Can AI Think Like Us? Kriegel’s Hybrid Model

Abstract

This review provides a systematic critique of the debate between two paradigms in the philosophy of mind—the Naturalist–Externalist Research Program (NERP) and the Phenomenal Intentionality Research Program (PIRP)—with particular focus on Uriah Kriegel’s reconciliation project. Following Kriegel’s view, attention is given to rational agents’ awareness of their mental states—a key issue since most current artificial intelligence systems aim to model rational thinking and action. Naturalist accounts derive mental content from brain activity and environmental interaction, emphasizing a constitutive dependence of the former on the latter. In contrast, phenomenological theories assert that the object of mental states is an internal semblance presented to the subject. Within this framework, I maintain that Kriegel attempts to naturalize mental content within the framework of a Same Order theory, but this limits his ability to demonstrate that intentionality is grounded in consciousness in the sense of the Phenomenal Intentionality Research Program. Compounding this issue, the idea that the mind arises from manipulating representations has been challenged by dynamical approaches to cognitive science, yet advanced representational models persist, often simulating phenomenological qualities through forms of internal data organization. Methodologically, the approach is primarily comparative and reconstructive, focusing on the structural tensions and theoretical commitments that distinguish NERP and PIRP.

1. Introduction

This study examines the fundamental theoretical commitments of both the Naturalist–Externalist Research Program1 and the Phenomenal Intentionality Research Program2 (hereafter NERP and PIRP), with particular attention to their connection with Kriegel’s self-representational theory, of which a brief illustrative sketch is preliminarily offered in this opening section [2,3,4,5,6]. This study critically evaluates the extent to which the latter can be coherently integrated into the PIRP framework3.

In the first instance, I would like to provide a general outline of my work: the first part presents key terms from the field of philosophy of mind. This is in order to critically assess Kriegel’s hybrid model of experience, which addresses the relationship between intelligence and consciousness—that is, between intentionality directed toward the world and internal awareness. I focus on the naturalist program (NERP) with a point concerning the debate with connectionism and the phenomenological program (PIRP). This is followed by a section in which I raise some objections to Kriegel’s proposal on intentionality and his self-representational theory of consciousness. Section 6 explores in what sense artificial machines can be considered thinking entities. The analysis of Kriegel’s work significantly impacts the overall research, shedding light on key concepts. Researchers are currently investigating whether it is possible to train artificial systems to achieve some level of understanding of their surrounding environment through sensory information. Similarly, within philosophy, some programs aim to ground thought in experience and in the felt life of the human subject. In some cases, philosophers explicitly focus on the naturalization of mind and consciousness, advocating functionalist and representationalist accounts of the mind. This entails considering what role experience plays in thought and raises the question of whether thought requires perception. In particular, within the field of AI, it is worth asking whether perception elevates thought to a higher level. For example, when one thinks of the concept of justice, one can represent it in abstract terms; however, one can also form a more refined understanding of justice after having experienced something just or unjust. We might ask what kind of representational vehicle would be required for representations with unique qualitative features—for instance, to represent the idea of justice in a way that goes beyond a merely symbolic depiction.

The analysis assumes that Kriegel’s attempt to reconcile phenomenal intentionality4 (PIRP) with tracking-based forms of naturalism5 (NERP) ultimately collapses into representational reductionism or fails to achieve coherent theoretical integration. Simply by adopting a PIRP-based framework, one can pursue naturalization of the phenomenal6 that avoids reductive strategies—thus maintaining compatibility with a scientific worldview while preserving the subjective character of conscious experience.

What stands in contrast to the PIRP perspective is an approach to the mind that posits a metaphysical separation between the phenomenal and the intentional mind. Representationalist physicalism recognizes that thoughts are conscious only indirectly, through sensory elements [8]. In this regard, Kriegel aligns himself with the PIRP, thereby distancing himself from the physicalist and representationalist views typical of higher-order theories. Furthermore, physicalism does not acknowledge the independence of anything mental from the physical realm, and this extends beyond the mere metaphysical issue of whether primitive and irreducible properties—such as qualia—exist in a dualistic sense7. According to the Higher-Order Theory and other forms of physicalist theories—including the self-representational theory, which can be considered a version of the Higher-Order Representational Theory [1]—the subjective feel of experience is explained functionally: what it is like to have a quale is just to have the right kind of higher-order representation.

