Bridging the Emotional Gap: Philosophical Insights into Sensual Robots with Large Language Model Technology

1 Introduction

Many robots are designed only for industrial or other purely pragmatic uses, where the sociality and sensuality of the robot is not important. The requirements to develop robots with emotional sensitivity, capable of interacting with humans in a human-like manner in a social context, have brought a number of new questions to human–robot interaction research. Addressing these questions requires clarification and rethinking of concepts such as perception, emotion, and sensuality – concepts that have traditionally been part of the philosophical investigation of human sociability and cognition.

The lack of emotionality and sensuality, which currently represents in many cases a gap in interaction between humans and robots, may be an opportunity for a new technology to transform their relationships and take robot–human coexistence to a new level. This gap can be gradually bridged as more knowledge is gained about the relationship of human speech and other aspects of human behavior to human perception of emotions and social relationships.

This article explores the intersection of sociability, emotionality, and sensuality in robots in the context of the new technology of large language models (LLM), which has great potential to enhance these properties. Starting from an exploration of the philosophical underpinnings of the concept of sensuality, we then analyze what is necessary for robots to realistically simulate social, emotionally convincing interactions with humans. We also cannot avoid some aesthetic and physiological considerations in this investigation.

The study also provides a brief overview of existing robotic technology and a brief description of large language model technology, focusing on how this new technology can help overcome the limitations of current robots. The text discusses the advances that have been made in the design of robots designed to work in social contexts, robots focused on nuanced interactions with humans, from the earliest attempts to state-of-the-art humanoid robots. We conclude that the application of large language model technology offers transformative potential for the creation of advanced robots whose capabilities will far exceed those of older generation robots, robots that will not only be significantly more socially competent, but also far more capable of interacting with human users in ways that are personal, sensory expressive, and emotionally resonant.

The rest of the article is structured as follows: Section 2 will be devoted to a general definition of the role of the senses and sensuality in the context of philosophy, both in humans and in the context of robotics. In Section 3, we will define aspects of human bodily sensuality independent of linguistic expression and provide a brief overview of prominent humanoid robots designed for social and emotional interaction, summarizing the main drawbacks of linguistic communication based on traditional algorithms and rules. Section 4 briefly introduces the technology of large-scale language models both from a technical perspective and in a philosophical context. Section 5 briefly delineates the physical and linguistic characteristics relevant to social robots, defining four fundamental characteristics necessary for robots to convincingly simulate social, emotional, and sensory interactions. This introduces the section on robots enhanced by large language model technology and their potential, while showing that the use of this technology addresses the shortcomings we identified in Section 3. The text closes with a brief conclusion.

2 Senses and Sensuality: From Human Senses to Robotic Simulations

The word “sense” denotes primarily our “physical abilities to see, hear, smell, taste, and feel” but also “an ability to understand, recognize, value, or react to something.” Similarly word “sensation” denotes “the ability to feel something physically, especially by touching, or a physical feeling that results from this ability.”^[1] In the second sense, as a physical feeling, a sensation is a mental state. These words come from the Latin word “sēnsus” meaning sensation or feeling. Humans have five basic senses – sight, smell, touch, taste, and hearing, which are provided by the sense organs. Other animal species may have different sensory organs and different senses, just as robots may have different senses, such as Kinect sensors.^[2] Related Old French word “sens” denotes not only the five physical senses but also wit, understanding. The Greek word “σύνεσις“ with a similar root “syn” means unity, but also the faculty of quick comprehension or intelligence. There is a different Greek word, “αισθήσεις,” which refers to our five senses and from which the word “aesthetics” is derived.^[3] This word in its modern meaning was introduced by the German philosopher Baumgarten in his work of 1735.^[4]

The related German word “Sinnlichkeit,” which Kant used as a central concept of his philosophy, is usually translated as “sensibility” (sometimes also as “sensuality”). This term is used for the capacity to acquire representations – the perceptive faculty, which is constitutive for the spatial and temporal organization of sensory experience.^[5] According to Kant, space and time are pure intuitions that lie a priori at the basis of the empirical experience; they are mere forms of our sensibility.^[6] He understood sensibility together with understanding (“Verstand”) as two stems of human cognition.^[7] Understanding is the term Kant used for the spontaneity of human cognition, the faculty for thinking of objects of sensible intuition.^[8]

The word “sensuality” is sometimes used in philosophy as a collective term for various human abilities that mediate perceptions.^[9] In common discourse, the term “sensuality” refers to the expression of sensory pleasures, usually in relation to the body and desires in a sexual context. The dictionary defines this term as “the expression or suggestion of physical, especially sexual, pleasure or satisfaction.”^[10] We can attribute sensuality to persons, to human performances and actions, but also to inanimate objects, especially those made by man and therefore, as we shall see, to robots.^[11]

In common discourse, “sensibility” often denotes an emotional or artistic awareness,^[12] understanding of, or ability to decide about what is good or valuable, especially in connection with artistic or social activities.^[13]

In summary, when we speak of sensuality, it is always in relation to the bodily senses, whose functions are provided by the sense organs, in the case of humans or animals, and by analogy, robots. Properly functioning sense organs enable the conscious perception of sensory sensations, which are shaped by the perceptual apparatus and thus constitute sensory experience. The sensuality then often refers to the expression of physical, especially sexual, pleasure.

