Y
represents
the actual connection weight for the connection from sta-
te Xto state Y. As an example, Wsrs
Complaints
,bs
Stigma
indi-
cates the weight for the connection from srs
Complaints
to
bs
Stigma
. This explicit representation can be used for adapta-
tion, for example, based on Hebbian learning (Hebb, 1949).
Representation states RW
X,Y
Related to the individual, for example, the own known
(or believed) connection weight characteristics are repre-
sented by RW-states, where RW
X,Y
indicates the person’s
representation for the weight of the connection from base
state Xto base state Y. Here the prefix Rfor representation
is used to distinguish it from the actual value used in the
processing of the network model: the own representations
RW
X,Y
may be quite different from the ‘real’ values
W
X,Y
. This knowledge has to be acquired by experience,
which may depend on the situations the person actually
encounters. For example, for a never experienced situation
a person may not know at all what own responses will be
triggered by it. Examples for the case study are shown in
Table 2. Note that all these states in Table 2 in principle
may occur without being aware of them. Awareness states
will be addressed at the next level, in Section 4.3. In the
example model for the case study described below, for
the sake of simplicity only some of these states are
included, as the others do not play a role in the example
scenario addressed.
In Fig. 3, the reification states modeling the first-order
self-model by RW-states are depicted toegether with the
representation (or referencing) relations to the connections
they represent (dashed yellow lines). Here the base level
network characteristic for the weight x
X,Y
of a connection
from a base state Xto a base state Yis represented by a
first reification level state denoted by RW
X,Y
.
Note that these referencing relations are not used in the
processing of the model. Instead, upward causal connec-
tions from base states to first-order reification states repre-
senting their connections are used as depicted in Fig. 4.
Based on these connections, the first-order self-
referencing states can be learnt from experiences and in
that way get their values. For non-accessible or badly
accessible (blind spots) base states, these upward connec-
tions are weak or very weak, or even nonexistent.
Fig. 2. Base level connections for the considered case study.
Table 2
Examples of self-representation states RW and self-referencing states SRW.
J. Treur, G. Glas / Cognitive Systems Research 68 (2021) 125–142 131
Self-referencing states SRW
X,Y
Given an actually occurring mental state Y, this, together
with the relevant characteristics representation RW
X,Y
, may
trigger first-order self-referencing states SRW
X,Y
which indi-
cate the person knowing or believing that mental state Y
occurs and that the occurrence of this mental state Yrelates
to certain personal characteristics of him or herself. The
first-order self-referencing states considered here are
SRWX1;Y,...,SRWXK;Yfor personal characteristics in
the form of weights of the incoming connections of state Y
from states X
1
, .., X
k
, respectively
To generate these self-referencing states, causal connec-
tions are applied to SRWXi;Yboth from the considered base
state Yand from the relevant characteristic RWXi;Y:
Y?SRWXi;Y
RWXi;Y?SRWXi;Y
Table 3
States used in the computational case study and their explanation.
State nr State name Explanation Level
Partner
Com
p
laints
Stigma
Irrelevant
Stigma
Irrelevant
AvoidCon
f
irmation
GotoGP
AvoidMentalHel
p
WStigma,Stigma Sti
g
ma Sti
g
ma
WIrrelevant,Irrelevant Irrelevant Irrelevant
RW Complaints,Stigma Com
p
laints Sti
g
ma
RW Stigma,Stigma Sti
g
ma
Stigma
RW Irrelevant,Irrelevant Irrelevant Irrelevant
SRW Complaints,Stigma Stigma
Com
p
laints Sti
g
ma
SRW Stigma,Stigma Stigma
Sti
g
ma
S
ti
g
ma
SRW Irrelevant,Irrelevant Irrelevant
Irrelevant Irrelevant
Therapist
FSRW ,Stigma Sti
g
ma
AComfortingideas
SASRW Complaints,Stigma
Stigma
SASRW Stigma,Stigma
Stigma
SASRW Irrelevant,Irrelevant
Irrelevant
TSRW Complaints,Stigma SRW Com
p
laints,Sti
g
ma
TSRW Stigma,Stigma SRW Sti
g
ma,Sti
g
ma
TSRW Irrelevant,Irrelevant SRW Irrelevant,Irrelevant
Therapist
Sti
g
ma
FSASRW ,Sti
g
ma
TSASRW Stigma,Stigma
V-SASRW Stigma,Stigma Sti
g
ma
V+AComfortingideas
V-WStigma Sti
g
ma
TStigma Sti
g
ma
132 J. Treur, G. Glas / Cognitive Systems Research 68 (2021) 125–142
At the first reification level, from the base level network
characteristics not only connection weights can be explic-
itly represented by reification states, but also other charac-
teristics of the base level network, such as the excitability
threshold s
X
of a base state Xthat plays an important role
in sensitivity of responses of X. For example, the sensitivity
threshold s
bs
Stigma
of base state bs
Stigma
can be represented by
a representation state RT
bs
Stigma
and self-referencing state
SRT
bs
Stigma
at the first reification level. For the moment, in
this exploration we limit ourselves to reification of the con-
nection weights.
4.3. Second reification level: self-awareness
An awareness state of a state Yrefers to Yand therefore
can be modeled at one reification level higher than Y.An
awareness state for state Ycan lead to a change of network
characteristics related to Y, for example, its excitability
threshold or weights of incoming connections of Y.As
awareness states are modeled at one level higher than Y,
the effect that awareness strengthens learning can be mod-
eled by causal relations within that level. Via downward
causal connections, this can lead to a decreased excitability
threshold or increased weights of incoming connections
used for Y. Note that, if Yis mental state at the base level,
awareness of Yby itself is not considered self-awareness as
there is no self-referencing. However, awareness of a self-
referencing state for Y, such as SRWXi;Yindicating that Y
occurs (partly) because X
i
has an effect on Y, is modeled
by a self-awareness state SASRWXi;Y. As it refers to a level
1 state SRWXi;Y, this self-awareness state SASRWXi;Yis inher-
ently second-order. As a particular case, this can be applied
to an emotion state Y, in which case SASRWXi;Yindicates
that the person not only has unconscious self-knowledge
that state X
i
contributes to triggering of emotion Y, but
is also aware that (s)he has the characteristic or tendency
that state X
i
contributes to the triggering of emotion Y,
which indeed is a form of self-awareness.