Ultimately, regarding qualia, Kriegel’s account coheres with the PIRP. From this perspective, qualia are not dualistic entities (as in Jackson’s view), but phenomenal properties that ground intentional contents. They are central, yet naturalizable in a certain sense—though not fully reducible.

Here, I offer a brief preliminary overview of the structure of Kriegel’s self-representational theory, introducing its hybrid nature between NERP and PIRP8.

Although Kriegel does not adhere to transcendental phenomenology in the strict Husserlian sense, he draws on some of its insights—for instance, the reflexive structure of experience.

Indeed, Kriegel’s self-representational theory is based on principles that rest on an analogy with Husserl’s theory of inner intentionality [10]. It is precisely through this comparison that Kriegel can be granted a certain reference to the internal structure of consciousness.

According to Husserl, the key point is that the subject’s internal perspective (so-called aspectuality) depends on phenomenal character. The phenomenal does not constitute the act–world relation, or directedness toward the world. Thus, the focus shifts to the structure of the mind—a turn that is also Kantian in its formulation. This involves an investigation into the conditions that enable the mind to refer to an object distinct from itself. It is by virtue of this structure that intentional acts possess the property of being directed toward objects. To Husserl [10], this refers to the phenomenological content internal to the act (at least in part). In line with phenomenology, such content is irreducible to a physicalist objectivity.

Kriegel, in his self-representational theory, adopts this structure of internal reflexivity within mental acts and claims that the manifest states appearing in our personal experience are self-present. According to Kriegel, when experiencing a blue sky, it is the “for me” of the experience that decides whether it is a conscious experience at all. For a mental state to have inner awareness, that is, a subjective character built into it, it is equal to having a directedness at itself9. His theory is also referred to as a ‘Same-Order Theory’ because it asserts the following:

“a mental state, called M, is conscious if the subject has a conscious counterpart of M, called M*, that appropriately represents M. That is, there is a constitutive relation between M and M*. This relation is one of part-whole, where the subject’s awareness is an integral part of the mental state M itself”

[11] (p. 153)

He explains that the conscious state M consists of a mereological sum of the monitoring state (M*) and the monitored state M.

According to him [4], the subjective mode of presentation of specific cognitive content is naturalizable within his self-representational structure of consciousness.

The internal awareness of being in a mental state is part of the state itself10.

In his 2011 account [4] (p. 102), Kriegel maintains that intentionality emerges from the combination of higher-order tracking theory and the experiential origin thesis; the latter conceives intentionality as grounded in the introspective encounter with mental states whose content is in virtue of their experiential character11.

Moreover, he distinguishes between two levels: S1, that is, a mental state directly tracking an object (e.g., a red apple), and S2, a higher-order state tracking of S1 while also representing the object. S1 and S2 include part–whole structures; however, the experience in S1 is conscious by virtue of being tracked by S212.

Indeed, these acts share the same object; however, since a part of the act is reflective and directed toward the relation that state S1 establishes with the apple as object, it also constitutes a reflective act aimed at what is immanent to consciousness. Thus, this reveals how the object is conceived within the subject’s act. Focusing on the relation of intending an object means observing that this is not an ordinary relation, because it is not determined by what is empirically true in the real world. In fact, there may be different intentional acts directed at the same object.

The point is that Kriegel, by adopting a theory such as Higher-Order Theory, reduces these intentional appearances to what is determined by the object or corresponds to the brain’s tracking of the object.

To summarize my entire proposal regarding Kriegel, I first assume that he aligns more closely with the Phenomenal Intentionality Research Program (PIRP) while also incorporating representational and physicalist assumptions that ultimately challenge a strictly phenomenological interpretation.

I observe that Kriegel’s theory and his attempt to naturalize phenomenal intentionality lead to a weak position—between, on the one hand, approaches that seek a physicalist objectivity in the investigation of human experience (NERP), and, on the other hand, those that, in principle, maintain that science must adopt a criterion of objectivity in which appearances in reality constitute an essential part of what exists (PIRP).

2. Intentionality

The issue of intentionality—central to the ontology of worldly objects—as recently framed [4,5] naturally brings into focus two classical programs: NERP and PIRP.

