1. Introduction
“Knowledge is Power” - Francis Bacon.
Engineering problems cannot be solved without applying knowledge.
Consequently, knowledge-intensive activities, such as knowledge
acquisition, representation, dissemination, utilization, and management,
play a vital role in engineering problem solving. The advent of systems
engineering (de Weck, 2018), engineering informatics (Tomiyama et al.,
2012), as well as the fourth industrial revolution (smart manufacturing,
Industry 4.0) (Zhou et al., 2018; Sinclair et al., 2019; Ullah, 2019)
has added a new dimension—digitization of knowledge-intensive
activities by applying advanced computing as well as information and
communication technologies. As far as Industry 4.0 is concerned, it
(Industry 4.0) employs some embedded systems (e.g., cyber-physical
systems) to perform such cognitive tasks as monitoring, understanding,
predicting, deciding, acting, and adapting (Zhou et al., 2018; Sinclair
et al., 2019; Ullah, 2019). Without applying digitized knowledge, these
systems cannot work. As a result, the digitization of
knowledge-intensive activities (knowledge acquisition, representation,
dissemination, utilization, and management) is critical for Industry
4.0. Before developing methods and tools, which are needed for achieving
the desired level of digitization of knowledge-intensive activities in
Industry 4.0, the following questions must be addressed. What is
knowledge? What are the types of knowledge? How to create knowledge? How
to represent knowledge? What is the difference between data/information
and knowledge? What is the role of human cognition in knowledge
formation? What is the role of experience in knowledge formation? Is the
attainment of “true” knowledge possible? Is analytical knowledge
better than experiential knowledge?
The abovementioned questions are difficult to answer since a relatively
unambiguous and circularity-free definition of knowledge is not yet
available. This can be understood from the commonly used definitions of
knowledge, as described below. A compressive account on the definition
of knowledge is presented in Section 2.
Consider the following three general views regarding knowledge. 1)
First, consider the most general view regarding knowledge, that is, a
piece of knowledge is a proposition that corresponds to justified true
belief (Gettier, 1963). The process of justifying the truthfulness of a
belief involves several intellectual resources, and the process capable
of making justification of a belief possible many not be known
beforehand. Thus, “true knowledge” may not exist. 2) Secondly,
consider the dictionary meaning of knowledge. For example, a
dictionary-based definition describes knowledge as an awareness,
understanding, or information, which either resides within a person’s
mind or is possessed by people and can only obtained via experience or
investigation (Knowledge definition and meaning , 2019). This is a
rather broad definition of knowledge involving other concepts requiring
prior definition. 3) Lastly, consider the definitions of knowledge given
by legislative bodies. For example, the European Union defines knowledge
as facts, principles, theories, and practices accumulated by learning;
both cognitive reflections and direct experiences of individuals or
groups contribute to the body of knowledge (Abele et al., 2017).
The remarkable thing is that all definitions of knowledge (including
those presented in Section 2) are based on several concepts. For
example, the last definition mentioned above associates concepts, such
as learning, cognitive reflection, direct experience, fact, principle,
theory, and practice, to define the knowledge. Such concepts must be
defined before defining knowledge. This results in a phenomenon called
circularity that must be avoided while defining knowledge (Zagzebski,
1999). Therefore, defining knowledge in clear terms, at the same time
avoiding circularly, is a challenging task. This paper aims to present a
circularity-free and unambiguous definition of knowledge that can help
build knowledge-based systems from the context of Industry 4.0.
The remainder of this paper is organized as follows. Section 2 presents
a comprehensive review of definitions of knowledge reported in extant
literature concerning epistemology, engineering design, manufacturing,
as well as organization science, education science, and information
science. Section 3 presents a revised definition of knowledge along with
its different types and categories. Section 4 describes the types and
categories of knowledge presented in Section 3 using some real-life
examples. The representation of knowledge using knowledge graphs
(concept maps) is also presented in Section 4. Section 5 discusses the
implications of this study by demonstrating the existence of different
types and categories of knowledge in a creative design process. This
section also suggests a framework for developing knowledge-based systems
for the advancement of Industry 4.0. Section 6 presents the concluding
remarks drawn from this study.
