Digital Twins (DTs) are often described as virtual replicas or copies of physical systems. However, we argue that this view not only obscures key aspects of what DTs actually do but also fails to capture their underlying ontological rationale. In this paper, we ofer a foundational account of DTs based on dolce, arguing that these technologies embed intentional representations at their core. We then qualify DTs as information artefacts shaped by stakeholders' goals and manifesting qualities, which we call 'diferential qualities', that have causal relevance. These are essential for accomplishing tasks typically performed by DTs in engineering, such as diagnostics, optimisation, and simulation.
Digital Twins (DTs) are often informally presented as kinds of virtual or digital models of physical
objects or systems, created to support activities like simulation, real-time monitoring, predictive
analysis, visualisation, and operational control. While DTs have recently garnered attention from
scholars working or interested in ontology at large (see, for instance, [
In this paper, we approach the conceptualisation of Digital Twins (DTs) from a foundational ontology
perspective, investigating the primitives required for their formalisation. The foundational ontology
adopted in this work is dolce (Descriptive Ontology for Linguistic and Cognitive Engineering) [
At first glance, as mentioned, one could be inclined to model DTs merely as copies or replicas of
physical entities. However, as we will argue, although it is weaker than the logical identity [
Rather than viewing DTs as kinds of copies, we propose to understand them as specific representations of physical endurants, i.e., intentional representations, which allows us to identify them as a subclass of technical artefacts, namely information artefacts, whose features depend (in part) on the goals of various agents involved as stakeholders (e.g., designers, engineers, users, etc.).
This goal-oriented approach is further complemented by a causal view of DTs. Regardless of the physical entities related to DTs, we claim that there exists a subset of features that consistently matter to identify them. We refer to these as ‘diferential qualities’. These properties, by having causal import, make a diference in how an entity changes over time, and therefore are relevant for understanding, simulation, optimisation, and diagnostic tasks that DTs typically perform.
This perspective is particularly important in engineering contexts, where DTs can be used to evaluate
and ensure the dependability of systems under a variety of conditions [
A well-known distinction present in the literature is that between Digital Twin Prototype and Instance. While the former is a description of a prototypical physical artefact and is aimed at then producing the physical twin, the latter describes a specific, already existing, physical twin, to which it remains linked throughout its life1.
However, although one can speak of DTs at the type level, thus addressing their conceptualisation to characterise the prototypical class of Digital Twins, in the current paper we prefer to limit our analysis to the token level. Beginning with individual tokens allows us to focus on practical examples and enables a step-by-step process of abstraction.
The paper is structured as follows: in Section 2, without aiming to disentangle the ontological
complexity of the notion of copy [
The label ‘Digital Twin’, which originated in the field of engineering, does not refer to a specific
technology. Rather, it encompasses a wide range of computational disciplines, applications, and
techniques, including computational logic, AI algorithms, ontology-based data modelling, the Internet
of Things (IoT), and sensors [
Ontologically, in a nutshell, one might say that Digital Twins (DTs) are technical artefacts2 representing
a physical entity. In this respect, they are not far from abstract models or dynamic simulations of
the world, which have been widely employed in science and technology for decades [
Not surprisingly, given the technological complexity of the topic, we lack a universally accepted
definition. Yet, the domain of discourse about DTs can be roughly understood as involving three main
entities [
Now, at a high level of abstraction, one might intuitively argue that, since each physical entity has a corresponding digital counterpart, that is, its ‘Digital Twin’, the very term itself suggests a kind of ‘copy’ relation is involved. However, this option is neither feasible nor desirable.
To start with, one might naively wander what features and how many, should be identified in order to
claim that a DT is a copy of a physical entity. Trivially, storing every single detail of the physical entity
is often impractical, if not impossible (epistemologically speaking). This immediately underlines that we
1It seems well-established in the literature that, regardless of whether it is created before or after its digital counterpart, the
existence of a physical twin is a necessary condition for the existence of the digital twin.
