How Consciousness Might Emerge From Thinking About Thinking

Source: R. Lewis - generated using Gemini AI / Google

This is Part 3 of a five-part series on cognition and consciousness. Part 1 explored what we mean by "cognition." Part 2 examined neuroscientist Joseph LeDoux’s account of mental models. Here we turn to LeDoux’s theory of consciousness.¹

When Models Become Conscious

What makes the difference between merely processing information and actually being conscious of it? The brain doesn’t just record perceptions and memories; it builds models of the world and of ourselves, most of which remain unconscious. Conscious cognition² is engaged when habits fall short, and flexible context-sensitive control is required.

Higher-order theories propose that consciousness arises when the brain generates higher-level representations of first-order representations (which are perceptions and memories), in effect, thinking about its own thinking. LeDoux builds additional important elements into this theory of meta-representations, or what he also calls redescriptions. He argues that prefrontal circuits re-describe entire models rather than just re-describing single representations (as explained in Part 2, models are assemblies of representations that perform simulations³). These "redescriptions" are still pre-conscious, according to LeDoux. They need to undergo additional steps to transform them into conscious experience: They are stabilized in working memory, enriched with memory content, and organized narratively into a self-related model. We will explore each of these processes here.

How Memory Types Map Onto Consciousness

Psychologist Endel Tulving had argued that different memory systems correspond to different forms of awareness. In Tulving's framework:

• Procedural memory (skills and habits) underlies anoetic consciousness—a basic, non-reflective, intuitive awareness of performing actions without explicit knowledge of past events.
• Semantic memory (factual and conceptual knowledge, and categorization) supports noetic consciousness—knowing and recognizing facts (but not personalized to time or place).
• Episodic memory (recollections of personal events) makes possible autonoetic consciousness—self-reflective awareness and the ability to simulate one's past experiences or potential future scenarios (mental time travel).

LeDoux, with Hakwan Lau, builds on this foundation but shifts its emphasis. For Tulving, each memory type corresponded to a distinct kind of consciousness. LeDoux and Lau agree but go further: Memory’s role is not just classificatory but mechanistic. It provides the material that higher-order prefrontal circuits redescribe and organize into narratives, thereby playing a role in generating conscious experience.

They add that anoetic consciousness infuses a tacit sense of familiarity and ownership to experience. "The fringe or penumbral quality of these goes unnoticed because anoetic states are often embedded in noetic and/or autonoetic states with actual conscious content."⁴

Drawing on Tulving’s categories but stressing that such forms of awareness can only be inferred from brain anatomy rather than directly demonstrated, LeDoux sketches a tentative evolutionary distribution: All mammals are likely capable of a basic form of anoetic consciousness and some degree of noetic consciousness (but this is probably limited, in non-primates); primates, with their enlarged prefrontal cortex, show richer noetic capacities; and humans—with a uniquely developed lateral frontal pole—are likely the only species with autonoetic consciousness, though he allows it may also exist in other great apes.^5,6

Working Memory as the Workspace

Meta-representations are generated when prefrontal circuits take entire models—those assemblies of perceptions and memories—and represent them at a higher, more reflective level. These pre-conscious redescriptions, however, would quickly fade if not stabilized. This is where working memory steps in. Working memory is the mental workspace controlled by the prefrontal cortex (and it can operate non-consciously as well as consciously). It holds meta-representations active, binding them with other relevant material such as ongoing perceptual inputs, retrieved memories, and self-related schemas. Without this workspace, meta-representations would remain fleeting and never acquire the coherence and narrative form that characterize awareness.

That narrative organization additionally depends on memory schemas.

Schemas and Narrative

Memory schemas give conscious content its shape. These schemas are bundles of conceptually interrelated memories (also integrating perception and bodily states) that combine into larger pre-conscious models of self and the world. When higher-order prefrontal systems redescribe these models, organize them narratively, and stabilize them in working memory, they can be experienced consciously. With Lau, LeDoux argues that narration is not just a retrospective portrayal but part of what constructs consciousness itself—a view sharpened by his earlier split-brain research, where one hemisphere routinely invented stories to explain actions initiated by the other. He suggests that this storytelling function is something our intact brains are constantly engaged in.

Source: R. Lewis, adapted from LeDoux

Emotion and Consciousness

Conscious emotions arise through the same redescription process as other states, but with two key additions: emotion schemas (knowledge of what fear, anger, or joy involves) and self-schemas (autobiographical memories that situate the event in relation to your own history). Consider encountering a snake: Your brain automatically builds a threat-detection model, but conscious fear emerges only when prefrontal circuits redescribe this model with personal context—"I'm afraid because snakes remind me of that time I nearly stepped on a venomous snake as a child"—and stabilize this narrative in working memory.

Symbolic Representation, With the Self as the Central Character

Consciousness feels like a storyline with ourselves at the center. LeDoux explains this as the brain redescribing its models in "mentalese"—philosopher Jerry Fodor's term for an internal symbolic code distinct from spoken language. This multimodal system integrates perceptions, memories, and bodily signals before we ever put thoughts into words.⁷ Actual language, for LeDoux, is a public code that piggybacks on mentalese, allowing us to express and share ideas.⁸

Cognition Essential Reads

AI and the Problem of Polished Prose

How to Throw the Right Kind of Party

By redescribing in mentalese, the brain weaves experience into a narrative model of the self. These narratives draw not only on perceptual and memory schemas but also on interoceptive signals from internal organs, the body’s homeostatic needs for stability and balance, and the value of current goals. Together, these elements ground awareness in lived experience and give conscious life its affective tone. This embodied narrative model of the self is most fully realized in autonoetic consciousness, which situates the self across time in a coherent story.

Consciousness Is a Trade-off: Costly, but Adaptive

According to LeDoux, the brain economizes by relying on habits and automatic responses, reserving the more energetically demanding conscious cognition for situations requiring flexibility. In those moments, higher-level redescriptions bind facts, feelings, and personal history into a self-narrative that enables reflective new options.

One Version of a Larger Debate

For LeDoux, whether in split-brain confabulations or daily life, consciousness is inseparable from the brain’s storytelling—the redescription of models into a narrative self. This recursive step—representations of representations—is the key mechanism that transforms mere cognition into conscious experience.

LeDoux’s account is one of many theories of consciousness,⁹ and like the others, it does not (nor claims to) completely solve the “hard problem” of why conscious experience feels like something (in fact, LeDoux considers the hard problem to be more of a philosophical question than a real scientific problem). It also rests on a strict definition of cognition. As noted in Part 2, LeDoux limits cognition to systems that can build mental models in a particular way—a criterion that applies to mammals and perhaps some birds, but not to many other animals that other scientists might regard as cognitive, and possibly conscious. And, in his view, the meta-representations needed for autonoetic consciousness may be uniquely human, or at most found in a few other large-brained species.¹⁰ His definition also excludes artificial intelligences (AIs), at least in their present design, because they lack the embodied, memory-integrated modeling he sees as central.

In Part 4, we will weigh the merits of a broader definition of cognition—one that centers around learning. And in Part 5, we will ask what these debates mean for AI, both in terms of cognition and the possibility of consciousness.