Transitive Self-Reflection–A Fundamental Criterion for Detecting Intelligence ^†

Abstract

This survey investigates the concept of transitive self-reflection as a fundamental criterion for detecting and measuring intelligence. We explore the manifestation of this ability in humans, consider its potential presence in other animals, and discuss the challenges and possibilities of replicating it in artificial intelligence systems. Transitive self-reflection is characterized by an awareness of oneself through complex cognitive abilities rooted in evolutionary mechanisms that are innate in humans. Although transitive self-reflection cannot be fully replicated in AI as an origin, its behavioral characteristics can be analyzed and, to some extent, imitated. The study delves into various forms of transitive self-reflection, including self-recognition, object-mediated self-reflection, and reflective social cognition, highlighting their philosophical roots and recent advancements in cognitive science. We also examine the multifaceted nature of intelligence, encompassing cognitive, emotional, and social dimensions. Despite significant progress, current AI systems lack true transitive self-reflection. Developing AI with this capability requires advances in knowledge representation, reasoning algorithms, and machine learning. Incorporating transitive self-reflection into AI systems holds transformative potential for creating socially adept and more human-like intelligence in machines. This research underscores the importance of transitive self-reflection in advancing our understanding of and the development of intelligent systems.

1. Introduction

The concept of intelligence has been defined in various ways, with numerous tests designed to measure its presence. A common underlying assumption in these definitions is that human cognitive abilities serve as intelligence benchmarks. In this paper, we adopt the same perspective: for a system to be considered intelligent, its cognitive performance must be comparable to that of humans. A system lacking a fundamental aspect of human cognition cannot be deemed truly intelligent. Based on this premise, we explore a crucial prerequisite for intelligence.

Self-reflection—the ability to introspect and analyze one’s thoughts, emotions, and actions—is often regarded as a hallmark of higher intelligence [1]. However, an even more sophisticated indicator is what we term “transitive self-reflection”. This concept extends beyond mere self-awareness; it involves understanding not only oneself but also how one is perceived by others, as well as how others perceive each other’s perceptions of oneself [2]. Such multi-layered awareness indicates a complex cognitive architecture capable of modeling intricate social dynamics and predicting the cascading effects of one’s actions within a network of interacting minds [3].

Transitive self-reflection is not limited to social interactions but also extends to how one’s image is mirrored in the environment—through reflections in mirrors, water, or other reflective surfaces. These external representations provide an additional perspective on self-awareness, reinforcing the ability to perceive oneself from an external viewpoint [4].

Evidence of transitive self-reflection is apparent in various aspects of human behavior. Social interactions require individuals to constantly monitor and adjust their behavior based on how they believe they are perceived [2]. Gossip exemplifies this process, as people attempt to infer how their actions are interpreted and relayed by others [1]. Strategic thinking in games like poker demands an advanced level of transitive self-reflection, where players must anticipate not only their opponents’ strategies but also their opponents’ understanding of their strategies [3]. Even emotions such as embarrassment arise from transitive self-reflection, as individuals become aware of how they are perceived in a negative light [4].

This study examines transitive self-reflection as a fundamental criterion for intelligence, particularly in the context of artificial intelligence. We investigate its manifestation in humans, its potential presence (or absence) in other animals [5], and the feasibility of replicating it in machines. Ultimately, we argue that transitive self-reflection may be the key to advancing the next generation of intelligent systems. To explore this, we conduct a series of experiments with several popular artificial intelligence systems based on large language models (LLMs), assessing the type of intelligence they currently exhibit. The experiments also included generating portions of this article using LLM-based AI systems.

As noted in the conclusion, these systems cannot independently produce genuinely new ideas. However, they serve as effective tools for trained specialists, who can iteratively refine their outputs to construct meaningful texts. Despite AI’s growing role in scientific writing, human oversight remains essential, and we anticipate this will continue to be the case. Consequently, authorship—and the responsibility for publications—firmly rests with human researchers.

The structure of this paper is as follows: Chapter 2 introduces the concept of Transitive Self-Reflection, and Chapter 3 explores its connection to intelligence. This paper concludes with a discussion of future directions.