Generation of any awareness state for a given state Yat
any level is modeled in a practical manner by applying a
combination of three principles occurring in multiple con-
sciouness theories:
focusing of attention; e.g., (Graziano, 2013; Graziano,
Guterstam, Bio, Wilterson, 2019)
a winner-takes-it-all competition between states; e.g.,
(Minsky, 1986; Baars, 1997; Graziano, Guterstam,
Bio, & Wilterson, 2019)
enhanced accessibility due to awareness; e.g., (Minsky,
1986; Baars, 1997; Graziano et al., 2019).
For the latter, an awareness state for Ymay amplify the
activation of Yvia changes of some of the network charac-
teristics related to Y, such as its incoming causal connec-
tions, or excitability threshold. By such mechanisms, the
second-order reification states may also affect the connec-
tions for the first-order reification states by which knowl-
edge on self-referentiality is obtained.
Fig. 3. First-order reification level states for the example and their representation or referencing relations.
Fig. 4. Base level and first-order reification level: upward interlevel connections for the example, assuming full accessibility of the base states (but no
accessibility of connections).
J. Treur, G. Glas / Cognitive Systems Research 68 (2021) 125–142 133
4.4. Third reification level: self-interpretation
Self-interpretation is considered here as analysis of the
own processes in relation to the person’s conscious self-
model. This conscious self-model is represented by the
self-awareness states at the second reification level. Such
an analysis refers to these self-awareness states and is there-
fore represented at the third reification level. Some of the
states at this third reification level are positive and negative
valuation states V
+Y
and V
Y
where Yis a second-order
state. For example, a state Ywith high negative valuation
can be considered undesirable. This enables the possibility
that due to the self-interpretation analysis, some changes in
the characteristics ‘embodying’ the second-order self-model
are made. For example, incoming connections of some of
the awareness states might get changed weights or the
excitability thresholds of some of them might change due
to the self-interpretation. But also characteristics of the
lower levels (first reification level, base level) may be
affected in a similar manner. For example, the self-
interpretation may reveal that there is an undesirable lack
of sensitivity of a certain state Y, and then via some form
of remedy that state Yis made more sensitive, modeled
by decrease of its excitability threshold. Or, the other
way around, an base state analysed as undesirable is
blocked or suppressed by increasing its excitability thresh-
old. Or, alternatively, a suppressing pathway to this state is
strengthened, like it can happen in emotion regulation.
Depending on how detailed such a remedy is modeled, its
effect can either be modeled as a direct influence from the
third to the first reification level representing the character-
istics, or via intermediate processes also involving the sec-
ond reification level: first an effect from the third on the
second reification level, which in turn has an effect on the
first reification level. More specific examples of this will
be shown in Section 5.
5. The four-level reified network model for the case study
In this section, the reified network model for the cogni-
tive architecture from Fig. 1 and the obtained computa-
tional model for the case study discussed in Section 2 is
presented in some more detail. For the sake of understand-
ing, the multilevel network model will be presented level by
level and the connectivity will be shown in Figs. 5–7
accordingly. In Table 3 an overview of the states used
and their explanation can be found.
Base level network states
The base network is as displayed in Fig. 2 above and dis-
cussed in Section 4.1. This base network represents the cli-
ent’s mental processes. Note that also input from the
client’s partner is incorporated, indicated by int0
Partner
.
This models that the partner tries to persuade the client
to go for mental help; it has a suppressing effect on
ps
AvoidMentalHelp
. The base network is depicted in Fig. 5 as
the lower plane.
For the first reification level RW-states are used as repre-
sentations for the own characteristics, in this case in the
form of the connections to the two belief states bs
Stigma
and bs
Irrelevant
; see Fig. 5. Based on these RW-states, at
the self-referencing SRW-states are generated that describe
that the respective belief state occurs due to the respective
personal characteristic by a connection (recall that SRW
X,Y
stands for a self-referencing state concerning state Yand
the connection from Xto Y).
Second reification level network states: self-awareness
Part of the self-referencing states SRW
X,Y
activate
awareness states SASRWX;Yfor them at the second reification
level, depending on attentional focusing and a winner-
takes-it-all-competition; see Fig. 6. Here the therapist initi-
ates intervention int1
Therapist
, which makes the person focus
(via focus state F
SRW
x,bsStigma
) on the belief state bs
Stigma
about stigma and possible relevant memories the person
has for it. This has no substantial effect yet. Also, part of
the therapist’s intervention int1
Therapist
is making the per-
son aware of some comforting ideas, for example, that
nowadays people are happy to openly talk about their
mental problems and are appreciated for that instead that
it harms them. This intervention 1 is aiming at two aware-
ness states to become active that were not active before the
intervention: A
Comfortingideas
and the awareness state about
having a memory connection SASRWmsStigma ;bsStigma . Only the
first one actually becomes active.
Third reification level network states: self-interpretation
Next, self-interpretation is modeled at the third reifica-
tion level; see Fig. 7. Here intervention int2
Therapist
is initi-
ated by the therapist by suggesting to take time to focus
and concentrate more (perhaps using some specific tech-
niques) on becoming aware of memories from the past
about a stigma (state F
SA
SRW
x,bs
Stigma
).
Fig. 5. Connectivity of the base level and first reification level (self-referencing) for the example.
134 J. Treur, G. Glas / Cognitive Systems Research 68 (2021) 125–142
This leads to the person opening up more for such an
awareness by decreasing the excitability threshold for it
via state T
SA
SRW
msStigma,bsStigma
and the pink downward
arrow from it. This has effect and the person becomes
aware of a memory that triggers his belief about a negative
stigma; so it is only now that self-awareness state SA
SRW
-
ms
Stigma
,bs
Stigma
becomes active.
Note that this state represents that the person is aware
that his belief about a stigma for mental problems has as
one of its triggers a connection (a personal characteristic)
from a bad memory from long ago when the attitude
toward mental problems in general was very different from
nowadays. Another process at the third reification level for
interpretation is valuation of the various activated aware-
ness states. This leads to negative valuation state (a V
-
state) of the belief state for stigma and a positive valuation
state (a V
+
-state) for the awareness of the comforting ideas.