First, intentionality refers to the essential feature of mental states that transcend consciousness and are directed toward an object, a property, or a state of affairs with a mind-independent nature [12]. Discussions of intentionality emphasize that perceptions and thoughts are often conscious and frequently possess intentional content13. Paradigmatic cases of intentional states include examples ranging from visual perception (“I see a book on the desk”) to propositional and doxastic states (“I see that there is a book”; “I believe that there is a book on the desk”).

Intentionality implies the instantiation of intentional properties, whereas consciousness entails the emergence of phenomenal properties [14]. Phenomenal properties, or qualia, are sentient phenomena through which certain subjective aspects of experience are taken to be conscious, as famously described in earlier accounts of consciousness14 [18].

Both representationalists15 and analytic phenomenologists16 understand intentionality in terms of the representational content of mental states or events and agree that experiential states include content17. To better grasp this notion, one must consider the mode of presentation of an object or concept—that is, an aspect of the content associated with conditions of satisfaction [14].

More specifically, a position on the use of intentional terms has been articulated [20], according to which two features are often regarded as necessary for intentionality: directness toward objects and aspectuality18. The first denotes the mind’s directedness toward what is observed in the world; the second captures the specific mode under which the object is presented to consciousness (Sacchi [24]; Voltolini [17,25]).

At this stage, one might ask whether any compelling theories currently offer a satisfactory account of intentionality. According to Chemero [26] (p. 626), no widely accepted definition of representation has emerged in philosophical analysis. As a result, many cognitive scientists working with computational systems advocate for anti-representationalism.

In practice, Chemero challenges representational theories of mind, favoring a dynamic account of adaptive behavior—one that does not require postulating internal representations. This view directly opposes tracking theories19, which posit a fixed causal link between internal content and external world states. Above all, it should be emphasized that this also appears to clash with positions such as those proposed by Kriegel’s self-representational theory. In contrast, the dynamic approach assumes that a cognitive system is best understood as a constantly evolving complex system, rather than as a collection of static internal representations.

Since the present discussion involves concepts central to philosophical naturalism and related foundational notions, I will postpone its elaboration to the next sections, following a set of preparatory remarks.

3. The Main Ideas Behind Naturalist–Externalist Program

NERP seeks to naturalize intentionality by explaining it within a broadly scientific framework grounded in empirical methods20. The central challenge is to account for representational content—that is, intentionality—in physical terms and to clarify how a physical organism can be intentionally directed toward the world [27]. According to this view, a given sequence of words—or an internal neural activation pattern—acquires its meaning from the causal relationships it establishes with elements of the external world. For instance, the term rabbit refers to rabbits precisely because the occurrences are typically triggered by the presence of rabbits in the environment [28].

Within this paradigm, an additional issue concerns how rationality presupposes intentionality and how causal transitions between mental states are to be explained.

Numerous proposals have attempted to identify mental representation with properties of mind embedded within the natural order [29,30]. With respect to the relation between intentionality and phenomenal aspects of cognition, reductive representationalism—as developed by authors such as Dretske, Lycan, and Tye—maintains that intentionality is metaphysically independent of phenomenal features. Moreover, as McGinn observes, the conditions for mental content are neutral with respect to whether the bearer has a conscious mental state. In principle, the same contents could be instantiated by non-conscious creatures; consciousness is therefore not constitutive of content.

Moreover, naturalism21 contrasts with radical realism, which denies reality to mind-dependent entities22.

Within the naturalist project, the central aim is to explain how physical mental states encode information about the external environment from which they derive their content23. The externalist approach stresses the role of environmental interaction and rejects the claim that intentionality is wholly internal24.

The causal theory of representation—including both an informational approach and an asymmetric dependency account—illustrates this naturalist strategy [34,35,36]. On such views, intentionality is a natural, non-semantic relation between symbols and the entities they represent, as in the classic example of smoke indicating fire. Further elaborations include teleosemantic and teleoinformational theories, which ground intentional content in evolutionary and functional explanations25.

A central objection concerns the reduction in phenomenological properties. Representational tracking theories, in particular, face serious difficulties. Internal states with qualitative features—such as pain or other unpleasant sensations—cannot be causally tracked, since only causally efficacious properties are available for monitoring. A related difficulty extends to tracking representationalism, a hybrid view that combines representationalism with tracking theory and aims to explain all phenomenal appearances [37].