2. Literature Review
Three commonly known definitions of knowledge are very briefly presented
in the previous section. However, the concept of knowledge and its
definitions have been studied by many stakeholders at great depth. This
section thus, first, reviews the definitions of knowledge found in
epistemology. Subsequently, it reviews the definitions of knowledge
found in the literature of engineering design, manufacturing, and other
relevant fields such as organization, information, and education
sciences.
2.1. Epistemology
Epistemology is the philosophical study that deals with the nature,
origin, and formulation of knowledge irrespective of the academic
discipline (Sosa, 2017; Steup, 2018). The definition of knowledge in
epistemology exhibits multiplicity, which has lasted since the period of
Aristotle. Multiplicity originates from such metaphysical concepts as
idealism, rationalism, empiricism, neutralism, pragmatism or
evolutionism, and explanationism. Each metaphysical concept corresponds
to certain truths that manifest knowledge. In particular, idealism
considers there exist unquestionable and transcendental truths that are
entirely independent of experiences. Rationalism considers there exist
rational processes that are somewhat independent of experiences, thereby
leading to some truths. Empiricism considers all truths to be dependent
on experiences; that is, the experience is the sole driver that
contributes to knowledge formation. Pragmatism adopts a skeptic or
evolutionary view toward truth; that is, usefulness of the perceived
truth determines its fate—whether or not it will be considered a piece
of knowledge. Consequently, truthfulness may vary with time. Neutralism
is similar to pragmatism, and considers that while finding truth, any
metaphysical concept from amongst idealism, rationalism, and empiricism
can be used. This implies that truth is not biased to a specific
metaphysical concept, and that any combination of metaphysics can be
used to formulate knowledge. Explanationism considers that a so-called
scientific truth evolves in accordance with the deductive–nomological
(D–N) explanation, inductive–statistical (I–S) explanation, or
statistical–relevance (S–R) explanation (Hempel, 1968; Salmon et al.,
1971; Shrader, 1977; Salmon, 2006; Woodward 2017).
In classical epistemology, definitions of knowledge proposed by Hume and
Kant have attracted significant attention. According to Hume, knowledge
corresponds to two propositions—relations of ideas and matters of fact
(Locke et al., 1960). Relations of ideas are a priori non-falsifiable
propositions (e.g., a triangle has three sides, summation of all
included angles of a triangle equals 180°, etc.). Matters of fact are
experience-dependent propositions that can be falsified if a
counterexample is available (e.g., apples are good for health, bachelors
are messy, etc.). Kant, on the other hand, analyzed the work of Hume and
proposed there exist three types of knowledge—analytic a priori,
synthetic a priori, and synthetic a posteriori (Kant, 2000). Analytic a
priori knowledge is always true, because these exist mere definitions of
ideas (e.g., a triangle has three sides, all unmarried males are
bachelors, etc.). Synthetic a priori knowledge is deduced from a set of
analytic a priori knowledge (e.g., 4 + 7 = 11, summation of all included
angles of a triangle equals 180°, etc.). Thus, knowledge gained from
mathematical and geometric derivations falls under the category of
synthetic a priori knowledge. Synthetic a posteriori knowledge
corresponds to knowledge gained through experience (e.g., apple is good
for health, bachelors are rich, etc.). Besides, Kant considered
existence of four concepts or categories of pure
understanding—quantity, quality, relation, and modality—which
correspond to the inherent ability of humans to organize their
experiences and formulate synthetic a posteriori knowledge. These four
categories, in turn, entail twelve concepts of judgment. Specifically,
quantity entails the concepts of unity, plurality, and totality; quality
entails reality, negation, and limitation; relation entails inherence
and subsistence, cause and effect, and community; and lastly, modality
entails possibility–impossibility, existence–nonexistence, and
necessity and contingency. Although Hume and Kant are considered an
empiricist and rationalist, respectively, their definitions of knowledge
possess certain similarities. For example, Hume’s relations of ideas
correspond to experience-independent knowledge, which Kant classified
into two categories—analytic a priori and synthetic a priori. Both
Hume and Kant categorized experience-dependent knowledge into a separate
category. Hume classified it as matters of fact, whereas Kant considered
it as synthetic a posteriori.