2This notion will be explored in more detail in Section 3.
are not looking for a strict notion of copy or a replica (e.g. a PDF copy) [
Furthermore, real-world social and biological systems are inherently emergent, requiring open and
non-linear models that move beyond simple copying processes [
Unlike a physical replica (e.g. biologically identical twins), a DT is goal-designed, i.e. it is created with specific purposes in mind, typically: diagnostic, prediction, optimisation. This agent-dependency entails that the Digital Twin selectively represents only those features of the physical entity that are relevant for achieving the intended goals attributed to the related DT. Its identity is therefore determined by a goal-oriented process. This marks a crucial ontological distinction: DTs are not merely what they are, but what they are for, a dimension absent in standard notions of replication and, to some extent, also of copying. Regarding the latter, the intentional dimension is involved insofar as the act of copying entails an artefactual process of creation or modification, and not to the idea of copy itself. We claim therefore that while the copy relation may seem intuitive, it is too rigid to fully account for the complexity of Digital Twins.
Along these lines, Korenhof et al [2021] warn us that framing the conceptualisation of DTs solely in terms of ‘realistic’ representations of a physical entities, and indeed as a kind of copy, obscures the ‘prescriptive’ dimension of DTs, which are a-priori designed to monitor or improve the physical counterpart. This is not a neutral process, of course. As a result, a DT always contains something that is of a diferent ontological kind w.r.t. the physical entity it refers to. For example, the numerical models that simulate the growth of a grapefruit tree are not part of the actual grapefruit tree. These models are instead used to cope with a physical reality that is perceived, if not as faulty, at least as something in need of optimisation. Ultimately, this leads to a reification of the DT: it is treated as a full-fledged entity that, for all practical purposes, substitutes the physical object.
For instance, a farmer who uses a DT to monitor and automate the feeding of cows, thus optimising
the process, engages with the DT as an interface. In the worst-case scenario, this might alienate the
farmer from the actual physical activity, delegating much of the labour to an automated system driven
by the DT [
While representations are a crucial component in discussions about DTs, they coexist with goals, prescriptions, and actions enabled or required by such technologies. This ultimately leads Korenhof et al [2021] to qualify DTs as ‘steering techniques’, aligning them with the traditional cybernetic paradigm in which technical systems are used to steer or guide behaviour. In this view, DTs are employed to realise specific goals in the world through feedback mechanisms that provide information and influence action.
The informational dimension introduces an additional nuance to the concept of ‘intentional’, that
must be distinguished from goal-oriented aspects or, in other words, from agents’ intentions4. An
intention or goal5, conceived as a mental state, is merely one among several such states, e.g. beliefs,
desires, and perceptions, that exhibit the property of ‘intentionality’ or ‘aboutness’, which has to do
with the very matter of meaning, whether mentally or linguistically conceived [
From what has been said, we can refine our ontological understanding of DTs by drawing on two main assumptions:
i) First, we can view the representation (i.e. aboutness) relation as the fundamental link between a physical entity and its DT. DTs refer to their physical counterparts in a meaningful way and carry information about them.
ii) Second, alongside this informational aspect, we must adopt a goal-oriented perspective. This accounts for how and why DTs, as a specific type of technical artefact, are used as ‘steeering’ technologies.
In the following sections, we will explore these assumptions in greater depth in order to provide a basic formal understanding of Digital Twins.
There is no doubt that DTs represent a complex technology, made up of multiple layers of material
and code. Some of these layers are directly perceivable by humans, such as the output displayed on a
screen, while most remain concealed beneath intricate arrangements of physical components, hardware
engineering, and logical systems (from high-level programming languages all the way down to the
electrical properties of copper and silicon) [
However complex such technology may be, ontologically we can distinguish two main layers: the
material and the artefactual, with the former constituting the latter. To account for this, we draw
on the formal theory of artefacts developed in [
The topic of capacities attributed to artefacts is well suited for our purpose, as it helps illuminate
the steering role of DTs in achieving human goals. Moreover, by adopting a multiplicativist approach
[
To account for this, we need to appeal to the so called ‘information entities’. This is a general label encompassing a variety of entities, ranging from semi-abstracta to meanings, ideas, and documentary entities. In order to highlight similarities and diferences among these entities, Sanfilippo [2021] presents a comparative table with column indices reporting on some ontological dimensions, such as existence in time and space, criteria of unity, generic dependence, and aboutness. For our purposes, it is worth noting that all the theories examined in [24] appeal to the notion of aboutness to characterise information entities, though without committing to a particular stance or providing a formal account 7. This lack of formal studies is not limited to the domain of information entities. In fact, the concept of intentionality or aboutness is primarily treated as a primitive and largely underanalysed in the applied field of Formal Ontology [26], despite being a fundamental notion across multiple domains.