Dretske and Tye respond by maintaining that a tracking relation26 can nonetheless account for qualitative experience. In the case of pain, for instance, the subject represents its negative character as bearing on their well-being, and this evaluative dimension explains the experience’s content27. What is tracked, then, is not merely the causal property itself but its relevance to the subject—a relevance that is inherently subjective and varies across individuals and cultural contexts [41].

A further difficulty concerns misrepresentation. Tracking theories appear to imply that whatever causes a mental state is thereby included in its content. Yet, such accounts struggle to accommodate cases of misrepresentation [42] (p. 422). We frequently misrepresent the world—for example, in illusions or hallucinations28.

There was considerable controversy over whether the details of specific tracking theories29 can be worked out, or whether they deliver the intuitively correct content assignments in specific cases [34].

Mendelovici argues that tracking theories of intentionality struggle to accommodate misrepresentation. For example, in cases of color illusion, some of our color experiences are mistaken, and it seems that the tracking theory can never misrepresent, since anything that causes or correlates with any of its tokens is automatically part of its content [42] (p. 422). In contrast, Phenomenal Intentionality Theory (PIT) explains misrepresentation more plausibly by grounding intentionality in subjective experience. Because PIT locates the source of intentionality within first-person phenomenology, it allows for the possibility that subjective appearances can deviate from objective reality.

Dretske, in his book [36], addresses the problem of misrepresentation by introducing the concept of teleological function. According to him, a mental state represents something not merely because it is caused by it, but because it has been selected—biologically or functionally—to represent it. What happens when you have a hallucination of a rabbit, but there is no rabbit in front of you? Our visual system has been designed (by evolution or by learning) to respond to the presence of rabbits. Even if, in a specific case (e.g., a strange shadow), your brain mistakenly generates the representation of a rabbit (a hallucination), the mental state still has the function of representing rabbits. It allows for explaining intentionality in naturalistic terms, without invoking consciousness or subjective experience.

Moreover, from the standpoint of causal-informational theories of content, Dretske [44] argued that intentionality requires a causal tracking relation to properties in the real world. Within the NERP framework, cognition can occur without the need for internal representations, as long as systems can track real-world properties—like magnetosomes detecting the direction of anaerobic water. Tracking, therefore, takes priority over symbolic manipulation.

This underpins strong critiques of AI systems—especially generative models—that rely on probabilistic processes rather than symbolic reasoning. Dretske argues that intentionality cannot be captured in purely syntactic terms; a symbol only has meaning if there is a causal or informational connection with the real-world entity it represents. However, models like Jukebox30 merely abstract statistical patterns from data, lacking such causal grounding—thus falling short of genuine intentionality.

AI systems might be capable of exhibiting human-like behavior, but can mere symbol manipulation ever result in genuine understanding, or intentionality, let alone consciousness?

John Searle’s Chinese Room Argument is, in fact, directed against a strong conception of artificial intelligence. He demonstrates that Strong AI31 is false because it claims that there is a program for Chinese such that if any computer system runs it, it thereby comes to understand Chinese. However, since it is possible to run a program for Chinese without thereby understanding Chinese, the initial assumption must be false.

As we see in the following sections, the key shift introduced by AI lies in moving away from a conception of intelligence based on symbolic manipulation, as typically assumed in classical cognitive science. First, in contrast to traditional algorithmic approaches32, connectionism33 seeks to emulate features of the biological brain through the construction of artificial neural networks (ANNs). These connectionist systems rely on intermediate processing units, commonly known as hidden layers, that enable the emergence of complex transformations34. ANNs simulate the connections between biological neurons in the brain in order to recognize patterns in complex or noisy input and are notably effective not only for pattern recognition but also for pattern generation35.

One of the main difficulties associated with connectionist structures is the so-called black box problem: there is no transparent account of how such networks generate their outputs. Despite this, research in the field has led to the development of deep convolutional neural networks (DCNNs)36, which have enabled a range of innovative technologies37.

4. The Phenomenal Turn: Phenomenal Intentionality Research Program

A central objection to the naturalist paradigm in the philosophy of mind and perception comes from the phenomenological investigation of experience. From this view, intentionality is understood as a property of conscious states—so-called underived intentionality. For many phenomenologists, this also implies that intentionality cannot be explained in semantic terms, since it is irreducible to language. Within this framework, attributing intentional content to machines is merely metaphorical: there is no content in the world without consciousness.