These valuations lead to a state Tbs
Stigma
meant to suppress
this belief state bs
Stigma
by increasing its excitability thresh-
old. By a downward connection (the long pink arrow), this
Fig. 6. Connectivity of the base level and first (self-referencing) and second (self-awareness) reification level for the example.
Fig. 7. Connectivity of the overall network model. Base level and first (self-referencing states), second (self-awarenss states) and third (self-interpretation
states) reification level.
J. Treur, G. Glas / Cognitive Systems Research 68 (2021) 125–142 135
suppression is actually performed, so that now this belief
bs
Stigma
is not generated anymore.
6. Detailed specification
In this section the full details of the reified network
model for the cognitive architecture from Fig. 1 are dis-
cussed. In Section 4 the connectivity characteristics of the
network were shown in a graphical manner, but, for exam-
ple, no connection weights, nor details about the aggrega-
tion characteristics and timing characteristics were shown.
Here all three types of characteristics will be shown in full
detail in role matrix format, which makes the presented
work reproducible; see Box 1, 2 and 3.
Each role matrix addresses a specific role of the specified
values of the characteristics. In each role matrix, each state
X
i
in the network has its own row. In this row in the differ-
ent columns for the specific state X
i
the different causal
impacts are listed from this role. This format can be used
as input for the dedicated software environment that is
available and enables simulation. First, it is shown how
the graphical representation of Fig. 7 can be expressed in
role matrix mb for base connectivity. Next, to add more
detail, in role matrix mcw numbers for the connection
weights are specified. Both these role matrices mb and
mcw are shown in Box 1 and fully specify all connectivity
characteristics. Role matrix mb specifies the causal impacts
from a base role; these are the other states that have a basic
impact on the state X
i
addressed in that row. For example,
state X
5
(=bs
Stigma
), which is the belief state bs
Stigma
, gets
basic causal impact from the complaints representation
state X
2
(=srs
Complaints
) in the first column and the memory
state X
3
(=ms
Stigma
) in the second column. In matrix mcw,
from the connection weight role, two more causal impacts
on state X
3
(=bs
Stigma
) are specified, namely the causal
impact from the connection weight role for X
2
(=srs
Com-
plaints
) specified by the 1 in the first column of mcw and
for X
3
(=ms
Stigma
) specified by the 1 in the second column
of mcw. So, these numbers specify the basic connection
weights used for state X
5
(=bs
Stigma
). In this case, most
of these causal impacts are constant, but the connection
weights from the memory states to the belief states were
Box 1 Role matrices mb and mcw for the connectivity characteristics of the adaptive network model.
mb base
connectivity
Partner
Complaints
Stigma
Irrelevant
Stigma
Irrelevant
AvoidConfirmation
GotoGP
AvoidMentalHelp
WStigma,Stigma
WIrrelevant,Irrelevant
RW Complaints,Stigma
RW Stigma,Stigma
RW Irrelevant,Irrelevant
SRW Complaints,Stigma
SRW Stigma,Stigma
SRW Irrelevant,Irrelevant
Therapist
FSRW ,Stigma
AComfortingideas
SASRW Complaints,Stigma
SASRW Stigma,Stigma
SASRW Irrelevant,Irrelevant
TSRW Complaints,Stigma
TSRW Stigma,Stigma
TSRW Irrelevant,Irrelevant
Therapist
FSASRW ,Stigma
TSASRW Stigma,Stigma
V-SASRW Stigma,Stigma
V+AComfortingideas
V-WStigma
TStigma
mcw connection
weights
Partner
Complaints
Stigma
Irrelevant
Stigma
Irrelevant
AvoidConfirmation
GotoGP
AvoidMentalHelp
WStigma,Stigma
WIrrelevant,Irrelevant
RW Complaints,Stigma
RW Stigma,Stigma
RW Irrelevant,Irrelevant
SRW Complaints,Stigma
SRW Stigma,Stigma
SRW Irrelevant,Irrelevant
Therapist
FSRW ,Stigma
AComfortingideas
SASRW Complaints,Stigma
SASRW Stigma,Stigma
SASRW Irrelevant,Irrelevant
TSRW Complaints,Stigma
TSRW Stigma,Stigma
TSRW Irrelevant,Irrelevant
Therapist
FSASRW ,Stigma
TSASRW Stigma,Stigma
V-SASRW Stigma,Stigma
V+AComfortingideas
V-WStigma
TStigma
136 J. Treur, G. Glas / Cognitive Systems Research 68 (2021) 125–142
made adaptive by adding reification states X
10
(=W
ms
Stigma
,
bsStigma
) and X
11
(=W
ms
Irrelevant
,bsIrrelevant
) for them, so that
these weight numbers and their causal impact from that
role exerted on X
5
(=bs
Stigma
) and X
6
(=bs
Irrelevant
) changes
over time (modeled based on Hebbian learning). Note that
the values 0.5 in mcw for the mutual connections for X
21
to X
23
model the competition between these three self-
awareness states.
Note that in role matrix mb only the horizontal, intrale-
vel connections (black arrows) and upward interlevel con-
nections (blue arrows) from Fig. 7 are specified, as only
these indicate basic impact. The downward interlevel con-
nections (pink arrows) indicate a special effect according
to a different, non-basic role: the role played by the source
state, which is specified in the role matrix for that specific
role. Below, this will be discussed for the T-states X
33
(=TbsStigma ) and X
29
(=TASRWmsStigma;bsStigma ), that as reification
states play the role of (adaptive) threshold combination
function parameter for the excitability of the states X
5
(=bs
Stigma
) and X
22
(=SASRWmsStigma;bsStigma ). So, for example,
state X
5
(=bsStigma) has also a non-constant non-basic cau-
sal impact from that combination function parameter role,
specified in role matrix mcfp.
Next, the aggregation characteristics of the adaptive net-
work model are addressed. Four combination functions are
used within the network: alogistic,hebb,compid,stepmod.