Apart from the Hume- and Kant-based definitions of knowledge, there
exist other definitions of knowledge in epistemology. According to
Russell (1911 (reprinted 2015) and 1914 (reprinted 2005)), there exist
two types of knowledge—by acquaintance and by description. Knowledge
by acquaintance implies knowledge gained by direct awareness or
experience of a knower, and is free from any intermediary inference
processes. Knowledge by description, in contrast, is a propositional
truth acquired via inferential, mediated, or indirect processes. Such
definitions of knowledge explicitly specify the role of the knower in
knowledge formulation. Many authors have investigated knowledge from
perspectives of acquaintance and description, and provided
epistemological descriptions of acquaintance and descriptive knowledge
(Gertler, 1999; BonJour, 2003; Fumerton, 2005). Meanwhile, new
metaphysics have also been added whilst formulating knowledge by
prioritizing the knower. For example, Zagzebski (1999) considered that
knowledge formulates when knowers try to build a relationship with a
portion of reality through their consciousness. Knowers might directly
or indirectly be related to a portion of reality. Therefore, knowledge
depends on knowers’ cognitive abilities and their emotional attachment
with a portion of reality; that is, the role of the knower must be
quantified while defining knowledge. Accordingly, Zagzebski (1999)
defined knowledge as a cognitive contact with reality arising out of
acts of intellectual virtue. This implies that “intellectual virtue”
is a metaphysical quantifier of the knower with regard to knowledge
formulation. However, intellectual virtue can be defined in two
different ways (Battaly, 2018) that depend on the concept of reliability
(Sosa, 1991; Greco, 1993) and responsibility (Zagzebski, 1999; Battaly,
2018). The above definition of knowledge is based on responsibility
(open-mindedness, courage, critical thinking, moral obligation, etc.).
2.2. Engineering Design and Manufacturing
Like its predecessors, in Industry 4.0, it is highly likely that
seamless execution of engineering and manufacturing (i.e., product,
system, and service conceptualization and realization) gets the highest
priority. Thus, how the concept of knowledge has been treated in
engineering design and manufacturing must be elucidated before
proposition a clear and circularity-free definition of knowledge.
First, consider engineering design (product, system, and service
conceptualization). Engineering design is a purely knowledge-intensive
activity (Tomiyama et al., 2013). Therefore, certain design theories
explicitly highlight the contribution of knowledge in the execution of a
design process. For example, let us consider the general design theory
(Yoshikawa, 1989; Takeda et al., 1990; Reich, 1995) and C–K theory of
design (Hatchuel and Weil, 2008; Ullah et al., 2012; Agogue and Kazakci,
2014; Le Masson et al., 2017). In accordance with the general design
theory (Yoshikawa, 1989), execution of a design process requires
knowledge manipulation, wherein “knowledge” may either to be of the
ideal or real types (Takeda et al., 1990; Reich, 1995). This ideal/real
knowledge plays its role through logical processes of deduction and
abduction (Takeda et al., 1990), as described in Section 3. Both
knowledge types assist in making necessary decisions concerning the
continuation of a design process under given circumstances (Takeda et
al., 1990). Nevertheless, ideal or real knowledge types can be defined
with respect to other concepts, such as the “entity” and “topology”
of the design space (Reich, 1995), which must be defined prior to
defining the knowledge types. This injects circularity in the definition
of ideal or real knowledge, in addition to certain logical ambiguities
caused by induction (Section 3). In addition to deduction and abduction
(refer to Section 3 for definitions), another logical process called
induction (Ullah, 2007) must be considered when processing real
knowledge. This is because induction extracts knowledge from experiences
and experimental data. Additionally, the role of induction is not
explicitly highlighted when processing real knowledge within the
framework of the general design theory, thereby imparting ambiguity in
the general-design-theory-based definition of knowledge. In contrast,
the C–K theory of design considers the simultaneous evolution of two
domains—concept and knowledge—when a design process continues
(Hatchuel and Weil, 2008; Ullah et al., 2012). Application of this
theory requires two knowledge types—existing and new—for continuing
a design process (Hatchuel and Weil, 2018; Ullah et al., 2012; Agogue
and Kazakci, 2014; Le Masson et al., 2017). New knowledge is necessary
to resolve epistemic uncertainties underlying creative concepts (Ullah
et al., 2012). Unlike the general design theory, the C–K theory does
not define new-knowledge creation or existing-knowledge utilization
processes in terms of deduction, abduction, induction, and so on.
