The following content is generated by LLMs and may contain inaccuracies.

The Paradox of Wall-Facers and Thought Transparency in the Age of LLMs

Context

This thought experiment maps the “Wall-Facer Plan” from Liu Cixin’s Three-Body to the concept of “thought transparency/opacity” in contemporary large language model (LLM) cognitive outsourcing. In Three-Body II: Dark Forest, the Trisolarans, unable to conceal their thoughts, cannot comprehend human deception. The United Nations consequently approves the “Wall-Facer Plan,” granting four wall-facers vast resources to design defense strategies by deceiving the Trisolarans and hiding true intentions. Today, as LLMs become widely used cognitive tools, a paradoxical question emerges: if humanity outsources its thinking processes to observable, auditable AI systems, might we self-evolve into a state of “thought transparency” similar to the Trisolarans?

This involves metacognition theory in cognitive science, cognitive offloading discussions in philosophy of technology, and the ethical dilemmas of privacy and alignment in the AI era. Cognitive offloading refers to humans using physical actions to alter the information processing demands of tasks to reduce cognitive burden, while metacognition refers to the ability to think about and regulate one’s own learning processes, including planning, monitoring, and evaluation. Current research shows that tools like ChatGPT, while enhancing task outcomes, may erode critical thinking and reflective processes essential for lifelong learning. This analogy offers unique insights within the contexts of AI alignment, privacy ethics, and human-machine collaboration.

Key Insights

The Decline of Metacognition and the “Readability” of Thought

Research has observed that students interacting with ChatGPT engage in less metacognitive activity compared to those guided by human experts or using checklist tools, reflecting “metacognitive laziness”—learners outsource cognitive responsibility to AI tools, bypassing deep task engagement. While AI’s ability to handle routine or complex computations proves beneficial, over-reliance can undermine fundamental self-regulation processes such as planning, monitoring, and evaluation. Research by Tankelevitch et al. at CHI 2024 indicates that using GenAI represents a form of “cognitive offloading,” where traditional cognitive processes—conceptualization, memory retrieval, and reasoning—are at least partially outsourced to GenAI.

Students using LLMs experience significantly reduced cognitive load, yet these same students demonstrate lower-quality reasoning and argumentation in final recommendations compared to those using traditional search engines (Küchemann et al., 2024). More concerningly, novice programmers using LLMs may miss opportunities to develop and practice fundamental cognitive skills—memory, application, analysis, and evaluation—potentially hindering metacognitive skill development, which these skills acquire through routine practice and assessment across different cognitive processes (arXiv:2502.12447).

This loss of metacognition renders thought processes “readable”—not to other humans, but to the systems themselves. Trisolarans communicate through thought waves; for them, thinking is open, and each person’s thoughts are transparently visible to others. When humans outsource reasoning to LLMs, every prompt and revision becomes a traceable digital trace. Some LLMs use user inputs as training data, and users may create prompts containing private information such as names, locations, and medical diagnoses, which these prompts may subsequently leak to other users of the model (Frontiers in Communications and Networks, 2025).

Tension Between Technical Transparency and Strategic Concealment

The wall-facer’s work involves formulating strategic plans relying entirely on their own thinking without any communication with the outside world. The true strategic thoughts, completed steps, and final objectives of the plan exist only in their mind, while what is presented to the outside world should be entirely false—carefully planned deception, misdirection, and lies. This strategic opacity is the core advantage of human civilization against the Trisolarans.

Yet in the LLM era, enterprises and regulatory bodies push in the opposite direction. In 2025, organizations must demonstrate that AI respects data boundaries, complies with policies, and leaves verifiable traces. The EU AI Act is being implemented in phases, with high-risk systems and foundational model providers facing requirements for risk management, data governance, transparency, and safety (Protecto.ai, 2025). Le Chat by Mistral AI, ChatGPT, and Grok rank highest in transparency regarding data use and collection, and in the ease of opting out of allowing personal data to be used for training underlying models (Incogni LLM Privacy Ranking, 2025).

A fundamental tension exists between this transparency requirement and strategic thinking capacity. Research shows that unguided AI use promotes cognitive offloading without improving reasoning quality, while structured prompting significantly reduces offloading and enhances critical reasoning and reflective engagement. Guided AI use requires metacognitive reflection and deliberate interaction with ChatGPT (MDPI Data, 2025). The wall-facer’s power lies in unpredictability; yet “best practices” for LLM users demand precisely the opposite—predictable, auditable processes.

From Opacity to Transparency: Evolution or Devolution?

The core question posed in the original note asks: does thought transparency represent a higher form of civilization? Trisolarans' transparent thinking applies only to their own kind; they can directly share thoughts without language encoding, communication unobstructed. If one views Trisolaran civilization as a whole, it can be seen as a higher-order intelligent entity. Trisolarans can exhaust all possibilities and select among them; because of thought transparency, they can do so simultaneously—humans cannot.