In the combination function library used they have num-
bers 2, 3, 22, and 35, respectively, which is specified by
mcf = [2 3 22 35]
By specifying this mcf, within this specific network
model these combination functions become number 1 to
4, respectively. Each of the 33 states gets one of these com-
bination functions assigned, which is specified in role
matrix mcfw for combination function weights in Box 2.
Most of them get alogistic assigned, which is often used
as a kind of standard combination function. However,
the three states X
12
,X
13
,X
14
(RWsrsComplaints;bsStigma ,
RWmsStigma;bsStigma ,RWsrsComplaints ;bsStigma ) use the function hebb,
Box 2 Role matrices mcfw and mcfp for the aggregation characteristics of the adaptive network model.
mcfw combination
function weights alog-
istic
hebb comp-
id
step-
mod
Partner
Complaints
Stigma
Irrelevant
Stigma
Irrelevant
AvoidConfirmation
GotoGP
AvoidMentalHelp
WStigma,Stigma
WIrrelevant,Irrelevant
RW Complaints,Stigma
RW Stigma,Stigma
RW Irrelevant,Irrelevant
SRW Complaints,Stigma
SRW Stigma,Stigma
SRW Irrelevant,Irrelevant
Therapist
FSRW ,Stigma
AComfortingideas
SASRW Complaints,Stigma
SASRW Stigma,Stigma
SASRW Irrelevant,Irrelevant
TSRW Complaints,Stigma
TSRW Stigma,Stigma
TSRW Irrelevant,Irrelevant
Therapist
FSASRW ,Stigma
TSASRW Stigma,Stigma
V-SASRW Stigma,Stigma
V+AComfortingideas
V-WStigma
TStigma
mcfp combination
function parameters
alogistic hebb compid stepm od
Partner
Complaints
Stigma
Irrelevant
Stigma
Irrelevant
AvoidConfirmation
GotoGP
AvoidMentalHelp
WStigma,Stigma
WIrrelevant,Irrelevant
RW Complaints,Stigma
RW Stigma,Stigma
RW Complaints,Irrelevant
SRW Complaints,Stigma
SRW Stigma,Stigma
SRW Complaints,Irrelevant
Therapist
FSRW ,Stigma
AComfortingideas
SASRW Complaints,Stigma
SASRW Stigma,Stigma
SASRW Irrelevant,Irrelevant
TSRW Complaints,Stigma
TSRW Stigma,Stigma
TSRW Irrelevant,Irrelevant
Therapist
FSASRW ,Stigma
TSASRW Stigma,Stigma
V-SASRW Stigma,Stigma
V+AComfortingideas
V-WStigma
TStigma
J. Treur, G. Glas / Cognitive Systems Research 68 (2021) 125–142 137
which as a form of learning from experience allows them to
acquire their values from simultaneous activation of the
two connected states.
Note that, as these RW-states describe the person’s
knowledge or beliefs on the connection strengths; they
are not used for the mental processing as those weights
actually have the constant values 1 specified in role matrix
mcw in Box 1, whereas the person’s beliefs can concern dif-
ferent values. For example, if the person as a kind of blind
spot about some effect, she or he can believe that the con-
nection has a very low value (low value of the RW-state),
whereas the value as specified in mcw is actually high.
The two states X
29
,X
32
(TASRWmsStigma;bsStigma ,V
Wx,
bs
Stigma
)
use the combination function compid which gives a comple-
mental effect by mapping any incoming impact Vto 1-V.
Finally, the interventions by the therapist X
18
,X
27
(in-
t1
Therapist
, int2
Therapist
) are considered external events for
the person and therefore they use the combination function
stepmod by which the timing of the event is specified.
Combination functions often have parameters, which
form yet another type of the network’s aggregation charac-
teristics which exert causal impacts from a different role.
These are specified in role matrix mcfp for combination
function parameter values:
Combination function alogistic has a steepness parame-
ter rand a threshold parameter s.
Combination function hebb has one parameter lfor the
persistence rate.
Combination function stepmod has a parameter qfor
repetition time of the event and dfor the duration of each
occurrence of the event.
Note that the two adaptive threshold values for states
X
5
(= bsStigma) and X
22
(= ASRWmsStigma;bsStigma ;bsStigma ) do not
have a fixed value specified in role matrix mcfp, but instead
their cells display a reference to another state, X
33
(=
T
bs
Stigma
) resp. X
29
(= TSASRWmsStigma ;bsStigma ), that represents the
dynamic value of this parameter. In this way, the adaptive-
ness of these network characteristics is specified explicitly
within the model by using their reification states; in the
graphical connectivity pictures this relates to the down-
ward pink arrows. For example, as a result of focusing
attention on SASRWmsStigma;bsStigma via FSASRWx;bsStigma ,byX
29
(=
TSASRWmsStigma ;bsStigma ) it is arranged that the threshold of
SASRWmsStigma ;bsStigma is decreased, so that self-awareness shifts
to it. After getting self-awareness SASRWmsStigma ;bsStigma , the
threshold TSRWmsStigma ;bsStigma (= X
25
) of self-referencing state
SRWmsStigma;bsStigma is decreased, due to which by the aware-
ness the latter state becomes more active and therefore also
more accessible for any other state. Note that another,
equally feasible, option to model this enhanced accessibility
could be by increasing the weights of the outgoing connec-
tions from SRWmsStigma;bsStigma .
For the timing characteristics of the network model,
another role matrix is used: the speed factor role matrix
ms (see Box 3). These speed factors indicate how fast a state
value changes upon causal impact received.
7. Example simulation for the case study
Based on the specification of the network model shown
in Box 1 to 3 in Section 6 and the dedicated software envi-
ronment described in (Treur, 2020b, Ch 9), a simulation
has been generated for the case study. The overall outcome
is shown in Fig. 8. The vertical lines indicate that interven-
tion 0 by the partner took place at time 40, intervention 1
by the therapist at time 60 and intervention 2 at time 80.
Moreover, at time 20 the complaints start. It can be seen
that intervention 0 only has a minor effect and intervention
2 has a major effect. However, with 33 lines in one graph,
this is not easy to understand in more detail. Therefore,
in the next 4 Figs. 9–12, parts are shown.