The area of sensory experience, sensation, sensuality, and other related concepts is complex and is conceptualized differently by different philosophical traditions and spans a number of philosophical disciplines, including aesthetics, ethics, metaphysics, and phenomenology.

From an epistemological perspective, sensation refers to how an individual acquires knowledge through the senses. Empirically oriented philosophers, beginning with Aristotle, have understood sensory experience as the basis of empirical knowledge.^[14] Robots can feign human sensuality by responding to stimuli in expressive ways that mimic human pleasurable responses to sensory input, but they do not experience these sensations in a human way. Traditionally understood human sensuality is based on a subjective, embodied capacity for experience that is absent in robots. Robots process data sensed from the environment with sensors that are functionally similar to human senses, but this processing is a syntactic computational transformation of input to output and is not accompanied by internal experiences similar to those of humans.^[15]

Sensuality is also important from a phenomenological point of view: for example, Merleau-Ponty considered perception as the primary way in which people relate to the world through their corporeality. Sensuality and perception are a fundamental part of human existence and are closely related to the experience of the body, as Merleau-Ponty makes explicit in his famous lecture The Primacy of Perception. Here he rejects the traditional distinction between substance and form, saying that “Matter is ‘pregnant’ with its form,” resulting in a “quasi-organic relation of the perceiving subject and the world.”^[16] Since perception is an original modality of consciousness, whether robots possess perception in the proper sense is inextricably linked to the question of whether they have consciousness.^[17]

Merleau-Ponty takes up the German “gestalt” psychology and develops it in a philosophical way. Gestalt is “a spontaneous organization of the sensory field,”^[18] which consists in the interrelation of elements and wholes, more or less stably organized. The laws of Gestalt have been precisely described by psychologists, and Gestalt theory has been studied in the context of computer vision, so it is generally possible that the perception of a robot is organized by the same Gestalt principles as human perception.^[19]

In contrast to Descartes and the common view of the body as a biomedical object or as a mechanical instrument of human consciousness, Merleau-Ponty takes a different position. The very nature of perception and its simple scientific reflection naturally leads to a Descratic objectification of the body, and it is necessary to show the limits of this approach. Merleau-Ponty phenomenologically studies the body in several domains, of which the synthesis of one’s own body, the relation of body to speech, and sexual being of the body are particularly relevant to our investigation of robots.

Each person’s own body is not just one object among many others; we are not observers of the relationships between parts of our body, limbs, “we are ourselves the unifier of these arms and legs, the person who both sees and touches them.”^[20] The human body thus has a double status – it is both object and subject. Our perception is embodied, and in this sense, the prerequisite for full-fledged robots is also their corporeality and perception integrated into their motor bodies, so that they too are not just observers in relation to their limbs.

When we move, we do not use our limbs as tools, as some object-means that we manipulate to achieve a goal, but our practical intentionality is pre-reflexive; we act directly so that we achieve the goal. Our body provides us not only with a perception of the world, but also with a kinesthetic sense of its own movements, which is based on what Merleau-Ponty calls the “body schema.”^[21] In this, our bodily experience is not isolated; we perceive other people (and potentially robots) through their bodily expressions. Current humanoid robots are able to acquire new motor skills by kinesthetic teaching.^[22]

Merleau-Ponty pays separate attention to the sexual nature of corporeality. Here again he argues against a simplistic schema in which sexual perception is based on the capacity to experience pleasure and pain together with a human-specific power of representation whereby “affectivity is not recognized as a distinctive form of consciousness.”^[23] Patients who suffer from physiologically based sexual affectivity disorder have a modified overall perceptual approach to the world: they are absentminded, they “do not throw themselves into what they are doing.”^[24] Merleau-Ponty also rejects the Descartesian model in the case of sexual perception: “Erotic perception is not a cogitatio which aims at a cogitatum; through one body it aims at another body, and takes place in the world, not in a consciousness.”^[25] In this sense, it is questionable whether any form of human-like sexual perception and sensuality is fully achievable for robots.

Merleau-Ponty’s work is also important to our exploration because it deals with the relationship between corporeality and speech. The mouth, the organ of speech, is not only fixated on sexual existence, but also on “those relations with others having the spoken word as their vehicle.”^[26] He maintains that “of all bodily functions speech is the most intimately linked with communal existence” or as he calls it in other parts of his text – with co-existence.^[27]

In this sense, even a robot, possessing the ability to communicate through speech, is able to participate in this co-existence of human beings, even if it will be deficient at the present technical level – deficient at the level of sexual communication linked to an erotic form of perception. The real ability to use language means the ability to transcend the individual and unique to see it as representative of an essence or a category.^[28] That is why he emphasizes the essentially intersubjective, intercorporeal dimension of language.^[29] The current level of large language models that can support robot communication can already be considered to meet this requirement, and in this sense, the linguistic competence of robots currently exceeds their perceptual competence.