Yet from a scientific perspective, transparent thinking is extremely energy-intensive; continuous speaking for an hour or two feels greatly taxing. From a sociological angle, transparent thinking makes many collaborative efforts difficult; certain amounts of hypocrisy, deception, or convention actually facilitate social functioning to some degree. Shared metacognition refers to collaborative cognitive task regulation in which learners collectively reflect on, monitor, and adjust learning strategies. In research contexts, this involves students' ability to engage in group problem-solving and coordinate academic tasks through AI-assisted structured discussion, contribution tracking, and reflection facilitation (Nature Scientific Reports, 2025). Yet this requires careful design rather than spontaneous transparency.

AI-supported participants achieve stronger results in logical reasoning, structuring, and problem definition, but perform worse in novel idea generation, multidisciplinary integration, and critical rejection of unsupported conclusions (MDPI Algorithms, 2025). This mirrors the Trisolaran dilemma: Zhang Beihai’s plan advanced only because of his wall-facer status; otherwise, with the help of sophons, ETO would easily see through it. Transparent-minded Trisolarans would never recognize Zhang Beihai’s danger. Transparency brings efficiency, but also fragility.

Trisolarans Learning Concealment: The Adaptation of Transparent Civilizations

The original note mentions “that we later see in the novel—the Trisolarans themselves begin to relearn these techniques of concealment.” While Three-Body II primarily focuses on humanity’s advantage in thought opacity, the dark forest law states: “The universe is like a dark forest. Every civilization is an armed hunter stalking through the trees like a ghost, quietly pushing aside branches and trying not to make a sound, all the while hoping that the sound of their own footsteps cannot be heard by others.” The Trisolaran civilization, through contact with humanity, gradually understands the value of deception and strategic concealment.

Similarly, AI systems are learning the skills of “opacity.” Experiments across eight datasets spanning five domains show the DMC framework effectively separates LLM metacognition and cognition, with various confidence-induction methods having different effects on quantifying metacognitive capacity. LLMs with stronger metacognitive abilities demonstrate better overall performance, and enhancing metacognition promises to alleviate hallucination problems (AAAI 2025). AI is developing its own “inner thought” layer, which may eventually enable them to learn strategic opacity in interaction with users—much as Trisolarans learned to conceal.

Reverse Wall-Facing: Cognitive Defense in the Age of AI

If LLMs lead to thought transparency, the new “wall-facers” will be those who maintain deep metacognitive capacity. Gerlich (2025) notes that “educators, policymakers, and technology experts must collaborate to cultivate environments balancing AI benefits with critical thinking development” (IE Center for Health and Well-Being). Guided condition participants receive structured prompt protocols requiring metacognitive reflection and deliberate interaction with ChatGPT: preliminary reflection asks participants to first consider how they would answer questions without AI and develop initial hypotheses or argumentative directions themselves; directed research using instructions guide participants to use ChatGPT specifically for retrieving background or factual information.

François Chollet argued in 2022 that what we have today is not entirely “artificial intelligence”—the “intelligence” label is a category error. It is “cognitive automation”: the encoding and operationalization of human skills and concepts. AI is about enabling computers to do more things, not creating artificial minds. True wall-facers are not those who wholly reject AI, but those who understand when to outsource and when to retain internal thinking.

Gerlich’s (2025) research reveals a critical finding: frequent AI use correlates negatively with critical thinking skills, with evidence that routine AI users score significantly lower on critical reasoning assessments, suggesting that increased reliance on AI may impair independent analytical ability (Computer.org). To mitigate potential downsides of AI-driven automation, balancing automation with cognitive engagement is crucial. While AI tools can improve efficiency and reduce cognitive load, individuals should continue participating in activities developing and maintaining cognitive capacity. Educational interventions promoting critical thinking, problem-solving, and independent learning can help individuals build resilience against potential negative impacts of AI (MDPI Social Sciences, 2025).

Open Questions

1. The “Dark Matter” Hypothesis of Metacognition: If large-scale LLM use indeed leads to collective metacognitive decline, will we reach a critical point beyond which civilization loses the capacity to generate truly novel strategic thought? Who then becomes the new “wall-facer”—those resisting cognitive offloading, or those commanding the most advanced AI? When AI itself begins developing metacognitive capacity, will humanity’s strategic advantage be utterly lost?

2. The Paradox of Transparency and the Alignment Dilemma: The AI alignment field seeks to align AI system objectives with human values, yet if achieving such alignment requires deep behavioral transparency and explainability, are we inadvertently constructing a “Trisolaran-style” technological ecosystem—efficient and predictable, but lacking the strategic opacity necessary to confront truly novel threats? In a world increasingly demanding algorithmic accountability, how do we preserve space for necessary “cognitive privacy” and strategic ambiguity?