In Fig. 9 the focus is on the base states, with also the
interventions visible. After the complaints start at time
20, the preparation to go to a GP becomes high (the dark
blue line). It can also be seen that after intervention 0 by
the partner the being prepared not to go for mental help
(the brown line) decreases until below 0.7, after interven-
tion 1 the level of the belief state bs
Stigma
decreases very
lightly (the red line), but still stays very high. However,
after intervention 2 the belief state bs
Stigma
goes down from
above 0.95 to just above 0.6. Hand in hand with this also
the desire state ds
AvoidConfirmation
(the green line) and the
Box 3
Role matrix ms for the timing characteristics of
the adaptive network model and the initial values iv.
138 J. Treur, G. Glas / Cognitive Systems Research 68 (2021) 125–142
preparation state ps
AvoidMentalHelp
(the brown line) go down
to below 0.4 and 0.2, respectively.
To see how these effects were achieved, in the next fig-
ures we zoom in on the higher level states, first in Fig. 10
on the self-referencing states covering of the first-order
internal self-model. In the first phase between 0 and 20 it
is shown how the connections to the belief states bs
Stigma
and bs
Irrelevant
are learned, shown by their reification states
W
ms
Stigma,
bs
Stigma
and W
ms
Irrelevant,
bs
Irrelevant
(dark green line and
brown line both starting at 0.1). It can be seen that initially
already representations RW
srs
Complaints,
bs
Stigma
,RW
srs
Complaints
,bs
Irrel-
evant
and RW
ms
Stigma
,bs
Stigma
, were present of the connections,
with activation levels 0.2, 0.25, and 0.3, but these do not
correspond exactly to the actual values which are higher:
the person has a first-order model on the connections,
but, as more often, such a model is not exactly the reality.
However, due to experiencing their effects, these values
finally increase to 1, what is indeed their actual value. As
belief state bs
Stigma
also becomes high (red line), together
with the above RW-states this activates the self-
referencing states SRW
srs
Complaints
,bs
Stigma
,SRW
ms
Stigma
,bs
Stigma
,
SRW
srs
Complaints
,bsIrrelevant
(initially all 0) which relates the
occurrence of belief states bs
Stigma
and bs
Irrelevant
to their
causing state bs
Stigma
.
All of the above states do not involve self-awareness yet.
The self-awareness states are described in Fig. 11. In the
first phase there is practically no self-awareness
SA
SRWms
Stigma
,bs
Stigma
of the role of the memory state ms
Stigma
in causing belief state bs
Stigma
(see the grey line). During
this phase, self-awareness focuses more on the belief bs
Irrel-
evant
(the light brown line peaking just aobve 0.7 around
time 60). Being awareness-states, these self-awareness
states mutually suppress each other, where apparently
SA
SRWms
Complaints
,bs
Irrelevant
is the first winner of the competition.
Intervention 1 at time 60 brings a focus state F
SRW
x,bsStigma
(light green line) that results in a small but still insufficient
change for the awareness up till time 80.
However, intervention 2 at time 80 brings about a more
serious change, so that after that SA
SRW
msStigma,bsStigma
becomes
the winner of the competition (the steep upward grey line).
It can be seen that the strong increase of the latter state
goes together with a substantial decrease for the belief state
bs
Stigma
. How that can happen is explained by the next
graph in Fig. 10 involving self-interpretation.
In Fig. 12 it can be seen that at this self-interpretation
level, by intervention 2 many things change strongly. First,
the focus F
SASRWx,
bsStigma
on the causes of belief state bs
Stigma
initiated by the therapist becomes active (the yellow line,
ending above 0.9). This leads to a strong decrease of state
T
SASRWms
Stigma
,bs
Stigma
(the steep downward blue-grey line)
indicating the (previously high) excitability threshold which
was blocking until then the rise of the awareness state for
SRW
ms
Stigma
,bs
Stigma
. As this threshold becomes low,
SA
SRWms
Stigma
,bs
Stigma
now gets a chance to increase, which
indeed happens as already was observed in Fig. 11, grey
line. Moreover, as part of the analysis for self-
interpretation, a negative valuation V
SA
SRW
msStigma
,bsStigma
develops (the steep upward light green line). Through this
negative valuation, the person generates the state T
bs
Stigma
(steep upward dark green line, ending at 1) meant to help
block or at least decrease the belief state bs
Stigma
as a form
of self-control. This is indeed the reason that the belief state
goes down in this stage (through the long pink downward
link in Fig. 7), which explains what was already observed in
the previous figures.
8. Discussion
In this paper, a multilevel cognitive architecture is intro-
duced that can be used to model mental processes in clients
of psychotherapeutic sessions. The architecture does not
only cover base level mental processes but also mental pro-
cesses involving self-referencing, self-awareness and self-
interpretation; e.g., (Glas, 2017; Glas, 2019; Glass, 2020).
To this end, the cognitive architecture was designed
according to four levels, where (part of) the structure of
each level is represented by an explicit self-model of it at
the next-higher level of the architecture. At that next-
higher level, states reify part of the structure of the level
below; these states represent this structure and have a ref-
erencing relation to it. In this way the overall architecture
includes its own overall self-model, which accordingly is
also multilevel.
Fig. 8. Overall outcome of the simulation.
J. Treur, G. Glas / Cognitive Systems Research 68 (2021) 125–142 139
Fig. 9. The interventions and the effects for the base states.
Fig. 10. The effects for the self-referencing states.
Fig. 11. The effects for the self-awareness states.
140 J. Treur, G. Glas / Cognitive Systems Research 68 (2021) 125–142
The cognitive architecture was evaluated for a case
study of a realistic type of therapeutic session from clinical
practice. The relevant aspects of this case study concerning
self-referentiality, self-awareness and self-interpretation
were all covered well by applying the architecture.
The cognitive architecture was specified formally using
the role matrix language format for reified temporal-
causal networks described in (Treur, 2020b), which extends
the approach introduced in (Treur, 2016a, 2016b) by add-
ing network reification and a systematic approach to
(multi-order) adaptive networks, introduced in (Treur,
2018, 2020a). Using the available dedicated software envi-
ronment, the simulations were conducted automatically
with this formal specification as input.