According to this view, a sequence of words or a neural activation pattern acquires its meaning through its connection to modes of presentation38 that uniquely identify particular objects or properties in the world. For instance, the word rabbit is taken to mean rabbit because it is tied to a descriptive cluster, such as a small, furry mammal with long ears, known for hopping, eating vegetables, and often depicted in children’s stories or cartoons [28].

Moreover, intentionality possesses a first-person dimension. The content of an experience and its subjective features are inseparable39 [27,53]. This concerns the very constitution of intentionality, including cognitive intentionality40. Building on this, there is a distinction between two kinds of content: conscious and non-conscious: “There is a kind of intentionality, pervasive in human mental life, that is constitutively determined by phenomenology alone” [47] (p. 520)41. Not all authors, however, agree that the two standard properties of intentionality—directedness and aspectuality—depend on consciousness. From one perspective, only aspectuality depends on consciousness [25] (p. 5) and [61]. A related debate concerns whether aspectuality42 should be considered in terms of subjective or objective appearances of the object of thought43. According to McGinn, phenomenal character—understood as a subjective aura or the felt cognitive character of experience—is a constituent and determiner of content. The objection to this view is that it risks rendering content subjective44 to such an extent that it becomes impossible for two subjects to entertain one and the same content45.

One position that is generally accepted for explaining first-person judgment is that of interactionist dualism [1] (p. 228). According to the interactionist theorist, when one judges oneself to be in pain—that is, when one has first-person awareness of it—this judgment can be explained by appealing to an effect caused by a causal interaction with the phenomenal property (e.g., redness or painfulness). This judgment can be caused by cognitive qualia.

On this view, experience arises from the capacity to articulate sentient aspects in phenomenal judgments, involving the use of semantic—and subsequently linguistic—categories. This process presupposes self-awareness as a cognitive resource. Above all, introspection is conceived as a possible means of access to one’s own states and beliefs.

5. Kriegel’s Main Arguments and Their Critique

At its core, Kriegel’s project seeks to explain intentionality by grounding it in consciousness—specifically through a Higher-Order Theory framework. However, he ultimately fails to resolve how the physicalist assumptions of Higher-Order Theory can be coherently reconciled with the phenomenological commitments of the PIRP framework.

Indeed, Kriegel explains that there is a monitoring relationship between the monitoring part of the mental state and the monitored part. However, one might question whether internal content, dependent on the mind, can be objectively identified or traced.

Thus, Kriegel may adopt a physicalist theory—namely, one that explains consciousness in terms of intentionality. However, this is the key point: Kriegel cannot coherently adopt a reductionist position while also maintaining a phenomenological commitment to subjectivity. This leads to a departure from a research program grounded in subjectivity46.

Moreover, since the subjective standpoint is a constitutive element of human experience, one may infer their inseparability; however, it does not follow that human experience is what the subjective standpoint is metaphysically reducible to.

In order to naturalize phenomenal intentionality—that is, the view that the intentionality of mental states is grounded in their phenomenal character, namely, the qualitative, subjective “what-it-is-like” aspect of experience47—Kriegel asserts that he must first naturalize consciousness itself.

The theme of the phenomenal mode in which something appears to the mind is also addressed in relation to non-existent objects, as part of a test case against higher-order tracking theories [4] (p. 146). While we cannot experience the properties of a square circle, we can experience an Escher Triangle48—precisely because it exists as a pictorial representation. The experience appears to track the property of being an Escher Triangle. However, if no such property exists, tracking cannot occur in the strict sense. Kriegel concludes that intentional inexistence resists explanation through tracking higher-order theories. Nevertheless, he ultimately considers this theory the most promising account of intentionality [4] (p. 173).

He further argues that there can be no scientific causal explanation of experiencing a chimerical colored dragon [65]. Experience may lack an object: no non-existent referent is needed, since the phenomenal mode of presentation is sufficient to structure it. Kriegel thus also rejects sense-data and abstracta.

Here begins the central point of my critique of Kriegel: his hybrid position and his endorsement of Higher-Order Tracking49. According to him, the mode of presentation is not grounded in mere external tracking, but in a conscious internal insight, reducible to a higher-order act. Consciousness thus grounds intentionality—there is no intentionality without phenomenal awareness.