Expressions such as ‘reference’, ‘intentionality’, ‘aboutness’ have considerable semantic overlap and
are sometimes used interchangeably for all means and purposes. However, depending on whether the
emphasis is on language or mind, we see a prevalence of terms like ‘reference’ and ‘aboutness’ in the
former case, and ‘intentionality’ in the latter. For instance, according to Yablo [2014], aboutness “is
the relation that meaningful items bear to whatever it is that they are on or of or that they address or
concern.”. Conversely, intentionality is, as mentioned, typically introduced as the property of mind, and
more specifically a kind of power or ability, to be about, represent or stand for something [
In this respect, the notion of aboutness can be employed to capture the link between DTs and their
physical counterparts. We will discuss this in Section 5, by leveraging the formalisation and the theory
of aboutness ofered in [
Before concluding this section, let us raise another important issue. Just as there is no single
technology that fully identifies a DT, there is likewise no fixed set of tasks it is meant to perform.
However, among the most common applications are diagnosis, simulation, and optimisation [
Digital Twins (DTs) are deployed in many heterogeneous domains, but in this paper we shall mainly
focus on engineering contexts, where causal and probabilistic aspects of artefacts’ behaviour are central.
In this section, we introduce the relation we dubbed ‘isDifferentialQualityFor’, specifically
devised to capture such aspects, and developed in a paper currently under review [
To this aim, let us briefly contextualise the main theory of causality we appeal to, and the reasons behind such an endorsement in light of our objectives. First of all, we aim to articulate the intuitive idea of ‘causal contribution’ between properties, in alignment with the technological domain under investigation. In this respect, a probabilistic approach to causation appears best suited to account for the behaviour of DTs, e.g. simulating physical artefacts in complex environments often involves dealing with multiple sources of uncertainty (including stochastic processes such as sensor noise, unpredictable external influences, and incomplete or imprecise data about the system’s components, the environment, etc).
Probabilistic accounts of causality belong to a broader family of approaches commonly referred to as ‘diference-making’ theories of causality. These are often contrasted with another group of approaches, known as ‘mechanistic theories’. The distinction lies in their difering uses of causal claims: the diference-making approach serves an inferential purpose and applies to types of entities, while the mechanistic stance is aimed at providing explanations and is used for token entities [27]. The mechanistic stance focuses on identifying the physical processes that explain the link (i.e., the mechanism) between causes and efects in individual cases and through inductive reasoning. For example, by connecting the insurgence of John’s lung cancer with his being a heavy smoker [28]. In contrast, the diference-making perspective emphasises the information required to infer efects from causes, e.g. by intervening on variables representing causes while controlling for variables representing efects, such as when constructing a Causal Bayesian Network from a dataset [ 29]. In brief, diferencemaking involves the idea of ‘change’, i.e. how is that the presence of causes brings about changes in their efects [cf. 28].
Hence, within the engineering context, both probabilistic and mechanistic factors are often required, as engineers are interested in both the predictive and explanatory uses of causal claims. From this standpoint, the epistemic theory of causality [29] is well-suited, as it unifies probabilistic and mechanistic evidence into a coherent framework of rational causal belief. Though being an epistemic theory (i.e. not committed to causality as a full-fledged entity), this view does not deny the existence of facts serving as truthmakers for causal beliefs [28, 30]. Rather, it openly acknowledges the evidential grounding of causal claims in the agent’s perspective.
In this framework, the relation isDifferentialQualityFor helps formalise how certain qualities
of artefacts contribute (both probabilistically and mechanistically) causally to changes observed in other
qualities. More specifically, referring to dolce’s categories 9, it links two qualities and ′ (inhering in an
8Readers more interested in the philosophical aspects might recognize in the notion of diferential qualities certain echoes
of the metaphysics of dispositions. This is a delicate point that deserves clarification. First, when we speak of diferential
qualities, our interest lies primarily in the type of relationship that exists between two (or more) qualities, and not in the
nature of those qualities themselves. The focus is on how such qualities behave within the dynamics of interaction between
the endurants to which they belong. In [
Ontologically, this setup asserts that when two events are taken as evidence for a causal claim, the time-dependent value change in the qualities of the participating endurants is also causally related. That is, a change in quality at time implies, within some probabilistic constraint and against some relevant mechanistic explanation, a change in ′ at time ′. For example, increased humidity (a quality of the environment) participating in a vaporising event can reduce the accuracy (′) of a sensor participating in a subsequent error reading event [32].