Consequently, C–K-theory-based definitions of knowledge are somewhat
informal.
Similar to engineering design, in manufacturing (product, system,
service realization), the concept of knowledge has always existed. It
appears more explicitly owing to the advent of Industry 4.0. As
mentioned before, Industry 4.0 employs embedded systems (e.g.,
cyber-physical systems) to execute such cognitive tasks as monitoring,
understanding, predicting, deciding, acting, and adapting. Some authors
reckon that the cyber-physical systems are nothing but an extensive and
self-growing knowledge base (Zhou et al., 2018), but knowledge is not
defined in clear terms. On the other hand, some authors consider that
the contents by which the embedded systems perform, take the form of
digital twins—exact mirror images of real-world objects, processes,
and phenomena—in cyberspace (Ghosh et al., 2019; Ullah, 2019). Some of
these twins consists of different types of knowledge (Ullah, 2019), but
these types are not clearly defined. Some other authors (e.g., consider
the work in Zheng et al., 2018) reckon that both data-bases and
knowledge-bases must populate the embedded systems where the demarcation
lines between “data” and “knowledge” are not clearly drawn. As a
result, in the literature of Industry 4.0, the concept of knowledge
remains ambiguous.
2.3. Other Relevant Fields
In addition to epistemology, engineering design, and manufacturing,
definitions of knowledge have also been reported in the literature of
other fields, such as organization, education, and information sciences.
A well-known definition of knowledge in organization science states that
knowledge can either be of the tacit or explicit types (Polanyi, 1958;
Nonaka, 1994; Nonaka and Takeuchi, 1995; Randles et al., 2012; Nonaka et
al., 2014). Tacit knowledge pertains to intuitions, experiences, and
know-how possessed by active individuals in their respective
organizations. Consequently, it is challenging to identify or even
codify such knowledge (Polanyi, 1958; Nonaka, 1994; Randles et al.,
2012). Explicit knowledge includes documented instructions for
facilitating organizational activities. It is, therefore, easy to
identify and share. Tacit knowledge dynamically transforms into explicit
knowledge and vice versa through social or teamwork-based interactions
(dialogue) among employees (Nonaka et al., 2014). It is remarkable that
such transformations do not require formal logical processes to be
performed (Nonaka, 1994; Nonaka and Takeuchi, 1995). This contradicts
the definitions of knowledge reported in other disciplines, such as
information science. However, there exists other schools of thought in
organization science related to knowledge (Albino et al., 2001) and its
formation (Bohn, 1994). For example, Albino et al. (2001) considered
that there exist five types of knowledge—scientific, quantitative,
qualitative, tacit, and intuitive. As reported by Bohn (1994), the
knowledge formation and validation processes follow a hierarchy, which
is not clearly defined.
In education science, the concept of knowledge has always existed along
with human-learning. For example, consider the definitions of knowledge
presented in Carson (2004), Kinchin et al. (2019), and Ullah (2019).