Merleau-Ponty does not regard words as external signs superimposed on internal concepts or ideas. The meaning of speech is formed in the act of speaking, which is an act of the body. Speech is not the product of a transcendent soul but is embedded in our body’s being in the world. The ability to speak is as much a bodily faculty as the body’s ability to grasp an object. Just as our body walks smoothly without our having to pay attention to each step, it is also capable of producing speech. In this sense, the integration of robotic corporeality and large-scale language model technologies, providing the ability to speak fluently, is also a prerequisite for bringing artificial intelligence closer to the human mode of communication. Human use of speech is based on human corporeality, in the same way speech produced by AI can be taken to a new level by embedding it in the physical body of a robot, localized in the world.

According to Merleau-Ponty, a prior thought is not a prerequisite for speech in time. Often, the reverse is true; speech comes first, and it is only by reflecting on what we have said that we fully grasp and develop in our thoughts the meaning of what has been said.

Robots, unlike humans, do not have a subjective “being-in-the-world”; they are not “at home” in the world, nor are they aware, in the human sense, of their perceptions.^[30] Thus, their sensuality is of a different character from the bodily human sensuality described by Merleau-Ponty. However, robots can simulate behavior that, from a human perspective, can be interpreted as sensual. If we were able to develop some form of subjective consciousness in robots as part of advanced AI research, robots might subsequently actually experience a kind of “machine sensuality.” The question remains whether this would necessitate a new definition of sensuality that would have to include these new, non-biological forms of experience.

The philosophy of Merleau-Ponty (and others, e.g., Lévinas) is criticized by Harman in his book Guerrilla Metaphysics. Here, he introduces the term “carnal phenomenologists,” by which he refers to thinkers in the phenomenological tradition who emphasize the embodied, sensory experience of phenomena. In the work of these authors, “we find ourselves mesmerized by the objects in the world, rooted in a carnal setting where our bodies meet with the voluptuous textures of entities.”^[31] The term “carnal” refers to this sensual, lived experience of the world, which these authors consider crucial.

Harman sees this conception as too human-centered, though he appreciates the ability of the phenomenological approach to describe human experience accurately. Harman believes that there is a need to break free from a purely human perspective and develop a metaphysics of objects themselves that includes also a description of how objects behave toward each other independently of humans. For Harman, objects are more than their phenomenologically describable qualities, and they also have an independent reality that is never accessible to us; he subscribes to a movement he calls speculative realism.^[32] Sensory objects thus play a distinctly different role in Harman than in Merleau-Ponty.

We may ask ourselves whether we should give preference to Merleau-Ponty’s phenomenology with its anthropocentric features, which emphasizes perception within a unified body and experienced sensuality, or to Harman’s speculative realism, which rejects anthropocentrism (for Harman, transcending anthropocentrism is key to the nonmodernist outlook^[33]) and whose ontology attributes reality to objects independent of humans.

Merleau-Ponty’s philosophy, on the other hand, is fundamentally anthropocentric, even though in perception we naturally encounter objects on a phenomenal level that are significant to us both epistemically and in terms of sexual perception. However, what is essential is always what happens in the phenomenal field, while Harman is also interested in how non-human objects interact when they are not perceived.

We can find a middle ground between these two philosophical approaches if we recognize that our perception is always bound to our body and is therefore subjective, while attributing independence to non-human objects. Such an approach will allow us to draw on both Merleau-Ponty’s phenomenological observations and Harman’s autonomous ontology when considering the corporeality of robots and artificial sensuality.

From an ethical perspective, sensuality is often discussed in moral philosophy because of its importance in the study of the moral value of pleasures. For the Epicureans, for example, moderate sensual pleasures were an essential part of a happy life; the Stoics, by contrast, favored restraint and rational self-control in relation to sensual pleasures.^[34] In the case of Aristotle, the role of sensuality in his conception of a happy human life is still debated today.^[35] In contemporary philosophy, research on sensuality has gained importance in different senses, especially in relation to identity or gender. Feminist writers, for example, have addressed the question of how sensuality is constructed, expressed or, conversely, repressed in the context of different cultural narratives.^[36]

Social robots that interact with humans, e.g., in a caregiving role, can be programmed to simulate emotionality and sensuality, allowing stronger emotional bonds to form. This raises a number of ethical issues related to the themes of authenticity, sincerity, and manipulation. Can the simulation of sensuality in robots be considered inherently deceptive?