With the introduction of diferential qualities, we are now ready to ontologically qualify Digital Twins.
In this section, we introduce a real-world example in the engineering domain to help understand the formalisation we will provide. We already mentioned that DTs are rooted in real-time data and grounded in physics-based simulations, making them powerful tools for early fault detection and decision-making.
The example is taken from Siemens’ use of DTs in monitoring large rotors in generators and compressors11. When one of such rotors fails, downtime can cost up to five million euros per day. Even scheduled maintenance results in weeks of inactivity (mainly due to safety reasons) and requires expert analysis.
In contrast to this traditional maintenance scenario, the Digital Twin-based approach ofers considerably greater flexibility. Every rotor has a Digital Twin created at the design stage that runs in parallel with the actual machine, constantly simulating ideal sensor readings (e.g., vibration data). When discrepancies arise between simulated and real data, the DT provides diagnostic. By simulating various fault scenarios (like imbalance forces or dust buildup), adjusting parameters (force location, magnitude), and minimising the diference between real and simulated data, the DT identifies the cause and location of the fault and allows enough time to schedule repairs.
Let us now introduce the dolce framework that will enable us to ontologically account for the
practical applications of Digital Twins, including the industrial scenario just described. To start with, we
recall that, of course, Digital Twins are artefacts, as they are the product of human intentional activities.
So far, in dolce only material artefact has been formalised in [
In our intuition, Digital Twins are still more specific kinds of artefacts, namely, able to carry information. We shall call this kind ‘information artefact’ and look for an ontological characterisation.
To this aim, we build on previous work, presented in [
html agents. Concepts can classify entities at a certain time12, but they can also exist without classifying anything. For those concepts that succeed in classifying some entities at a given time, we use in our framework also the relation of 2, which is a ternary relation involving an agent’s mental state , an entity (instance) , and a time (moment or period) at which such an entity is classified by that very concept.
So, the overall picture is that when agents are in a certain mental state (like belief or intention), such mental state is always directed towards a concept. The fact that a concept is always involved captures the common-sense idea that thinking is always thinking about something. In our framework, the relation of being directed towards a concept is called 1 and, while this relation holds, the concept may be called the ‘intentional content’ of the mental state. Now, in some cases the concept does not classify any entity or, in other words, the content does not denote any entity, such as for fictional entities. In such cases, 1 is the only relation that holds; in all other cases, there will also be a relation of 2 between an agent’s mental state and an individual entity in the‘world’ or ‘domain’ of interest, that we shall call ‘intentional object’, which is then classified by the concept towards which the mental state is directed.
It is worth noticing the two senses of aboutness formalised in [
Now we can relate the theory of aboutness to the characterisation of information artefacts. The basic idea is that an information artefact is a technical artefact that encodes information. However, for such information to be meaningful, it must be ‘cognitively graspable’ by an intentional agent, therefore here is where the notion of 1 comes into play. This is tantamount to say that the ‘aboutness’ is not so much intrinsic to the concepts encoded by the informational artefact per se, but rather arises indirectly through the agent’s cognitive intervention [34]. The encoding is possible insofar as the physical object (or part of it) that constitutes the artefact is arranged in a way that makes it cognitively accessible; in other words, insofar as it functions as a symbol (e.g., ink patterns in books, voltage states in computer gates). We stress that, generally speaking, many information artefacts only need to appeal to 1, that is, a relation to content; e.g. in the case of books about fictional characters. Likewise, there are cases in which such content, through the mediation of an intentional agent who entertains some mental attitudes towards it, can refer to something in the world. Those are the cases in which 2 is at stake 13. Examples of such information artefacts are medical records, dashboards in industrial settings and Digital Twins as well.
To provide the definition of information artefact, we introduce the primitive binary relation Encodes,
which holds between the technical artefacts and an ‘informative content’(concept) at a certain time14.
We have therefore the following:
Axiom 1: Encodes(, , ) → TechArt() ∧ C() ∧ ()
Definition 1: InfoArt() ↔ Encodes(, , ) ∧ ∃(MS() ∧ About1(, , ))
12In the seminal work on concepts [33], these were introduced to classify endurants. However, in [
Axiom 1 minimally constrains the arguments of the relation Encodes,while Definition 1 states that is an information artefact if is a technical artefact that encodes a concept at a certain time and there exists an agent’s mental state that is About1 concept at 15.