Carson (2004) has proposed nine categories of knowledge—empirical,
rational, conventional, conceptual, cognitive-process skills,
psychomotor, affective, narrative, and received. All these categories of
knowledge form in the intertwined domains, and ultimately transform to
conventional knowledge. Kinchin et al. (2019) have proposed four types
of knowledge, namely, novice knowledge, theoretical knowledge, practical
knowledge, and professional knowledge. All these types of knowledge
possess different degrees of “semantic gravity.” Ullah (2019) has
proposed five types of knowledge—analytic a priori, synthetic a
priori, synthetic a posteriori, meaningful, and skeptic—for
discipline-based education. The first three types follow the Kantian
epistemology described in section 2.1 and form in the cognitive and real
worlds, whereas the last two types of knowledge form in the pragmatic
world where the preferences of the knowledge formulator and the purposes
of applying knowledge become the main ingredients of knowledge.
Nonetheless, the definitions of knowledge in Carson (2004), Kinchin et
al. (2019), and Ullah (2019) are somewhat informal—defined
linguistically, only.
In information science, the concept of knowledge has always existed
along with the concepts of information and data, and these three
concepts have been used interchangeably. These concepts have started to
play an explicit role in engineering problem solving when several
machine-learning approaches have been introduced (Quinlan, 1979;
Hayes-Rotb et al., 1983; Quinlan, 1986; Nagao, 1990; Heckerman et al.,
1995; Studer et al., 1998; Rathman et al., 2017) to facilitate learning
from a given dataset. This enables expert systems to solve
domain-specific problems (Hayes-Rotb et al., 1983). At the core of these
systems lie certain rules (e.g., if…then… rules) extracted
from a given dataset using probabilistic reasoning and fuzzy logic
(Zadeh, 2002; Ullah and Harib, 2008). Therefore, in information science,
machine-learning-enabled rules have been playing the role of knowledge.
A few authors in information science have formally defined knowledge
with regard to data and information. But these definitions suffer
circularity. For example, consider the definitions reported in Nagao
(1990) and Mizzaro (2001). Nagao (1990) has classified knowledge into
two types—factual and inference-based. Factual knowledge is obtained
objectivity, accepted widely, and can be expressed as a sentence or
symbolic equation, wherein each term is clearly defined. To represent
factual knowledge, other concepts (referred to as primary and secondary
information) (Nagao, 1990) can be used as semantic annotations for ease
of digital-media-based information processing. On the other hand, using
inferences (deductive, inductive, or probabilistic) (Ullah et al.,
2012), cognitive reasoning (analogical, common sense, and qualitative),
and heuristics, new knowledge can be acquired from factual knowledge.
Such knowledge is called inference knowledge (Nagao, 1990, p.9-16).
Mizzaro (2001) has introduced a concept called knowledge-state to draw
demarcation lines among knowledge, information, and data. Although the
temporal nature of a knowledge-state has been studied (Mizzaro, 2001),
the types of knowledge have not been shown explicitly in terms of
knowledge state or data/information.
3. Revised Definition of Knowledge
The previous section describes the multiplicity and circularity
underlying the definitions of knowledge as elaborately as possible
referring to different fields (epistemology, engineering design,
manufacturing, organization science, education science, and information
science). This section presents a revised definition of knowledge from
the perspective of Industry 4.0. Before presenting the revised
definition, the following points are highlighted for the sake of better
understanding.
Despite the multiplicity and circularity in the definitions of knowledge
as described in the previous section, there are some common grounds. One
of the noteworthy common grounds is related to the representation of
knowledge—knowledge graphs (e.g., concept maps) can represent
knowledge irrespective of its types, human-learning, machine-learning,
and academic fields. To understand this, recall information and
education sciences, as described above. In information science, where
machine-learning is the main concern, knowledge representation boils
down to concept mapping. For example, the types of knowledge called
factual and inference knowledge (Nagao, 1990) are represented using
semantic networks of concepts wherein the logical operations (AND, OR,
NOT, and alike) connect the relevant concepts manifesting a set of
if…then… rules (Nagao, 1990). In education science, where
human-learning is the main concern, knowledge representation also boils
down to concept mapping. For example, Kinchin et al. (2019) and Ullah
(2019) represented various types of knowledge using concept maps. A
concept map here is a personalized ontology of human understanding
regarding a given issue (Ausubel et al., 1978; Ausubel, 2000; Novak,
2002). These types of maps ultimately refer to the assimilation
hypothesis of human learning (Seel, 2012). The remarkable thing is that
the technology of semantic web wherein the machine and human readable
knowledge and linked data will reside for using them in Industry 4.0
will rely on the knowledge graph-based data-format (i.e., concept
mapping) (Kim, 2017; Fionda et al., 2019).