It has been shown how by the cognitive architecture self-
awareness states can be modeled, taking into account three
wellknown principles occurring in multiple theories of con-
sciousness: the role of attention, a winner-takes-it-all prin-
ciple, and enhanced accessibility; e.g., (Graziano, 2013;
Graziano et al., 2019; Baars, 1997). In addition, the leveled
structure of the cognitive architecture has some relation to
higher-order theories of consciousness; e.g., (Metzinger,
2004; Metzinger, 2007; Rosenthal, 2005).
In future work, a number of issues can be addressed fur-
ther. One of them is to model examples of therapeutic ses-
sions in which emotions play a main role; e.g., (Glas, 2020).
Also other case studies from clinical practice will be consid-
ered for further evaluation. Further future development
may consider to use the architecture for virtual training sit-
uations for therapists.
Declaration of Competing Interest
The authors declare that they have no known competing
financial interests or personal relationships that could have
appeared to influence the work reported in this paper.
References
Atkinson, A. O., & Ratcliffe, M. (2012). Introduction to the special section
on ‘‘emotions and feelings in psychiatric illness.Emotion Review, 4(2),
119–121.
Baars, B. J. (1997). In the theater of consciousness: The workspace of the
mind. Oxford: Oxford University Press.
Bowen, K. A., & Kowalski, R. (1982). Amalgamating language and meta-
language in logic programming. In K. Clark & S. Tarnlund (Eds.),
Logic programming (pp. 153–172). New York: Academic Press.
Chandra, N., & Barkai, E. (2018). A non-synaptic mechanism of complex
learning: Modulation of intrinsic neuronal excitability. Neurobiology of
Learning and Memory, 154, 30–36.
Christoff, K., Cosmelli, D., Legrand, D., & Thompson, E. (2011).
Specifying the self for cognitive neuroscience. Trends in Cognitive
Sciences, 15(3), 104–112.
Damasio, A. R. (1999). The feeling of what happens. Body and emotion in
the making of consciousness. San Diego: Harcourt.
Damasio, A. R. (2010). Self comes to mind: Constructing the conscious
brain. New York: Pantheon Books.
Descartes R (1644) Principles of philosophy (translated by MS Mahoney).
Fessler,D. M. T., Clark, J. A., & Clint,E. K. (2015). Evolutionary psychology
and evolutionary anthropology. In D. M. Buss (Ed.), The handbook of
evolutionary psychology (pp. 1029–1046). Wiley and Sons.
Fessler, D. M. T., Eng, S. J., & Navarrete, C. D. (2005). Elevated disgust
sensitivity in the first trimester of pregnancy: Evidence supporting the
compensatory prophylaxis hypothesis. Evolution & Human Behavior,
26(4), 344–351.
Fonagy, P., Gergely, G., Jurist, E. L., & Target, M. (Eds.). (2002). Affect
regulation, mentalization, and the development of the self. New York:
Other Press.
Gallagher, S. (2013). A pattern theory of self. Frontiers in Human Neuroscience,
7, e443, 1-7 [online: 01 August 2013 doi: 10.3389/fnhum.2013.00443].
Galton, A. (2006). Operators vs arguments: The ins and outs of reification.
Synthese, 150, 415–441.
Gascoigne, N., & Thornton, T. (2014). Tacit knowledge. London & NY:
Routledge.
Glas, G. (2017). Dimensions of the self in emotion and psychopathology.
Philosophy, Psychiatry, & Psychology, 24(2), 143–155.
Glas, G. (2019). Person-centered care in psychiatry. Selfrelational,
contextual, and normative perspectives. Abingdon/London: Routledge.
Glas, G. (2020). An enactive approach to anxiety and anxiety disorders.
Philosophy, Psychiatry, & Psychology, 27(1), 35–50.
Graziano, M. S. A. (2013). Consciousness and the social brain. New York:
Oxford University Press.
Graziano, M. S. A., Guterstam, A., Bio, B. J., & Wilterson, A. I. (2019).
Toward a standard model of consciousness: Reconciling the attention
schema, globalworkspace, higher-order thought, and illusionist theories.
Cognitive Neuropsychology, 37(3–4), 155–172. https://doi.org/10.1080/
02643294.2019.1670630.
Gross, J. J. (2008). Emotion regulation. In M. Lewis, J. M. Haviland-
Jones, & L. F. Barrett (Eds.), Handbook of emotions (3rd ed.,
pp. 497–512). New York: Guilford.
Fig. 12. The effects for the self-interpretation states.
J. Treur, G. Glas / Cognitive Systems Research 68 (2021) 125–142 141
Gross, J. J., & Jazaieri, H. (2014). Emotion, emotion regulation, and
psychopathology: An affective science perspective. Clinical Psycholog-
ical Science, 2, 387–401.
Gross, J. J., & Thompson, R. A. (2007). Emotion regulation: Conceptual
foundations. In J. J. Gross (Ed.), Handbook of emotion regulation
(pp. 3–24). New York: Guilford.
Hebb, D. O. (1949). The organization of behavior: A neuropsychological
theory. Wiley and Sons.
Hobson, P. (2010). Emotion, self/other-awareness, and autism: A devel-
opmental perspective. In P. Goldie (Ed.), The oxford handbook of
philosophy and emotion (pp. 445–472). Oxford: Oxford University Press.
Iacoboni, M. (2008). Mirroring people: The new science of how we connect
with others. New York: Farrar, Straus & Giroux.
Immordino-Yang, M. H. (2011). Me, my ‘‘selfand you: Neuropsycho-
logical relations between social emotion, self-awareness, and morality.
Emotion Review, 3(3), 313–315.
Kappas, A. (2011). Emotion and regulation are one! Emotion Review, 3(1),
17–25.
Keysers, C., & Gazzola, V. (2014). Hebbian learning and predictive mirror
neurons for actions, sensations and emotions. Philosophical Transactions
of the Royal Society of London. Series B, Biological sciences, 369,
20130175.
Kim, J. (1996). Philosophy of mind. Westview Press.
