Meta and Mind: Tracing the Journey of Thinking about Thinking

Thinking about Thinking: A Timeless Fascination

For as long as we have written history, humans have been fascinated by the idea of thinking about thinking. The ancient Greeks saw self-reflection as a path to wisdom: Socrates urged his students to “know thyself”, while Aristotle suggested that the mind could even grasp its own activity. Centuries later, philosophers and logicians took this further, asking whether knowing something also means knowing that you know it. In the 1960s, Jaakko Hintikka captured this in a famous principle of logic: if an agent knows a fact, it should also know that it knows it. Fast forward to today, and this same idea has found new life in artificial intelligence, where researchers explore how machines might be designed not just to think, but to reflect on their own thinking.

But what exactly is this cognitive capacity, and how is it described in literature? Here we encounter considerable conceptual confusion. The same cognitive capacity has been labeled with a wide variety of terms. In some cases, different authors use different words to refer to essentially the same phenomenon, for example, introspection, reflection and consciousness (Flavell, 1979; Brown, 1987; Koriat, 2007). In other cases, the same term is used in conflicting ways, for instance, meta-reasoning has been cast both as a process and as an object (Russell & Wefald; Ackerman & Thompson, 2017). By this we mean that some authors describe meta-reasoning as the activity of monitoring and regulating one’s reasoning in practice, while others treat it as an object of study to be modeled or formalised. The literature also encompasses a cluster of related notions, including metaknowledge, metamemory, self-adaptation and self-awareness (Davis & Buchanan, 1977; Nelson & Narens, 1990; Dunlosky & Metcalfe, 2009; Fleming & Frith, 2014). Both considered, this terminological diversity reflects both the richness and the ambiguity of research on cognitive capacity, and it poses a challenge for anyone seeking conceptual clarity.

Metacognition as the Dominant Term

Despite the diversity of terms used to describe this higher-order capacity, the prevailing usage in literature has been to designate it as metacognition. Since Flavell’s seminal work in the 1970s (Flavell, 1976; 1979), the concept has been widely adopted in psychology and cognitive science to denote the monitoring and regulation of cognitive processes. Nelson and Narens (1990) advanced this understanding by introducing a formal two-level model, distinguishing between monitoring and control functions. Brown (1987) further asserted the multifaceted nature of metacognition, encompassing processes of planning, evaluation and regulation. More recently, comprehensive reviews by Dunlosky and Metcalfe (2009) and by Fleming and Frith (2014) have consolidated this perspective, reinforcing the status of metacognition as the umbrella term for research on self-monitoring and self-regulation in cognition.

Etymologically, the component meta- is derived from the Greek preposition meta- (μετά), meaning “beyond” or “about”, while the root cognoscere comes from Latin, meaning “to get to know” or “to recognise”. In philosophical and scientific discourse, the prefix meta- has long been used to denote a higher-order stance: metaphysics, for instance, refers to inquiry “beyond physics”, while meta-language refers to a language that can describe another language (see Chalmers, 2002; van Benthem, 2007). In psychology and cognitive science, cognition is typically defined as the set of processes by which information is acquired, processed, stored and used, including perception, memory, learning and reasoning (Neisser, 1967; Anderson, 2010). When combined, meta and cognition literally denote “cognition about cognition”. Flavell (1976, 1979) introduced this term to describe knowledge and regulation of one’s own cognitive processes, an understanding that has since become the dominant definition across psychology and cognitive science. In this sense, metacognition refers to the reflective capacity by which individuals monitor and control their own cognitive states and operations (Nelson & Narens, 1990; Dunlosky & Metcalfe, 2009).

The Rise of Metacognition in Psychology

After Flavell’s pioneering work, the concept of metacognition rapidly gained traction in cognitive science and psychology during the 1980s. Researchers increasingly converged on the view that metacognition should be understood not as a single process, but as a broad umbrella concept encompassing a wide range of higher-order cognitive skills, including memory, learning, reasoning and problem solving. (Flavell, 1979; Brown, 1987; Veenman, Van Hout-Wolters, & Afflerbach, 2006). By the mid-1980s, the literature on metacognition had expanded to such an extent that it was famously described as a “many-headed monster of obscure parentage” (Brown, 1987, p. 106), reflecting the field’s proliferation of overlapping definitions and constructs. Brown and her colleagues highlighted that metacognition includes both knowledge about cognition and the regulation of cognitive activity, linking it closely to the emerging construct of executive control (Brown, 1987; Weinert & Kluwe, 1987).