However, Kriegel maintains that conscious states are those in which the subject is, in some way, aware of being in such states. He argues that a mental state is conscious insofar as it monitors itself. Yet such self-monitoring already presupposes a functionalist framework.

Every X that attempts to explain Y must be ontologically or epistemically prior to its explanandum, Y. That is, X must contain the concepts, structures, or tools that allow us to define and recognize Y. If X is a functionalist theory of consciousness and Y is the phenomenon to be explained, then Y is identified through X. But if we can only know what Y is via X, Y will necessarily be described as a functional system. As a result, if Y is a functional system, it must be construed as intentional. Therefore, this leads to the conclusion that intentionality precedes consciousness. This questions Kriegel’s assumption that consciousness is prior to intentionality.

If consciousness is prior, can it be explained by functionalism? If they are the same, one cannot explain the other. Furthermore, the self-monitoring of a mental state presupposes intentionality, in contrast to the Experiential Origins of Intentionality thesis.

To illustrate the point, Kriegel’s hybrid account rests on several core commitments:

(1)

Endorsement of phenomenal properties50 as carrying aspectual accuracy conditions, which structure intentional content51 (pp. 88–93);

(2)

Recognition that experiential character is determined by how the subject is aware of her internal state [4] (p. 90);

(3)

Acknowledgment of the Higher-Order Tracking Theory as an approach to conscious experience [4] (p. 97), despite the fact that the higher-order view posits intentionality as the underlying structure to which consciousness is reduced;

(4)

Acceptance of the compatibility between the higher order tracking theory and the naturalistic approach to conscious experience52 (p. 97).

As he writes, “a mental state is conscious in virtue of representing itself in a distinctive way” [4] (p. 90); this self-representation, in turn, grounds its intentional content53. Two options follow: either the phenomenal mode of presentation is taken as a phenomenal property grounding representation, in which case Kriegel aligns with PIRP, or it is claimed to be naturalizable through self-representation.

Thus, Kriegel’s attempt to ground intentionality in phenomenal properties proves unconvincing.

At this point, one influential interpretation of presence-to-the-subject (inner awareness) [2,3,67,68] as an intentional phenomenon comes into focus. Kriegel objects to McGinn’s thesis, according to which intentionality depends on consciousness, arguing instead that inner awareness is itself intentional54 and that its presence in intentional states does not imply such dependence55. In his reading, Searle represents the stronger view, grounding intentionality in consciousness through aspectuality, while McGinn ties it to presence-to-the-subject in terms compatible with naturalism. Therefore, Kriegel’s theory cannot be included in PIRP, since phenomenal experience is reduced to intentionality, confined to the act itself.

Regarding the view that intentionality is grounded in and reducible to subjective presentations, Kriegel proposes three possible solutions:

(1)

Subjective representations are indispensable and genuinely exist, possessing an ontological status over and above objective representations;

(2)

Subjective representations can be explained by objective representations;

(3)

Objective representations can be reduced to subjective ones [39] (p. 175).

He defends the view that subjective representations may ultimately be reducible to objective ones. Moreover, he stresses that, although tracking theories of mental representation do not yet provide a fully reductive account, they nonetheless keep open the possibility of such a reduction. Thus, the naturalization of phenomenal intentionality seems to hinge on whether subjective mental representations can be reduced to, or identified with, objective ones—a claim maintained despite the difficulties this poses for tracking theories.

6. Aiming Toward a More Scientific Account of Intentionality and Thinking Entities

In parallel, Kriegel claims that intentionality is a “primitive property” accessible through introspection56 [4]. Critics argue that intentionality is not directly sensed but a theoretical construct, real like atoms in physics, yet absent from immediate experience. However, it may be noted that in a position seeking to ground the understanding of reality in perspectives that integrate the subjective standpoint, even an entity such as the atom—though abstract—is nonetheless something upon which our experiences depend.

Yet, what ultimately matters is that, since experience may be misleading or incomplete, it cannot provide a reliable foundation for conceptual development.