From an aesthetic point of view, sensuality is associated with the perception of beauty through the senses, such as sight and hearing. Beauty here can be both the beauty of natural origin and the beauty of works of art. In the context of the beauty of works of art, sensuality is often a topic of discussion in the various art forms, because each art form requires the involvement of different senses, and this involvement is essential for the full experience of the works of art. One can often perceive artists’ efforts to involve as many senses as possible simultaneously in the perception of their work.^[37]

Natural beauty is related to the beauty of the human body and is related to the themes of erotic attraction and sexuality. The word sensuality in this context often implies a total devotion to the pleasures of the physical senses. In this more intimate context, the term sensuality refers to those aspects of the body that can evoke physical sensual attraction of an erotic nature. For example, Bataille grounds erotic attraction in the “inner experience of desire.” Attraction does not come from some objective quality, but must “somehow touch our inner being.”^[38] He understands eroticism to denote that form of human sexuality which transcends animal sexuality. The emergence of man from his pre-human ancestry is linked to the emergence of sexual taboos, which are always related to certain forms of behavior and linked to forms of perception. Although it is also possible to define forbidden forms of action or to directly taboo certain areas related to sexuality in a robot, this cannot be linked to the directly lived experience of desire in contemporary robots.

Bataille also discusses the emergence of erotic perception in human prehistory, relating it to the burial of the dead, a cultural activity whose remains can be found in archaeological research. He thus considers the ancestors of man, in whom it is already possible to show this cultural behavior and therefore to infer the presence of other, albeit primitive, cultural patterns. He suggests that sexual taboos, which are conditioned by sexual perception, were not present in the ancestors of man but appeared at the emergence of the first humans.^[39]

It might be tempting, in this context, to regard robots as in some ways similar to human ancestors, but there are important differences: robots equipped with the current technology of large language models communicate at least as well as contemporary humans, whereas they do not possess the inner experience that Bataille sees as crucial to the human form of perception and understanding. We humans are aware of our own mortality, the finitude of our lives, and we want to surpass our limited existence. In this experience, we are shown the inevitable void.^[40] This dimension of human existence is apparently inaccessible even to the most advanced robots.

3 Human and Artificial Sensuality: Biological, Psychological, and Cultural Dimensions

In this section, we briefly outline aspects of sensuality in humans and, indirectly, in robots that are independent of linguistic expression. We first focus on the constituents of sensuality based on factors given by evolutionary psychology, and then on cultural influences that have also historically contributed to the perception of sensuality.

An important part of sensuality in humans is the shape of the human body. The importance of body shape, i.e., the ratio of individual parts to the whole, in relation to sensuality, is rooted in human evolution and reproductive signaling. The perception of certain significant bodily features has significant psychological effects on whether the body is perceived as sensual or attractive. The human body, as sensual, is understood as a source of aesthetic pleasure, sometimes in a romantic or erotic context. Sensuality can also be expressed in terms of formal characteristics of specific bodily features, such as waist-to-hip ratio (WHR), which also play a key role in assessing the sexual attractiveness of women to men. A figure that has a shape likened to that of an hourglass, which is characterized by a lower WHR, is perceived by male individuals in a statistically overwhelming number of cases as attractive.^[41] The rationale, from an evolutionary psychology perspective, is that this shape signifies higher levels of estrogen. Higher levels of this hormone are in turn related to fertility – such a body shape is advantageous for reproductive success. Other features, such as a prominent abdomen, may signal lower estrogen levels, reduced fertility, and other health risks, which may psychologically reduce attractiveness.^[42]

This recognition of physical sensory signals is active at a psychologically subconscious level in humans. It is related to a cognitive assessment of health, vitality, and mate potential. Perceptiveness with respect to these attributes is due to evolutionary pressure to select mates with optimal genetic potential.^[43] Of course, in the case of human–robot contacts, reproductive potential does not play a role, but unconscious psychological functions on the part of humans are still active. A robot that exhibits the sensory attributes perceived by humans as signs of health and vitality may be received more positively and acceptably. At the same time, social contact with such a robot may promote the human’s own mental health and naturally positive attitudes. A robot that exhibits sensory attributes associated by humans with illness, fatigue, deformity, etc. may elicit a negative or depressive disposition when socially interacting with a human, for example, in a caregiver role. This social transmission of depression is referred to as emotional contagion or social contagion.^[44]

From a cultural perspective, the perception and aesthetic evaluation of the shape of the human body in terms of sensuality is formed by locally dominant social norms and prevailing artistic representations. For example, a study of British literature from the sixteenth to eighteenth centuries shows that men consistently preferred the narrow-waisted female figure over the centuries.^[45] This research suggests that biology provides the basis for perceptions of sensual attractiveness, and culture reinforces and normalizes these preferences.

Different cultures may prefer different body shapes as sensual, although certain features are shared across cultures. The hourglass figure is a universally accepted archetype of the sensuality of the female body, whereas specific physical features such as height or weight vary across cultures or subcultures.^[46] The cultural body ideal is articulated and reinforced through art, media, and fashion, creating a universally accepted social standard of physical beauty and attractiveness. Cultural narratives thus combine and interact with biological predispositions to create standards in a given local society that determine what bodily aspects and proportions are accepted as sensual.

In general, the sensuality of certain shapes of the human body is the result of natural biological instincts as well as a socially constituted phenomenon. It stems, on the one hand, from evolutionary processes that constitute our cognitive makeup and, on the other hand, from cultural influences that further shape the human psyche depending on the nature of the society in which the individual grows up and lives.