The following step is that of characterising Digital Twins (DTs) as a subclass of information artefact. We mentioned (Section 2) that DTs are used in goal-oriented contexts, underlining their being ‘steering technologies’. In addition, we introduced a class of qualities (Section 4), called ‘diferential qualities’ helpful to identify features of the physical entity useful for DTs’tasks (e.g. simulation, optimisation, etc). Now let us resume the example about the Siemens’ rotor to contextualise the ontological scenario.
Consider a system in which the DT of a rotor includes a parameter representing the rotor’s orientation angle. This parameter is modeled as a quality inhering in the DT. The DT participates in a simulation event , during which it predicts friction due to dust accumulation. As a result, it updates the value of the quality ′ (i.e., the angle) that inheres in the actual rotor. This update causally triggers an adjustment in the physical rotor’s orientation, mitigating dust build-up. Thus, the quality of the DT can be said to be a diferential quality for ′ related to the Physical Twin. In addition, the simulation event serves as a satisfier 16 of an instrumental goal . The occurrence of brings about another event ′, i.e., a decision-making event that, in turn, satisfies an end goal ′. This complex event ultimately results in scheduling a cleaning intervention without requiring system shutdown, thereby achieving the end goal of minimising operational costs.
Now, in order to formalise the causal dimension, thanks to which the DT modifies a quality of the
actual rotor, we leverage the primitive relation of ‘causal contribution’ introduced in [
Q(′) ∧
Axiom 2 states that if causally contributes to ′, then both and ′ are events, and there exist qualities and ′ such that is a diferential quality of ′ (a change in the value of brings about a change in the value of ′). These qualities inhere in the endurants and ′, respectively, with participating in and ′ participating in ′. Now we introduce the goal-oriented dimension which, together with the relation of causal contribution, allows us to qualify what a Digital Twin is: Axiom 3: dgTwin() → InfoArt()∧PED()∧SD(, )∧goal(, , )∧∃′′(︀ goal(, ′, ′)∧ ≺ ′ ∧ (∃. sat(, , ) ∧ PC(, , ) → ∃′. sat(′, ′, ′) ∧ PC(, ′, ′) ∧ causalContr(, ′)))︀ Axiom 3 characterises a Digital Twin as an information artefact that is specifically dependent on a physical endurant (i.e., any material entity, including other artefacts) for its existence. Moreover, the DT participates in an event which satisfies an agent’s (instrumental) goal concerning a concept at time . Event causally contributes to event ′, which satisfies the agent’s (end) goal related to a concept ′ at time ′, where precedes ′. In addition to qualifying Digital Twins as steering technologies that exert a causal impact on their physical counterparts, Axiom 3 also emphasises their existential dependence on the corresponding Physical Twins. Finally, we also provide a characterisation of the notion of a Physical Twin: Axiom 4: phyTwin() → PED() ∧ ∃, , , (︀ dgTwin() ∧ Encodes(, , ) ∧ About1(, , ) ∧
About2(, , ))︀
15Notice that must be the same in the two relations in Definition 1, because in the periods in which one of the two conditions
fails, the technical artefact still exists, but it does not convey information, thus the information artefact ceases to exist.
16In [
Axiom 4 says that if a physical endurant is a Physical Twin, it is the reference of an information content encoded by the Digital Twin. This should render the fact that an entity is a Physical Twin only when there is at least an agent holding a mental state directed to an information content encoded by the Digital Twin that classifies (is made concrete by) such entity. In other terms, together with the existential dependence of the Physical Twin from the Digital Twin, it is also the case, roughly speaking, that an entity becomes (plays the role of) a Physical Twin, only when it has a Digital Twin.
In this paper we argue against the use of the concept of copy to account for the ontology of Digital Twins. While intuitively appealing, the notion of copy proves too rigid and obscures the prescriptive nature of these technologies. Instead, we propose a goal-oriented approach which, supported by a class of qualities we dubbed diferential qualities, highlights how DTs function as paradigms for modifying, controlling, and ultimately guiding the behaviour of their physical counterparts. Moreover, although a relation of existential dependence holds between the DT and the physical entity (see Axiom 3), we contend that aboutness is the main relation linking the two. It is through its representational character that the DT carries information about its corresponding Physical Twin. This, in turn, allows us to characterise DTs as a particular kind of technical artefact, namely, information artefacts.