Regarding this issue, Kreigel clarifies that his account of intentionality concerns our epistemic relation—how we define it and seek a more scientific understanding of the phenomenon. He treats intentionality as a natural concept, claiming it can be directly perceived and that its instances function as observational concepts, identifiable through nomologically possible observations [4] (p. 23 and p. 54). Similarities among cases, he argues, are observer-independent and grounded in underlying features revealed by empirical investigation [31,69,70,71]. Given this type of input, the moment one knows this sort of portrait, intelligence agents can predict any similar phenomenon, despite changes and multiple variables. This implies that psychological systems can reliably identify instances of a given type of object. Yet Kriegel’s commitment to phenomenal, sense-derived properties ties intentionality to constant observation, making it difficult to secure the stability required for reliable prediction.

Interpretive insights may be found in studies on machine learning within artificial intelligence programs. This may involve designing optimal programs for intelligent agents.

Self-monitoring theories, developed within neuroscience and cognitive science, describe functional models in which a cognitive system can monitor, evaluate, or regulate its own mental states and behaviors57. Instead, self-attention architectures are core components of Transformer models58 in generative artificial intelligence, where each input element can dynamically access and assign weights to all other elements, including itself59.

These models can offer only formal analogies or heuristic tools for exploring how self-ascriptive capacities might emerge in cognitive systems, without presupposing genuine first-person awareness60. It is noteworthy that the latter is expressed differently from the NERP and PIRP perspectives.

Unlike the idea of a personal mind, within the NERP framework, awareness of a mental state may arise either from a higher-order judgment or from the state itself. Higher-Order Thought theory holds that awareness of a mental state originates at the sub-personal level. Consciousness can be approached bottom-up, emerging from sub-personal processes, but not from a top-down perspective, as in phenomenology, in which it arises through introspective access to mental states. On the former account, there is no introspective route by which unconsciously ‘cognitive’ mental states become conscious. The difference emerges at the theoretical level in relation to the explanation of experience. Indeed, within a phenomenological framework, a conscious mind is posited when we hold that we perceive not a computer as such, but something as a computer. Such experience requires a self, understood as the source of one’s own bodily movements.

Foundational models represent another type of system. The most striking consequence of foundational models is that they appear to operate according to internally driven representational logics that, in both structure and function, recall the fundamental features of the Phenomenal Intentionality Theory (PIT). As in foundational models, meaning emerges from learned internal structures rather than being explicitly encoded; similarly, in the PIRP, meaning is intrinsic to phenomenal experience rather than imposed from without. These foundational models are large-scale neural networks61 trained on massive datasets, designed to be adaptable across a wide range of domains. While dynamical models, as defended by Chemero [26], reject internal representations altogether, foundational models operate differently. They are systems that build internal latent representations encoding statistical regularities across large-scale corpora. This means that the system does not follow a genuine program, yet it can produce outputs that elicit beliefs in users closely aligned with general expectations, and sometimes even exceed them. Consider, for instance, simulated human “chats” in systems such as ChatGPT (version 5).

Could a machine think?62 Searle argues that thinking is possible for machines that possess the requisite internal causal capacities, such as the brain. Nevertheless, no program alone is sufficient to constitute genuine thought. In this context, intentionality in human beings is a product of the brain’s causal properties—specifically, the causal relations between mental processes and brain activity. However, the explanation of how the brain produces intentionality is not that it does so through a program. Intentionality does not derive from programs, because both a human brain and a machine can learn a code without understanding that code63.

A human agent could operate a program without possessing external intentionality. This is precisely one of the objections raised, for instance, by proponents of the NERP.

Kriegel, in turn, with his thesis of underived intentionality64 raises the question of whether consciousness can truly explain unconscious intentionality—that is, the intentionality of non-conscious or sub-personal states (e.g., visual processing)65.

According to Kriegel, it is clear that we can regard unconscious intentionality as identifiable and explicable, provided we adopt a third-person standpoint66.

That study focuses on different types of aboutness or intentional directedness. This opens up several interpretive levels. In particular, we may ask whether aboutness grounded in phenomenological modes of presentation differs from aboutness that is merely derived. This raises questions concerning functional systems and self-monitoring theories, in which a cognitive system can regulate its own mental states through self-monitoring mechanisms—without necessarily involving a phenomenology of self-reference.

Different from Kriegel, to better understand the differences and similarities between these systems and the human mind, we might consider an example of the embodied mind. Consider the case of a wine expert whose refined palate and ability to infer subtle differences between wines emerge through sensory experience and internal processing. Had the same information been conveyed through images or theories, the expert’s taste likely would not have changed. This shows that embodied experience and sensory nuance create real changes in perception and knowledge.