We can now look in more detail at some of the more significant attempts to build robots capable of social, emotional, and sensual interaction with humans. In the recent history of robotics, there have been many attempts to build social or sensual robots – as opposed to purely purposeful robots, such as industrial robots. We first describe in detail a robot called Robosapien, and then briefly mention a number of other robots with similar characteristics.

Robosapien, developed in 2004 and designed by Mark W. Tilden, was a humanoid robotic toy that was one of the first major steps in biomorphic robotics.^[47] Biomorphic design emphasizes making robot movements resemble human behavior, which makes the robots thus created more credible and engaging. Robosapien had a fully articulated biomechanical design that allowed it to walk in a pendulum-like motion that is visually comparable to human walking. At the same time, its memory contained 67 pre-programmed actions, such as dancing, greeting, and suggesting karate moves, but none of them were particularly sensual.

For the subject of our text, the most significant characteristic of Robosapien was his vocal expression. Due to the time of his creation, he had not yet used large language models to create the content of his vocal delivery. However, his author concentrated mainly on the sensual-emotional aspect of his voice and called his specific speech “International Caveman Speech.” This voice was acoustically adjusted by the author to be easily recognizable and was characterized by thick tones. Robosapien also produced non-linguistic speech that included yawns, burps, and whistles. The voice was designed using humorous and exaggerated intonation to give Robosapien a playful personality. In addition, the robot’s individuality is emphasized by the catchphrase “Rosebud,” which is based on the classic movie “Citizen Kane,” which users intuitively associate with a nostalgic mood. From a technological point of view, Robosapien included other advanced features for its time, such as touch sensors and an infrared remote control.^[48]

Despite all the positive things that can be written about Robosapien’s vocal expression, it must be admitted that, from today’s perspective, the content of his vocal expression was very simple. It was programmed at a time when today’s large language models were not yet available, so it was based on a system of rules, and its communication was not comparable to human language. This, of course, limited the range of emotions it could respond to and express. Likewise, his perception of his surroundings as a sensory being was limited as a result.

In addition to Robosapien, several other similar robots have been created. The first group consists of robots that, like Robosapien, do not try to resemble humans in appearance very closely, and the emotionality of their physical expression and the sensuality of their bodies are thus limited to some extent by the constraints of their highly technical appearance. These robots include Pepper,^[49] Kismet,^[50] and Ameca.^[51] Another group consists of robots that try to resemble humans as much as possible. Among them, very well-known are Sophia^[52] and Erica.^[53] Other robots are more animal-like than human in appearance. An example is the robotic dog Aibo^[54] and penguin-like Lovot.^[55]

A specific group of sensual robots, which have a sophisticated human body and whose authors try to resemble human bodily sensuality as much as possible, are robots that are designed to be the social or intimate partner of their user. These are robots that use artificial intelligence to interact intimately and sensually with their user. Examples include RealDoll X.^[56]

As in the case of Robosapien, the ability of these robots to communicate in human language is severely limited to a few basic interactions programmed in the form of rules or simple algorithms. As a result, we can consider the following drawbacks as the main disadvantages of the above presented robots that use traditional rule-based technologies or simple algorithms to implement communication with humans: (a) communication with these robots is not intuitive and often not fully understandable, (b) communication is limited to patterns included in the rules and algorithms and thus not flexible enough, (c) communication with the robot is limited to the current situation and communication history is usually not included and taken into account, (d) robot knowledge is limited to pre-programmed content; even basic common knowledge is inaccessible to the robot, (e) robots are unable to respond to subtle nuances of language, (f) they are unable to personalize interaction with different users, treating all users according to the same patterns, and (g) emotional and sensual interaction at the verbal level is unconvincing and templated.

As we will show in the following, many of these drawbacks can be solved by using large language model technology. Nowadays, such improvements are already happening in some cases, and newer versions of the robots described above may have a module based on the principle of large language models, or a direct connection to one of the commercially available implementations of this architecture (Ameca).

The opposite of the aforementioned robots are the virtual social conversational chatbots being developed today. They possess a high quality of conversational skills, including sensitivity to emotional context and sensual context, but they lack any physical body, hence the absence of bodily sensuality.

A well-known example of such embedded technology is SIRI, a communication agent developed by Apple and built into mobile phones, which does not use large language model technology, but rather a combination of older technologies based primarily on rules. The company plans to integrate new technologies in 2026.^[57] The situation is similar in the case of Amazon’s Alexa software. The software was originally created exclusively on the basis of a combination of traditional algorithms, using mainly rules. At the beginning of 2025, a new version was introduced, which is supposed to use large language model technology.^[58] User experience with these simple software programs has clearly shown that older rule-based systems are insufficient for communication between humans and computer systems, and therefore were also insufficient for use in robots.

Some new chatbots have been developed with integrated large language model technology, and some even have a virtual body, which is, however, usually very limited in its possible expressions. An example of such a chatbot with a virtual body is Replika, an AI-equipped conversational app that aims to provide the user with a virtual friend. According to its creators, it uses a combination of large language models trained on a large dataset of conversations inducing positive sentiment to ensure that the resulting text output supports positive thoughts in users. In addition to large open-source language models, Replika also uses custom scripts to provide more security and a professional approach.^[59]

Although virtual social conversational chatbots have higher quality conversational skills, the absence of a physical body limits them in their sociality and sensuality. Their advantage is low cost compared to traditional bots. Thus, in the future, they will still exist alongside full-fledged social robots as a cheaper and more accessible alternative.