Unlike some common information artefacts, such as a science cfition book, in which the encoded concepts need not refer to something actual, DTs belong to a specific class of information artefacts that are always connected with a material entity, in the specific sense of 2 introduced in Section 5. This is not surprising, given that DTs are designed to causally afect the physical world: there must always be a physical counterpart upon which they act, either directly (e.g., when the DT autonomously performs an action) or indirectly (e.g., through human intervention). Our analysis ultimately reveals that, among all types of information artefacts, DTs are arguably among the very few that also exert a causal impact on the world they represent. Finally, a further interesting feature of DTs – and of information artefacts in general – that we have highlighted in our framework is that not only the intentions of the designer are relevant for their characterisation, but also of the agents that can grasp the information that the artefact encodes.
In this paper, we have focused specifically on the concept of the DTs at the token level. However, a comprehensive ontological account cannot disregard the type level, that is, the prototypical dimension that underpins the design, modelling, and implementation of Digital Twins as a class of artefacts. In this respect, the present work should be considered as a primer that serves as a base for a future type-oriented level analysis.
This research has been funded by the Project PRIN 2022 SMARTEST (Project nr. 202223E8Y4X), awarded by the Italian Ministry of University and Research (MUR). Roberta Ferrario has also been supported by the Project PRIN2020 BRIO – Bias, Risk and Opacity in AI (2020SSKZ7R), funded by MUR and by the project FAIR (Project nr. PE 00000013), funded by MUR.
During the preparation of this work, the authors used ChatGPT in order to: grammar and spelling check, paraphrase and reword. After using this service, the authors reviewed and edited the content as needed and take full responsibility for the publication’s content. Philosophy, spring 2023 ed., Metaphysics Research Lab, Stanford University, 2023. https://plato. stanford.edu/archives/spr2023/entries/intentionality/. [22] Y. Hui, What is a digital object?, in: H. Halpin, A. Monnin (Eds.), Philosophical Engineering:
Toward a Philosophy of the Web, John Wiley & Sons, 2013, pp. 52–67. [23] C. Masolo, E. M. Sanfilippo, Technical artefact theories: A comparative study and a new empirical approach, Review of Philosophy and Psychology 11 (2020) 831–858. URL: https://doi.org/10.1007/ s13164-020-00475-9. doi:10.1007/s13164-020-00475-9. [24] E. M. Sanfilippo, Ontologies for information entities: State of the art and open challenges, Applied
Ontology 16 (2021) 111–135. [25] W. Ceusters, B. Smith, Aboutness: Towards foundations for the information artifact ontology, in:
Proceedings of ICBO 2015, volume 1515, 2015, pp. 1–5. [26] F. Toyoshima, A. Barton, O. Grenier, Foundations for an ontology of belief, desire and intention, in: Formal Ontology in Information Systems, 2020, pp. 140–154. [27] J. Williamson, Causal pluralism versus epistemic causality, Philosophica 77 (2006). [28] F. Russo, J. Williamson, Interpreting causality in the health sciences, International Studies in the
Philosophy of Science 21 (2007) 157–170. [29] J. Williamson, Bayesian Nets and Causality: Philosophical and Computational Foundations, Oxford
University Press, Oxford, 2004. [30] M. Wilde, J. Williamson, Evidence and epistemic causality, in: Statistics and Causality: Methods for Applied Empirical Research, Wiley, 2016, pp. 31–41. [31] P. Gärdenfors, Conceptual Spaces: The Geometry of Thought, MIT Press, Cambridge, MA, 2004. [32] Y. Zou, J. D. Clark, A. A. May, A systematic investigation on the efects of temperature and relative humidity on the performance of eight low-cost particle sensors and devices, Journal of Aerosol Science 152 (2021) 105715. [33] C. Masolo, L. Vieu, E. Bottazzi, C. Catenacci, R. Ferrario, A. Gangemi, N. Guarino, Social roles and their descriptions, in: D. Dubois, C. Welty, M.-A. Williams (Eds.), Proceedings of the 9th International Conference on the Principles of Knowledge Representation and Reasoning (KR), AAAI Press, Menlo Park, CA, 2004, pp. 267–277. [34] J. R. Searle, Intrinsic intentionality, Behavioral and Brain Sciences 3 (1980) 450–457.