The embodied mind67 makes inferences precisely because it is embodied. Consequently, if we maintain that foundational models involve a non-embodied mind, which cannot replicate the embodied learning and transformation that occur in an agent’s physical interaction with the world, then we must acknowledge that the relationship with the physical environment plays a crucial role.

The extended mind theory outlined by Chalmers and Clark [74] even accepts the possibility that insights can originate externally and that, in Otto’s case, the insight is retrieved from an external source that interacts with his mind. The extended mind theory holds that Otto possesses an insight that counts as part of his mind—even though that insight is retrieved through information stored in his notebook. Similarly, the wine expert’s refined perception is the product of a mind that is extended into the body itself. In other words, foundational models can hardly reflect the phenomenal content as an embodied subject experiences it.

The critique also involves Kriegel’s model of intentionality [4]. Referring to the distinction between non-derived and derived intentionality, we can argue that this distinction—when applied to simulated mind models—does not provide a workable definition of intentionality, since it offers no clear criteria for application. For example, foundational models do not fit the definition of non-derived intentionality (the phenomenal mode of presentation), as argued earlier. Consequently, the concept of derived intentionality also falls short for these models, since it only gains meaning in contrast to its opposite—namely, in models of mind that support non-derived intentionality68. Thus, Kriegel’s distinction of mental intentionality cannot be applied within a theoretical paradigm that aspires to bridge the boundaries between human and artificial minds.

7. Final Remarks

Theories of intentionality divide broadly into two strands. The Phenomenal Intentionality Research Program (PIRP) emphasizes the irreducibility of intentionality to empirical research, grounding it in the structure of conscious experience. In contrast, naturalist accounts aim to reduce intentionality to empirical terms, often through causal or tracking models.

A leading approach within the PIRP framework seeks to base intentionality on phenomenal cognition [4,5]. However, it often attempts to account for phenomenal properties themselves by appealing to a conscious tracking style. This move risks diluting the very subjective foundation that the PIRP claims to prioritize. Indeed, more precisely, this strategy appears to align with the positions of intentionalists, who argue that phenomenal consciousness can ultimately be explained in terms of intentionality—that is, they hold that phenomenal states are themselves a form of intentional states. However, as a hybrid theory that emphasizes phenomenal aspects, it risks collapsing intentionality into phenomenal properties while failing to provide a coherent account of naturalization.

Moreover, if phenomenology is conceived by Kriegel as constitutive of directedness, the object cannot be independent of its phenomenological character. Perception would be infallible, yet its degree of universality would be absent, entirely private69.

The main conclusion of this article is that recent AI architectures offer interesting heuristic proposals for grasping aspects of the mind that continue to attract interest and remain subject to ongoing debate [26,74,75,76].

For example, foundational models lack embodied minds, yet they produce forms of artificial intelligence and, in this sense, qualify as minds. Certainly, the AI’s understanding of meaning is purely statistical, rooted in pattern recognition rather than conscious awareness. Yet, this does not mean that it is not creative.

What remains distinctly human, instead, is our responsibility to calibrate their use and contextualize their applications. In fact, representation learning does not know what it is representing or for what purpose, and it is not aware of or responsible for that purpose—nor does it have any understanding of the meaning behind its outputs.

8. Materials and Methods

This conceptual–analytic review employs a philosophical methodology grounded in textual analysis and systematic comparison. Primary sources include works by Uriah Kriegel and authors associated with both the Naturalist–Externalist Research Program (NERP) and the Phenomenal Intentionality Research Program (PIRP). The analytical approach consists of reconstructing key arguments, highlighting points of divergence, and critically assessing the explanatory power of Kriegel’s hybrid model in relation to both paradigms. Special attention is given to the phenomenological tradition, including transcendental methods that consider pre-reflective and unconscious dimensions of intentionality.

In addition, the article situates Kriegel’s model in the context of current philosophical discussions on artificial intelligence systems, arguing that contemporary debates on the intentional and phenomenal capacities of non-biological agents offer a productive lens for critically evaluating the limitations of hybrid accounts like Kriegel’s.

No empirical data were collected or generated for this study. This analysis remains entirely theoretical and interpretative.