In the next section, we briefly introduce large language model technology, which holds the prospect of significant advances in conversational capabilities for the further development of social robots.

4 Large Language Models: A New Paradigm of Linguistic Competence

Currently, the most advanced approach to providing linguistic capability to an artificial digital system is through language models (LMs). In this section, we will first briefly describe their technical basis and then briefly discuss this technology in the context of a philosophical exploration of the possibilities of artificial intelligence.

The language models were developed using artificial neural network theory, which is inspired by biological neural networks. The basic example of a biological neural network is the human brain. The first small neural language model, on which most others are now based, was published by Bengio et al.^[60] Its refinement came from the application of recurrent networks and the method of long short-term memory architectures.^[61] In further research, these capabilities were extended by the application of convolutional networks^[62] and pre-trained word embeddings,^[63] and the introduction of the attention mechanism.^[64] The use of transformation networks to record and generate meaning from linguistic patterns in a wide range of textual contexts has also been beneficial.^[65]

Large Language Models (LLMs) refer to systems using the advanced techniques described above, along with the use of extremely large artificial neural networks with on the order of millions to billions of connections/parameters between neurons. Historically, the first large-scale language models included the Generative Pre-trained Transformer 1 (GPT-1) developed by OpenAI and BERT developed by Google in 2018.^[66]

From a more general philosophical perspective, large language models are an advanced form of artificial intelligence, aimed at processing and producing textual output of a quality equivalent to that produced by human authors. Unlike older linguistic systems, which were built on the principle of using grammatical and other rules, large language models are based on probabilistic approaches. Their starting point is the initial processing of extremely large sets of textual data, based on which probabilistic patterns are identified and encoded into the parameters of an artificial neural network. Subsequently, the model thus prepared is used to process the input text and to construct meaningful linguistic expressions. These linguistic expressions have a probabilistic basis and are not based on explicitly encoded grammatical rules or explicit instructions.

Thus, the basic method of large language models is to predict the probability of occurrence of a word or a more complex expression based on many preceding words, i.e., the textual context. This context can range up to several thousand words. The large artificial neural network used for this prediction typically has several billion parameters, which allows achieving very accurate results and producing meaningful word concatenations. Currently, the most popular applications of large language models are chat applications – such as ChatGPT designed by OpenAI, Llama produced by Meta, and Claude created by Anthropic.

Chat applications provide coherent, meaningful, and contextually relevant answers to a wide variety of questions, with high fluency and high relevance and content competence. Although the basis of how large language models work is probabilistic calculi, they synthesize text in new ways, restructuring the content that constituted the training data to a significant degree, so that the output text is not a mere repetition or reordering of the input data.^[67]

If we look at the big language models from a philosophical point of view, then we are led to revise traditional views about the nature of language and thought. These software artifacts produce sophisticated linguistic output that is, in terms of competence, comparable to that of a skilled human speaker. The output has a logical structure and can include advanced rhetorical figures of speech only mastered by professional human language users.

However, it is still necessary to keep in mind that their underlying mechanism is statistical in nature, not conceptual in nature, underpinned by an awareness of the meaning of the text produced. Large language models lack the subjective conscious phenomenal experience, semantic grounding, and fully experienced intentionality that are hallmarks of human thought and cognition. Therefore, large language models cannot be said to “understand” language in the sense that humans do. Some authors have therefore compared them to a stochastic parrot that merely repeats text without understanding its meaning.^[68] That is a bit of an exaggeration. It is true that large language models operate on a probabilistic basis – they do not understand the text contained in the training data, nor the text that these models produce, in the human sense. At the same time, however, their function is not one of mere repetition or reorganization, which is why other authors reject the comparison to a parrot.^[69]

5 The Future of Sensual Robots: Merging Form, Function, and Emotion

As mentioned above, sensual robots should meet certain physical characteristics, and we can also make demands on them with respect to their linguistic expression. In total, we can divide these requirements into four groups:

1. shapes and appearance of the body and face,

2. body movements and dynamic facial expressions,

3. tone of voice and its acoustic and phonetic properties, and

4. semantic content of communication and proper grammar.

In the following parts of this section, we briefly mention the first three topics, before focusing on the fourth requirement in the rest of the text, which is related to the introduction of large language model technology into robotics and is the main focus of this study.

6 How Large Language Models Revolutionize Robotic Communication

In the previous text, we noted that robots that use rules or traditional algorithms to generate language output often do not communicate in a way that is easily comprehensible to humans in the context of the real situation and the previous conversation.