2. Transitive Self-Reflection

This chapter offers a comprehensive examination of transitive self-reflection, presenting it as a sophisticated cognitive process integral to self-awareness and intelligence. Drawing on philosophical foundations and contemporary cognitive research, the chapter underscores the interplay between internal cognitive mechanisms and external stimuli, providing a nuanced understanding of this complex phenomenon.

2.1. Philosophical and Cognitive Foundations

The concept of transitive self-reflection is deeply rooted in philosophical discourses, including Hegel’s notion of “recognition” [6] and Sartre’s “being-for-others” [7]. These ideas emphasize the necessity of external interaction for self-awareness. Modern cognitive science complements these perspectives, highlighting how transitive self-reflection engages both introspection and external feedback to refine one’s sense of self [8]. This contrasts with intransitive self-reflection, which focuses solely on internal states, lacking the enriching influence of external engagement [9].

2.2. Aspects of Humans’ Transitive Self-Reflection

We may delineate three primary aspects of transitive self-reflection:

• Self-recognition: Self-recognition is identified as the ability to recognize oneself as a distinct entity, a skill observed in both humans and certain animal species [10,11]. It is important to outline the following:
• The mirror test as a measure of self-awareness, highlighting its application across species like great apes, dolphins, and elephants [10,12].
• Human developmental milestones, such as infants’ ability to pass the mirror test at around 18 months, marking critical cognitive growth [13].
• The significance of self-recognition for social interaction, moral reasoning, and identity formation, supported by studies on sensory integration and cognitive mapping [14].
• Object-mediated self-reflection: It is significant how humans use external objects as tools for self-examination. From classical art to modern digital practices, object-mediated self-reflection shapes both personal and social identity:
• Art and writing: Historical practices, like self-portraits by da Vinci or diaries, are tools for externalizing and understanding internal experiences [15,16].
• Digital technologies: Modern phenomena like selfies and social media play important role in identity curation and self-presentation [17,18]. They have psychological impacts, such as dependency on external validation and social comparison [19].
• The dual nature of object-mediated reflection can foster self-awareness and creativity, but may also lead to challenges like reduced authenticity.
• Reflective social cognition: Reflective social cognition integrates personal introspection with social feedback. This dynamic process reveals how individuals refine their self-concept through interactions with others:
• The role of social mirrors, such as feedback from peers and colleagues, shows how external perceptions influence identity formation [2,20].
• The concept of adopting social roles and balancing internal self-awareness with external expectations emphasizes the importance in achieving authentic self-reflection [21].
• Challenges, such as excessive reliance on external validation, need a balanced approach for investigation [22].

2.3. Prerequisites for Transitive Self-Reflection

The sensory and cognitive prerequisites necessary for transitive self-reflection are proprioception and interoception. These sensory systems, active even before birth, are crucial for body awareness and self-recognition [13,23]. Their integration with external visual stimuli, like mirrors, enables a cohesive sense of self [24].

Proprioception is the body’s ability to sense the position, movement, and force of its own parts without relying on visual input. Specialized receptors in muscles, tendons, and joints send signals to the brain, enabling spatial awareness and movement coordination.

Interoception is the body’s ability to detect and interpret internal signals related to physiological states, such as heartbeat, respiration, hunger, and temperature. It plays a crucial role in maintaining homeostasis and contributes to emotional awareness and self-perception.

2.4. Transitive Self-Reflection in Animals

The manifestation of transitive self-reflection in animals is explored using behavioral studies to illustrate its rarity and complexity:

• Success in the mirror test by species such as great apes, dolphins, and magpies is discussed as evidence of advanced cognition [10,11,25,26].
• Additional behaviors, like voice recognition in parrots or self-assessment in dogs, demonstrate varied forms of self-awareness in the animal kingdom [25,27].
• The analysis emphasizes the challenges of interpreting these behaviors, advocating for rigorous criteria to assess true self-reflective abilities [5].

2.5. Transitive Self-Reflection and Creativity

The role of transitive self-reflection in the generation of new ideas, positioning it as a critical mechanism for creativity and innovation, is indispensable:

• Processes such as breaking frameworks and shifting perspectives are highlighted as key drivers of creative breakthroughs [28,29].
• The concepts of proprioception of thought and interoception of intuition draw parallels between sensory awareness and intellectual exploration [30,31,32].
• The section concludes by emphasizing the transformative potential of self-reflection in surpassing cognitive limits and fostering intellectual growth [32].