To bring theoretical coherence to the field, Thomas Nelson and Louis Narens (1990) proposed one of the most influential frameworks for understanding metacognition. They formalised the idea that cognitive processes operate on two hierarchically related levels: an object level, which carries out the primary cognitive activities such as reading, problem-solving or remembering, and a meta level, which monitors and regulates the object-level processes. In this model, the meta-level maintains a dynamic representation of the object-level’s state and it can issue control signals that alter the course of ongoing cognitive activity. For instance, while reading a difficult text (object level), a person might periodically ask themselves “Do I understand this?” (monitoring). If the answer is no, they might slow down, re-read or adopt a new strategy (control). Nelson and Narens’ framework thus identified two core components of metacognition: monitoring (or the evaluation of cognitive states and control) or the regulation of those states in light of monitoring outcomes. This model has been widely cited because it provides a unifying language and structure for otherwise disparate metacognitive phenomena across domains such as memory (Koriat, 2007), perception (Fleming et al., 2012), decision-making (Ackerman & Thompson, 2017), learning (Dunlosky & Metcalfe, 2009) and reasoning (Cox & Raja, 2011). By clearly distinguishing between the object- and meta-levels, Nelson and Narens offered a theoretical foundation that continues to shape contemporary debates about the nature of self-monitoring and self-regulation. At this stage, it becomes clear that metacognition should not be treated merely as a loose label for a cluster of related notions, but rather as a specific function: the capacity to monitor cognitive processes and to exert regulatory control over them.

Entering the World of AI

In parallel with developments in psychology, the concept of metacognition also entered the field of artificial intelligence (AI) in the 1970s. Early expert systems demonstrated the need for a meta-level of knowledge that could oversee and refine the system’s reasoning. A notable case was the TEIRESIAS system, an extension of the medical expert system MYCIN, in which Davis and Buchanan (1977) introduced the notion of meta-level knowledge—defined as “knowledge about knowledge”. This allowed the system to “know what it knows” and to reason about its own knowledge base. By incorporating meta-rules, TEIRESIAS could evaluate and revise its reasoning rules, marking one of the first concrete implementations of self-reflective capabilities in AI (Davis & Buchanan, 1977).

During the 1980s, the idea of meta-level reasoning gained momentum within AI and computational cognitive science. Cognitive architectures such as Soar (Newell, 1990) integrated mechanisms for systems to reflect on and improve their own problem-solving through strategies like chunking and introspection. A decisive step was taken by Russell and Wefald (1991), who introduced a decision-theoretic account of how an intelligent agent should decide when and how to think about its own thinking under resource constraints. Their framework formalised meta-level acts—such as choosing which subproblem to prioritise or whether to allocate more time to computation—as rational decisions in an “outer loop” that monitors and regulates the “inner loop” of object-level reasoning. This work made explicit the need to study not only reasoning itself but also reasoning about reasoning—a perspective that soon came to be described as meta-reasoning.

Meta-reasoning Today

By the 2000s, a subfield explicitly devoted to these issues had emerged, with researchers employing a variety of terms such as computational metacognition, meta-reasoning, meta-cognitive reasoning and just metacognition. Scholars such as Cox (2005) and Cox and Raja (2011) emphasised the importance of self-monitoring and self-regulation in intelligent systems, explicitly borrowing the language of metacognition from psychology but applying it directly to reasoning processes. Architectures like MIDCA (Metacognitive Integrated Dual-Cycle Architecture) embodied this integration, with one cycle for object-level cognition and another for meta-level monitoring and control (Cox, 2005). These systems could detect failures in their reasoning, identify new goals and revise strategies accordingly, bringing a structured meta-layer into computational design.

Today, the AI community makes use of this family of terms, but meta-reasoning has become especially prominent in work that focuses directly on reasoning as the central cognitive process to be monitored and regulated (Russell & Wefald, 1991; Cox, 2005; Zilberstein, 2008; Cox & Raja, 2011; Ackerman & Thompson, 2017). Applications extend to machine learning, where algorithms evaluate their own uncertainty and adapt strategies (Graves et al., 2016; Lakshminarayanan, Pritzel, & Blundell, 2017), and robotics, where agents assess their performance in dynamic environments (Winfield, 2018), and more recently, large language models (LLMs). In LLMs, meta-reasoning has been invoked to explain and improve mechanisms such as Chain-of-Thought prompting, Tree-of-Thought reasoning and self-reflection methods, which aim to enhance models’ abilities to monitor their outputs, detect errors and refine their reasoning strategies. The coexistence of different terminologies, meta-reasoning and computational metacognition reflects the conceptual diversity of the field. Nevertheless, when the focus is specifically on reasoning processes, the use of meta-reasoning highlights the function of monitoring and regulating inference rather than cognition in general.