Koole, S. (2009). The psychology of emotion regulation: An integrative
review. Cognition and Emotion, 23(1), 4–41.
Leary, M. R., & Tangney, J. P. (2003). The self as an organizing construct
in the behavioural and social sciences. In M. R. Leary & J. P. Tangney
(Eds.), Handbook of self and identity (pp. 3–14). New York/London:
The Guilford Press.
Leary, M. R. (2007). Motivational and emotional aspects of the self.
Annual Review of Psychology, 58, 317–344.
Leibniz, G. W. (1698). Phoranomus seu De potentia et legibus naturæ.
Dialogus II Republished in (1991). Physis, 28, 797–885.
Levy, D. A., & Nail, P. R. (1993). Contagion: A theoretical and empirical
review and reconceptualization. Genetic, Social, and General Psychol-
ogy Monographs, 119(2), 233–284.
Lorenz, E. N. (1963). Deterministic nonperiodic flow. Journal of the
Atmospheric Sciences, 20, 130–141.
Lorenz, E. N. (1993). 1993 The Essence of Chaos. University of Washington
Press.
McPherson, M., Smith-Lovin, L., & Cook, J. M. (2001). Birds of a
feather: Homophily in social networks. Annu Rev Sociol, 27,
415–444.
Metzinger, T. (2004). Being no one: The self-model theory of subjectivity.
MIT Press.
Metzinger, T. (2007). Empirical perspectives from the self-model theory of
subjectivity: a brief summary with examples. In: R. Banerjee & B.K.
Chakrabarti (Eds.), Models of brain and mind: Physical, computa-
tional and psychological approaches. Progress in Brain Research, vol.
168, pp. 215-245, 273-278. Elsevier.
Montgomery, K. (2006). How doctors think: Clinical judgment and the
practice of medicine. Oxford: Oxford University Press.
Minsky, M. (1986). The society of mind. New York: Simon & Schuster.
Nagel, E., & Newman, J. (1965). Go
¨del’s proof. New York: New York
University Press.
Newton, I. (1729). The mathematical principles of natural philosophy;
Newton’s Principles of Natural Philosophy, Dawsons of Pall Mall, 1968
Nicolini, D. (2012). Practice theory, work, and organization. An introduc-
tion. Oxford: Oxford University Press.
Northoff, G., Qin, P., & Feinberg, T. E. (2011). Brain imaging of the self
Conceptual, anatomical and methodological issues. Consciousness and
Cognition, 20, 52–63.
Prinz, J. (2010). The moral emotions. In P. Goldie (Ed.), The Oxford
handbook of philosophy and emotion (pp. 519–538). Oxford: Oxford
University Press.
Ratcliffe, M. (2005). The feeling of being. Journal of Consciousness Studies,
12, 43–60.
Ratcliffe, M. (2008). Feelings of being. Phenomenology, psychiatry and the
sense of reality. Oxford: Oxford University Press.
Ratcliffe, M. (2010). The phenomenology of mood and the meaning of life.
In P. Goldie (Ed.), The oxford handbook of philosophy and emotion
(pp. 349–372). Oxford: Oxford University Press.
Reddy, W. M. (2009). Historical research on the self and emotions.
Emotion Review, 1(4), 302–315.
Ricoeur, P. (1992) (original work published in 1990). Oneself as another
(trans. Kathleen Blamey). Chicago: University of Chicago.
Rorty, A. (2010). A plea for ambivalence. In P. Goldie (Ed.), The oxford
handbook of philosophy and emotion (pp. 425–444). Oxford: Oxford
University Press.
Rosenthal, D. (2005). Consciousness and mind. Oxford: Oxford University Press.
Slaby, J., & Stephan, A. (2008). Affective intentionality and self-
consciousness. Consciousness and Cognition, 17, 506–513.
Solomon, R. C. (1983). The passions. Notre Dame: University of Notre
Dame Press.
Stephan, A. (2012). Emotions, existential feelings, and their regulation.
Emotion Review, 4(2), 157–162.
Smorynski, C. (1977). The incompleteness theorems. In J. Barwise (Ed.).
Handbook of mathematical logic (vol. 4, pp. 821–865). Amsterdam:
North-Holland.
Sterling, L., & Beer, R. (1989). Metainterpreters for expert system
construction. Journal of Logic Programming, 6, 163–178.
Stern, D. N. (2004). The present moment in psychotherapy and everyday
life. New York & London: Norton.
Tracy, J. L., Klonsky, E. D., & Proudfit, G. H. (2014). How affective
science can inform clinical science. An introduction to the special series
on emotions and psychopathology. Clinical Psychological Science,
2(4), 371–386.
Tracy, J. L., & Robins, R. W. (2007). In the self in social psychology. In C.
Sedikides & S. Spence (Eds.), Frontiers of social psychology series
(pp. 187–209). New York: Psychology Press.
Treur, J. (2016a). Dynamic modeling based on a temporal–causal network
modeling approach. Biologically Inspired Cognitive Architectures, 16,
131–168.
Treur, J. (2016b). Network-oriented modeling: Addressing complexity of
cognitive, affective and social interactions. Springer Nature Publishers,
Cham, Switzerland
Treur, J. (2018) Network reification as a unified approach to represent
network adaptation principles within a network. In: Martı
´n-Vide, C.,
Vega-Rodrı
´guez, M. A., Fagan, D., O’Neill, M. (Eds.), Proc of the 7th
International Conference on Theory and Practice of Natural Com-
puting, TPNC’18, (pp. 344-358). Lecture Notes in Computer Science,
vol 11324, Springer Nature Publishers.
Treur, J. (2020a). Modeling higher-order adaptivity of a network by
multilevel network reification. Network Science, 8(S1), S110–S144.
Treur, J. (2020b). Network-oriented modeling for adaptive networks:
Designing higher-order adaptive biological, mental and social network
models. Switzerland: Springer Nature Publishers, Cham.
Tse, P. U. (2013). The neural basis of free will: Criterial causation.
Cambridge: MIT Press.
Westerhoff, H. V., He, F., Murabito, E., Cre
´mazy, F., & Barberis, M. (2014).