On the other hand, there are some disputed grounds. For example,
consider the role of logical operations in knowledge formation. There is
a split in this regard. To understand this, recall the recall
information and organization sciences, as described above. Information
science demands application of logical operations for knowledge
formulation and transformation, e.g., factual–inference knowledge
transformations require sophisticated logical operations (Nagao, 1990).
Organization science demands application of social
interactions—tacit–explicit knowledge transformations require social
interactions only (Nonaka, 1994)).
The demarcation lines among knowledge, data, and information have been
an important issue. Though knowledge-state (Mizzaro, 2011) may be used
to solve this problem but without the types of knowledge it would be
difficult to implement it in real-life scenarios. Though the semantic
networks of linked data and knowledge (i.e., knowledge graphs or concept
maps) integrate the relevant data, information, and knowledge to solve
problems in engineered systems (Mordecai and Dori et al., 2017;
Mordecai, 2019; Chen et al., 2019), the maps are created without making
any distinctions among knowledge, data, and information. As a result,
when the so-called knowledge state (Mizzaro, 2011) changes, its
influence randomly propagates to whole network. This means that what has
been affected (knowledge, data, or information) to what extend remains
obscure. On the other hand, human-learning described so far is based on
the assimilation hypothesis (Seel, 2012), which states that without
existing knowledge, nothing new can be learned. This contradicts the
concept of new knowledge given by the C–K theory of design (Ullah,
2012), and is not desirable because new knowledge is the primary
ingredient for creating artifacts. Thus, education-science-driven
definitions of knowledge tend to ignore the main ingredient of
creativity. On the other hand, which segment of a knowledge graph or
concept map is knowledge and which part is not must be known beforehand.
This is not possible until a clear demarcation line exist between
knowledge and other relevant contents (e.g., data and information). At
the same time, the relevant system must aware of the co-existence of
different types of knowledge. This is drawback of the exiting knowledge
and data prospection using semantic web (Kim, 2017; Fionda et al.,
2019).
Based on the abovementioned considerations, this section presents an
Industry 4.0- and semantic web-friendly definition of knowledge, which
is free from circularity and ambiguity.
The proposed definition of knowledge is as follows. A piece of knowledge
(denoted as K ) comprises three elements—knowledge claim
(Kclm ), knowledge provenance
(Kprv ), and knowledge inference
(Kinf ). In general, these elements demonstrate
the following relationship.
\(K=\left\{K_{\text{clm}},\ K_{\text{prv}},K_{\inf}\right\}\text{\ \ \ \ \ }K_{\text{prv}}K_{\text{clm}}\)(1)
In the above expression, Kclm denotes a
manifestation of K ; that is, Kclmrepresents a piece of knowledge, such as a proposition, an equation, or
any other piece of information. Other elements may or may not be
reported explicitly; that is, Kprv andKinf may remain empty, butKclm ≠ ∅. Kprv helps
identify the truthiness of Kclm . There exist no
restrictions that Kclm must be “completely
true” or “completely false.” Partially true or partially falseKclm can be used to manifest K . This
implies that Kprv may not fully justifyKclm . Kinf refers to the
inferential process involved in gaining Kclm in
the presence of Kprv . In addition,Kinf helps categorize K into different
types and categories. In some cases, Kprv andKinf may remain empty; i.e.,Kprv , Kinf = ∅ is allowed.
The definition of knowledge given by equation (1) yields four
fundamental knowledge types—(1) definitional; (2) deductive; (3)
inductive; and (4) creative—which are summarized in Table 1 along with
their main characteristics and descriptions.
Table 1. Definition of Knowledge