From Socratic calls to “know thyself” to modern AI systems that can evaluate and revise their own reasoning, the idea of “thinking about thinking” has traced a remarkable journey. What began as a philosophical reflection on self-knowledge evolved into psychological theory of metacognition, and later further evolved into computational frameworks for meta-reasoning. Today, as researchers design machines that monitor, regulate and even justify their own inferences, we are reminded that this age-old capacity is not only central to human intelligence but also to the future of artificial intelligence. Tracing this trajectory—across philosophy, psychology and AI—shows that the quest to understand and implement meta-level capacities is ultimately a quest to understand both mind and machine.

References:

Ackerman, R., & Thompson, V. A. (2017). Meta-reasoning: Monitoring and control of thinking and reasoning. Trends in Cognitive Sciences, 21(8), 607–617.

Anderson, J. R. (2010). Cognitive Psychology and Its Implications (7th ed.). Worth.

Aristotle. De Anima. Trans. and commentary in Shields, C. (2016). Aristotle: De Anima. Clarendon.

Baker, L., & Brown, A. L. (1984). Metacognitive skills and reading. In P. D. Pearson (Ed.), Handbook of Reading Research (pp. 353–394). Longman.

Benson, H. H. (2000). Socratic Wisdom: The Model of Knowledge in Plato’s Early Dialogues. Oxford.

Brown, A. L. (1987). Metacognition, executive control, self-regulation, and other more mysterious mechanisms. In F. E. Weinert & R. H. Kluwe (Eds.), Metacognition, Motivation, and Understanding (pp. 65–116). Lawrence Erlbaum Associates.

Chalmers, D. (2002). “On Sense and Intension.” Philosophy and Phenomenological Research.

Cox, M. T., & Raja, A. (2011). Metareasoning: Thinking about Thinking. MIT Press.

Dunlosky, J., & Metcalfe, J. (2009). Metacognition. Sage Publications.

Flavell, J. H. (1976). Metacognitive aspects of problem solving. In L. B. Resnick (Ed.), The Nature of Intelligence (pp. 231–236). Lawrence Erlbaum Associates.

Flavell, J. H. (1976, 1979). Foundational papers introducing metacognition.

Flavell, J. H. (1979). Metacognition and cognitive monitoring: A new area of cognitive–developmental inquiry. American Psychologist, 34(10), 906–911.

Fleming, S. M., & Frith, C. D. (Eds.). (2014). The Cognitive Neuroscience of Metacognition. Springer.

Hintikka, J. (1962). Knowledge and Belief. Cornell University Press.

Koriat, A. (2007). Metacognition and consciousness. In P. D. Zelazo, M. Moscovitch, & E. Thompson (Eds.), The Cambridge Handbook of Consciousness (pp. 289–325). Cambridge University Press.

Neisser, U. (1967). Cognitive Psychology. Appleton-Century-Crofts.

Nelson, T. O., & Narens, L. (1990). Metamemory: A theoretical framework and new findings. The Psychology of Learning and Motivation, 26, 125–173.

Online Etymology Dictionary. (n.d.-a). Meta-. In D. Harper (Ed.), Online Etymology Dictionary. Retrieved September 10, 2025, from https://www.etymonline.com/word/meta-.

Online Etymology Dictionary. (n.d.-b). Cognition. In D. Harper (Ed.), Online Etymology Dictionary. Retrieved September 10, 2025, from https://www.etymonline.com/word/cognition.

Plato. Apology. In Cooper, J. M. (Ed.), Plato: Complete Works. Hackett, 1997.

Pressley, M., & Ghatala, E. S. (1990). Self-regulated learning: Monitoring learning from text. Educational Psychologist, 25(1), 19–33.

Russell, S., & Wefald, E. (1991). Do the Right Thing: Studies in Limited Rationality. MIT Press.

Schneider, W. (2008). The development of metacognitive knowledge in children and adolescents: Major trends and implications for education. Mind, Brain, and Education, 2(3), 114–121.

van Benthem, J. (2007). “Logic in Games.” (discussion of meta-language & meta-levels).

Veenman, M. V. J., Van Hout-Wolters, B., & Afflerbach, P. (2006). Metacognition and learning: Conceptual and methodological considerations. Metacognition and Learning, 1(1), 3–14.

Weinert, F. E., & Kluwe, R. H. (1987). Metacognition, Motivation, and Understanding. Erlbaum.

Williamson, T. (2000). Knowledge and Its Limits. Oxford University Press.

Further reading/watching/listening:

Books & Articles:

Cox, M. T., & Raja, A. (Eds.). (2011). Metareasoning: Thinking about thinking. MIT Press.

Dunlosky, J., & Metcalfe, J. (2009). Metacognition. SAGE Publications.

Flavell, J. H. (1979). Metacognition and cognitive monitoring: A new area of cognitive–developmental inquiry. American Psychologist, 34(10), 906–911.

Videos & Podcasts:

Metacognition in LLMs: Can AI Think About Thinking? – Shun Yoshizawa & Ken Mogi.