In the case of natural language communication based on software powered by large language models, the situation is quite different. Robots equipped with large language model technology have significant advantages in their interactions with real humans. Through the use of this technology, robots so equipped have the ability to process natural language expressions significantly better, making interaction with them more intuitive and comprehensible for humans. Compared to rigid, pre-programmed conversational patterns, such equipped robots can understand complex questions and respond better with contextually embedded answers. The result is a much more human-like and friendly interaction and natural conversation, even when input from a human user is unpredictable.

At the same time, robots with large language model technology will manifest greater flexibility in communicating with humans. By not being restricted to a predefined set of responses, they instead generate more varied and relevant responses, sometimes with a creative character, which increases the quality and clarity of interactions. These robots can also adapt the vocabulary and emotional coloring of the language produced to the conversational situation, making them more suitable for environments such as nursing or healthcare, where empathetic communication is a necessity.

In addition, robots equipped with large language model technology can store and process information related to the context of previous communication over time and can therefore build on this conversational context in meaningful ways. As they remember details and context from previous communications, they follow up on those conversations and adjust their responses according to their details, allowing for a stronger emotional connection with human participants in the conversation. Such behavior is an advantage, for example, in educational or companion robots or virtual assistants, where maintaining conversational continuity enhances the user experience.

The advantage of integrating large language model technology into robots also leads to the ability to quickly process previously unthinkable amounts of information. Large language models are trained on huge text corpora involving large amounts of knowledge and information of various kinds. Robots equipped with this technology can access and summarize complex information, explain concepts intelligibly, and answer queries in real time, making them valuable tools for education or technical support. They can even serve as expert advisors to democratic representatives.^[81] They can also learn new information through updating their language models and adapt to evolving language or trends, ensuring they remain relevant and useful.

Large language model technology will allow robots to respond to subtle nuances of language, which can improve the robot’s ability to interact even in intimate or emotionally charged environments. Such enhanced robots will be able to process even less obvious conversational cues and respond sensitively, in accordance with the context of the situation. These robots will have a number of additional benefits, particularly in increasing the emotional connection and intimacy of the conversation, leading to communication with the robot more closely resembling that with a real person.

The question is whether robots equipped with large language model technology will be able to accommodate the levels of meaning that appear primarily in spoken language and rarely in the written texts from which they learn to communicate. This may be the case for vulgar speech, which contains many associations that are not common in written language, but a robot may encounter such speech. Lacan asks in this context, “Why is it the image of the vulva that surfaces to express a number of different acts, including those of escaping, of fleeing, of cutting and running (se tailler) […]?”^[82] Meaning frames and associations in the case of similar expressions are extremely broad and often permeate vulgar speech. Although such expressions may not be considered acceptable in ordinary society, they may still appear in spoken communication or may refer to them by association. Lacan says: “It should be noted that the use of a term that originally meant ‘coitus’ is capable of being extended virtually infinitely, that the use of a term that originally meant ‘vulva’ is capable of generating all sorts of metaphorical uses.”^[83] The question is whether robots will understand such metaphorical uses.

Robots with large language model technology will also be capable of more advanced communication personalization. The robot can store a communication history with each individual human and, upon re-recognition, can build on the previous communication or respond in a way that is appropriate not only to the context but also to the previously expressed wishes and needs of that particular user. This can create a fuller experience when communicating with a robot, similar to communicating with a real human. This adaptive learning capability refines the illusion of emotional connection and closeness, which can help, for example, users who suffer from loneliness and expect the robot to be a social complement or substitute for interactions with real people that they lack.

Another advantage of applying large language model technology to the development of social robots is the ability to generate communication outputs that realistically simulate emotional interactions. Using this technology, robots can engage in flirting, sensual dialogue, or romantic conversations that feel realistic and convincing to the human user. The ability to sensitively choose the emotional undertones of words enhances the sensuality of the conversation. Robots equipped in this way can simulate a tender, seductive, or playful tone of communication to create the illusion of emotional connection. Such communication more closely resembles communication with an emotionally intelligent partner than communication with a machine, which can be more engaging and satisfying for the user. Such an experience, when communicating with a robot, resonates emotionally with the user and can have a therapeutic effect in the case of care-oriented robots.

For example, Leach criticizes the concept of robots designed to emulate human femininity (gynoid robots) because they can function as sex robots, they can be programmed or instructed by the user to demonstrate through sensual behavior and speech sexual desire for the user’s gratification.^[84] Opposition to sex robots has a long history in feminist literature – as early as September 2015, there was a widely publicised Campaign Against Sex Robots (CASR),^[85] although some other writers have opposed this campaign.^[86] Leach uses the term gynoid to mean “an android that appears to be conforming to culturally contingent female gender stereotypes” and believes that this term is more accurate than the more commonly used “female android.”^[87] She also addresses the question of whether a nonhuman can have a gender; she claims that other artifacts such as razors, perfume, and toys are gendered because they were usually intended for only one gender. With respect to linguistic gender, this is not true for many languages; many words in them have gender regardless of what gender of the artifacts they were used by. There is also the problem that female versions of robots can come across as caricatures of femininity, and when acting in full accordance with cultural stereotypes attributed to women, such imitations can amount to parody.^[88]

The advantage of integrating large language model technology into robots will also be that the resulting interactions with robots are free of judgment and personal biases and prejudices that users may fear when interacting with humans. Conversations with robots thus provide a safe space in which users can share their hidden emotions, desires, or fantasies without having to worry about negative reactions or rejection. This can be particularly beneficial for individuals who struggle with social anxiety that prevents them from self-expression and open communication with real people. Even in this context, interaction with a robot can have a positive effect on the user’s psyche.