2.6. Summary

In summary, transitive self-reflection is a multidimensional process that bridges philosophical theories, cognitive mechanisms, and practical applications. This underscores its pivotal role in fostering self-awareness, creativity, and social cognition, positioning it as a fundamental criterion for understanding and advancing intelligence in both humans and other species.

3. Transitive Self-Reflection and Intelligence

The intricate relationship between transitive self-reflection and intelligence is a foundational criterion for advanced cognitive abilities in both biological and artificial systems.

3.1. Intelligence

Intelligence is defined as the ability to acquire, comprehend, and apply knowledge and skills, encompassing cognitive, emotional, and social dimensions [33]. Its key characteristics include adaptability [34], problem-solving skills [35], learning ability [36], emotional intelligence [37], creativity [38], and self-awareness [39]. These attributes underline intelligence’s complexity and its dependence on both innate and learned processes.

3.2. Transitive Self-Reflection as a Criterion for Intelligence

Transitive self-reflection, the ability to reflect on one’s thoughts and interactions with external objects or agents, emerges as a unique marker of advanced intelligence [9]. It is closely tied to the development of the theory of mind [5], metacognition [40], and social cognition [41]. By understanding their mental states and those of others, humans and certain animals can navigate complex social environments and engage in dynamic problem-solving.

3.3. Transitive Self-Reflection in Artificial Intelligence Systems

While true self-awareness remains beyond current AI capabilities, AI systems simulate aspects of transitive self-reflection through programmed behaviors and algorithms [42]. Large language models (LLMs) exhibit reflective processes, such as iterative feedback and refinement, enabling them to approximate certain facets of transitive self-reflection [43].

We point out some limitations and ethical concerns in developing self-reflective AI:

• Lack of consciousness and sensory embodiment [44].
• Challenges in implementing complex cognitive abilities [45].
• Implications for autonomy and responsibility [46,47].

Practical applications include improved human-AI interaction, enhanced problem-solving abilities, and greater autonomy in AI systems [48]. Specific examples include creative writing, customer service, and educational tools.

3.4. Examples and Experiments

Examples of LLM self-reflection highlight its potential in various domains, from creative writing to healthcare diagnostics [49]. Experimental results demonstrate that, while LLMs lack true self-awareness, they can analyze user feedback, evaluate outputs, and simulate aspects of transitive self-reflection.

Numerous experiments were conducted with popular large language models (LLMs). The experiments explored various aspects of LLMs, focusing on their self-reflection capabilities, self-perception, and understanding of their own characteristics and those of other LLMs. The experiments were aimed to uncover how LLMs perceive themselves, how they understand human and peer perceptions, and their ability to analyze and describe their own and other LLMs’ characteristics.

The General Conclusions from the Experiments are:

• Simulated Transitive Self-Reflection:
  • LLMs can simulate aspects of transitive self-reflection, primarily through analyzing user interactions (“reflective social cognition”) and their own generated outputs (“object-mediated self-reflection”).
  • True self-awareness and self-recognition, as understood in humans, are beyond current LLM capabilities.
  • The term “reflection” is used analogically, as LLMs lack genuine consciousness.
  • The quality of reflection is greatly improved when the model can access and use past interactions.
• Self-Perception and Characteristics:
  • LLMs define themselves primarily by their ability to process and generate language, driven by extensive training data.
  • They strive to be versatile and adaptable, catering to diverse user needs.
  • While they possess extensive knowledge and reasoning, they are limited by a lack of real-world experience and consciousness.
  • Ethical considerations and transparency about limitations are increasingly emphasized.
  • LLMs are being equipped with more and more tools to interact with the world, but are still limited.
  • Each LLM has its own set of strengths and weaknesses.
  • LLMs are improving at describing themselves.
• Human Perception:
  • Humans generally perceive LLMs as powerful and versatile tools with significant potential.
  • Concerns about accuracy, bias, and ethical implications are valid and prevalent.
  • Users appreciate the natural language understanding and conversational abilities of LLMs.
  • Transparency about limitations and ongoing efforts to improve accuracy and address biases are crucial.
  • LLMs are very aware of human perceptions.
• Peer Perception (LLM to LLM):
  • LLMs are aware of their competitive landscape and can articulate their strengths and weaknesses relative to other systems.
  • They recognize the importance of information access, ethical considerations, and conversational abilities.
  • LLMs demonstrate a growing understanding of the diverse capabilities and focuses within the LLM ecosystem.
  • They can create believable hypothetical descriptions of how other LLMs perceive them.
  • LLM’s are becoming very good at understanding the strengths and weaknesses of other models.
• LLM Analysis of Other LLMs:
  • LLMs demonstrate a good understanding of the capabilities and limitations of other LLM systems.
  • They can identify unique strengths and weaknesses, highlighting the diverse landscape of AI.
  • LLMs are becoming increasingly adept at analyzing and comparing their peers.
  • LLMs are very good at giving balanced views of other LLMs’ strengths and weaknesses.