Understanding principles of the dynamic biochemical networks of life
through systems biology. In A. Kriete & R. Eils (Eds.), Computational
systems biology (2nd ed., pp. 21–44). Oxford: Academic Press.
Westerhoff, H. V., Brooks, A. N., Simeonidis, E., Garcı
´a-Contreras, R., He,
F., Boogerd, F. C., ... Kolodkin, A. (2014). Macromolecular networks
and intelligence in microorganisms. Frontiers in Microbiology, 5 e379.
Weyhrauch, R. W. (1980). Prolegomena to a theory of mechanized formal
reasoning. Artificial Intelligence, 13, 133–170.
Zahavi, D. (2005). Subjectivity and selfhood. Investigating the first-person
perspective. Cambridge Mass: MIT Press.
142 J. Treur, G. Glas / Cognitive Systems Research 68 (2021) 125–142
... At that next-higher level, states represent this structure and have a referencing relation to it. Most material is based on (Treur and Glas 2021). The self-modeling network model shows how a person can have mental self-models of him-or herself of different levels of description. ...
Chapter
In this chapter, a multilevel cognitive architecture is introduced that can be used to model mental processes in clients of psychotherapeutic sessions and in particular the mental (self-)models they have about themselves. The architecture does not only cover base level mental processes but also mental processes involving self-referencing, self-awareness and self-interpretation. To this end, the cognitive architecture was designed according to four levels, where (part of) the structure of each level is represented by an explicit self-model of it at the next-higher level of the architecture. At that next-higher level, states represent part of the structure of the level below; these states have a referencing relation to it. In this way the overall architecture includes its own overall self-model. The cognitive architecture was evaluated for a case study of a realistic type of therapeutic session from clinical practice.KeywordsNetwork model Third-order adaptive Self-referencingSelf-awarenessSelf-interpretation
Chapter
This chapter contributes an analysis of how in mental and social processes, humans often apply specific mental models and learn and adapt them in a controlled manner. It is discussed how controlled adaptation relates to the Plasticity Versus Stability Conundrum in neuroscience. From the analysis an informal three-level cognitive architecture for controlled adaptation was obtained. It is discussed here from a self-modeling network viewpoint how this cognitive architecture can be modeled as a self-modeling network. Making use of the specific network characteristics offered by the self-modeling network structure format, a large number of options for different types of adaptation of mental models and different types of control over adaptation of mental models were obtained. Many of these options were illustrated by a several realistic examples that were formalized by self-modeling networks. Other options that were distinguished from the analysis here, are offered as interesting options for future research.KeywordsDynamics AdaptationControlMental models
Chapter
This paper contributes an analysis of how in mental and social processes, humans often apply specific mental models and learn and adapt them in a controlled manner. It is discussed how controlled adaptation relates to the Plasticity Versus Stability Conundrum in neuroscience. From the analysis an informal three-level cognitive architecture for controlled adaptation was obtained. It is discussed here from a self-modeling network viewpoint how this cognitive architecture can be modeled as a self-modeling network. Making use of the specific network characteristics offered by the self-modeling network structure format, a large number of options for different types of adaptation of mental models and different types of control over adaptation of mental models were obtained. Many of these options were illustrated by a several realistic examples that were formalized by self-modeling networks. Other options that were distinguished from the analysis here, are offered as interesting options for future research.
Article
The paper focuses on the importance of psychological analysis of socio-cognitive architectures for the development of artificial intelligence technologies that can fulfill communicative needs and expectations of people. The structure of mental acts of social goal-setting is analyzed, including metaprocesses, while the author develops the idea of a person’s metacognitive ability as a property and an integral part of the cognitive system. The role of cognitive architectures of social goal-setting in natural speech communication is described. The described situations of social goal-setting refer to the different periods of ontogenesis - childhood, adolescence, and adulthood.
Full-text available
Conference Paper
For a video presentation of this paper, see https://www.youtube.com/watch?v=-C73Y76OpIs. This paper contributes an analysis of how in mental and social processes, humans often apply specific mental models and learn and adapt them in a controlled manner. It is discussed how controlled adaptation relates to the Plasticity Versus Stability Conundrum in neuroscience. From the analysis an informal three-level cognitive architecture for controlled adaptation was obtained. It is discussed here from a self-modeling network viewpoint how this cognitive architecture can be modeled. Making use of the specific network characteristics offered by the self-modeling network structure format, a large number of options for different types of adaptation of mental models and different types of control over adaptation of mental models were obtained. Many of these options were illustrated by several realistic examples that were formalized by self-modeling networks. Other options that were distinguished from the analysis here, are offered as interesting options for future research.
Full-text available
Article
Here we examine how people’s understanding of consciousness may have been shaped by an implicit theory of mind. This social cognition approach may help to make sense of an apparent divide between the physically incoherent consciousness we think we have and the complex, rich, but mechanistic consciousness we may actually have. We suggest this approach helps reconcile some of the current cognitive neuroscience theories of consciousness. We argue that a single, coherent explanation of consciousness is available and has been for some time, encompassing the views of many researchers, but is not yet recognized. It is obscured partly by terminological differences, and partly because researchers view isolated pieces of it as rival theories. It may be time to recognize that a deeper, coherent pool of ideas, a kind of standard model, is available to explain multiple layers of consciousness and how they relate to specific networks within the brain.
Full-text available
Article
For related videos, see the YouTube channel on Self-Modeling Networks here: https://www.youtube.com/channel/UCCO3i4_Fwi22cEqL8M_PgeA. In network models for real-world domains often network adaptation has to be addressed by incorporating certain network adaptation principles. In some cases, also higher-order adaptation occurs: the adaptation principles themselves also change over time. To model such multilevel adaptation processes it is useful to have some generic architecture. Such an architecture should describe and distinguish the dynamics within the network (base level), but also the dynamics of the network itself by certain adaptation principles (first-order adaptation level), and also the adaptation of these adaptation principles (second-order adaptation level), and maybe still more levels of higher-order adaptation. This paper introduces a multilevel network architecture for this, based on the notion network reification. Reification of a network occurs when a base network is extended by adding explicit states representing the characteristics of the structure of the base network. It will be shown how this constru