To summarize the advantages of robots equipped with large language model technology, we can briefly express them in the following list:

1. more intuitive and understandable interactions,

2. greater flexibility in communication,

3. the ability to build on past conversational context in meaningful ways in communication,

4. the ability to tap into vast amounts of knowledge, very quickly,

5. the ability to respond to the subtle nuances of language,

6. advanced personalization of communication,

7. realistic simulation of emotional and sensual interaction, and

8. interacting without judgement and rejection.

In summary, robots possessing large language model technology offer more intuitive and personalized interaction, building on previous communication history, and can flexibly adapt to changes in the conversational situation while understanding subtle nuances of language. In doing so, they are able to communicate emotionally, intimately, and sensually without judgment or rejection, achieving an emotional connection to the user that traditional technologies have not been able to. At the same time, these robots have a wealth of knowledge at their disposal at all times, which they are able to use very quickly in communication.

Large language model technology is thus the next step in bridging the gap between man and machine, allowing robots to communicate with humans in a human-like way. Of course, it is necessary that robots are also equipped with the other technologies mentioned above, i.e., the ability to move their bodies sensually, the ability to use facial expressions, and the ability to modify the tone and acoustic properties of the voice. These enhancements will make robots not only more efficient, but also more social and sensual, which is very important for their involvement in everyday life in human society.

7 Conclusion

Although human encounters with robots in everyday environments are currently rare, the frequency of such encounters will gradually increase. Therefore, robotics needs to be addressed not only by engineers but also by designers and experts in the field of human–computer interaction. It is not just a question of making robots technically advanced, but of making them closer to people and more emotionally responsive.

For example, in ageing societies where there is a shortage of human carers, robots could gradually supplement or replace their services. Robots could provide not only basic assistance, and medical advice, but also emotional comfort, and their presence could act in a reassuring way if their outward appearance were sophisticated. In this sense, the use of the robot’s external positive sensual elements is important, understood here not in a sexualized sense, but in the sense of providing positive emotional attunement to the human persons being cared for.

Robots can also care for children, for example, in therapeutic or educational contexts. Such a designated robot can be equipped with an engaging voice, expressive movements, and a friendly yet authoritative appearance, which will help to attract and maintain the attention of children. For example, Robosapien already fascinated children with its presence and attracted their attention even though it was not an educator but a robotic toy.

In the case of a robot for the mainstream population, it may be a friend or partner, and such a robot may take on a sensual form in a sexualized sense. The appearance and behavior of such a robot should provide emotional support, especially for lonely people. At the same time, given its knowledge, it could also help them in everyday life situations with advice or comments.

Generally speaking, the main target groups are vulnerable groups – children, the sick, or elderly, or lonely people. In their case, a robot company can be a solution to their situation, either temporarily or completely. In the long run, it is to be expected that interaction with robots and their integration into society will become more and more widespread, so efforts should also be made to accompany this process with the development of robots on an emotional and sensory level.

The current development of robotics is shifting from a focus on functional or industrial robots to social robots, which requires a major effort in combining technology and design aimed at simplifying human interaction. Newly designed robots are no longer just functional entities, but are also beginning to act as human companions, expressing emotion and sensuality, requiring new forms of philosophical reflection. In this article, we have attempted to map and define the concept of sensuality in both human and robotic contexts, with a view to integrating new, advanced technology of large language models into robots.

Large language model technology is revolutionizing the ability of robots to communicate with humans, with full simulation of emotional intelligence. Robots equipped with this technology can communicate with humans while maintaining context, communication history, yet with flexibility and personalized access to different individuals, bridging the gap between machine functionality and human emotional and sensory interaction. These capabilities can help robots establish deep emotional connections with users and help them meet their social needs. As a result, this technology pushes the boundaries of what is possible in human–robot interaction.

However, the development of new social robots not only brings its own benefits, but also raises a number of new ethical questions about the implications of robots that are able to mimic human emotions but do not actually experience them becoming a normal part of society, with some people interacting with them on a daily basis, or even robots becoming their intimate companions. Unfortunately, there was not enough space in this article to address these important issues. It is a task for future studies dedicated to ethical issues to address the development of these robots and formulate ethical principles to ensure that their operation is trustworthy and respects the basic moral principles of human society.^[89]

In conclusion, we have shown that robots enhanced by large language model technology represent a major advance in the possibilities of machine–human interaction, bringing a range of new possibilities and the promise of a future in which robots will not only help and serve humans in the practical areas, but will become part of society, partners capable of being responsive companions on an emotional and sensual level. Although many questions remain unresolved and there are many open challenges, the deployment of this new technology in robotics represents a major transformative step in the evolution of the relationship between humanity and artificial intelligence.