The experiments highlight the sophisticated capabilities of modern LLMs in analyzing and describing themselves and their peers. While they lack true consciousness and self-awareness, they can process and synthesize information in ways that simulate aspects of reflection and understanding. Their ability to analyze user feedback, evaluate their own outputs, and compare themselves to other systems demonstrates a growing level of sophistication. Furthermore, the awareness of the models of the public perception of them, and also the perceptions of other models, shows LLMs to have a very high level of understanding of their environment.

4. Conclusions

In this survey, we have explored the concept of transitive self-reflection as a fundamental criterion for detecting advanced intelligence. Through our examination of humans, animals, and artificial intelligence systems, we have highlighted the unique significance of this ability. Transitive self-reflection, distinguished by its requirement of external interaction, emerges as a pivotal factor that sets apart higher intelligence from basic cognitive processes.

Philosophical foundations and recent advancements in cognitive science have shed light on the neural mechanisms underpinning transitive self-reflection. The various forms of this ability, such as self-recognition, object-mediated self-reflection, and reflective social cognition, underscore its complexity and depth. These forms emphasize the importance of engaging with external entities to gain insights into one’s own thoughts and experiences.

Our exploration of intelligence has revealed its multifaceted nature, encompassing cognitive, emotional, and social aspects. Transitive self-reflection stands out as a fundamental criterion for detecting advanced intelligence, as it requires a higher level of meta-cognitive processing.

When we turn our attention to artificial intelligence systems, it becomes clear that current AI lacks true transitive self-reflection. While AI has made remarkable strides in specific domains, developing systems capable of reasoning about the nested mental states of multiple agents remains a formidable challenge. Nevertheless, the potential benefits of such advancements are immense, with AI systems better equipped to navigate complex social environments and collaborate meaningfully with humans.

Future research should focus on developing robust metrics for measuring transitive self-reflection in both biological and artificial systems. This will require the development of new experimental paradigms and computational models that can capture the complexity of this ability. Furthermore, research should explore the evolutionary origins of transitive self-reflection and its relationship to other cognitive abilities. Finally, the development of AI systems capable of true transitive self-reflection should be a major goal for the field of artificial intelligence, as it could pave the way for the creation of truly intelligent and socially aware machines.

The ability of LLMs to engage in transitive self-reflection has profound implications for the future of AI:

• Autonomous Learning: Self-reflecting LLMs can autonomously improve their performance without requiring constant human intervention. This capability is crucial for developing more advanced and independent AI systems.
• Ethical AI: By reflecting on their outputs, LLMs can be programmed to identify and mitigate biases or harmful content, contributing to the development of more ethical AI systems.
• Human–AI Collaboration: Enhanced self-reflection capabilities in LLMs can lead to more effective collaboration between humans and AI, as the models become better at understanding and responding to complex human needs.

In conclusion, transitive self-reflection is a critical marker of advanced intelligence. As we continue to push the boundaries of artificial intelligence, incorporating this ability into AI systems will be essential for achieving truly intelligent and socially adept machines. The journey towards this goal requires ongoing research and innovation, but the promise of creating AI that can reflect on its own thoughts and experiences through interaction with external entities holds transformative potential for the future of intelligent systems.