An Emerald Door Is Never Just A Door, but AI Should Remain Just A Tool

An unidentified library, featuring a biography of H.G. Wells. Photo courtesy of openverse.

Before sitting down to write this article, I read H.G. Wells’s short story “The Door In The Wall,” which details a man’s interactions with a mysterious door, dating back to his childhood. The short story is told by a narrator who describes his friend, Wallace’s, growing obsession with the door that slowly erodes his highly ambitious life as a rising politician. The door appears to him only in moments of serious stress or despair, taunting him from afar with memories of childhood bliss. The ending is unambiguous, with the narrator stating that Wallace mistook a construction site entrance for the beloved door, falling to his death. 

What appealed to me, aside from the text’s emphasis on childhood innocence as escapism, was Wallace’s relationship with the emerald door. In addition to how his obsession influenced his political career and personal ambitions, Wallace became almost completely reliant on the door, and its promises, to function. The very idea of passing through the door leads him to lose himself completely, and because of his childhood experience discovering what laid beyond the door, he does not question its veracity. He steps through, as he did before, only this time he plummets. As I read H.G. Wells’ story, I couldn’t help but think about our society’s current relationship with AI. 

Are we all falling to our deaths? Well, intellectually, we might be. In the age of artificial intelligence, many of us solve mundane tasks with an unregulated aid that feeds on the very questions we ask, using our inquiries to expand its database of knowledge. This tool uses our questions, prompts, and other seemingly trivial interactions to learn what information we seek, not only to best answer our questions, but also to legitimize what we want to hear. Many do not view the use of ChatGPT or other forms of generative AI as a loss of autonomy, but it certainly poses that risk. Currently, humanity is in a precarious position that dystopian, sci-fi novels have pondered for generations: what should we do when technology threatens to have overt influence over us by shaping our inquiries and thoughts? My answer is more simple than one might think: we need to start reading more. 

AI, in today’s society, is often used at the expense of cognitive development, despite scholars’ hopes for its greater learning potential. A recent MIT study focused on the accumulation of cognitive debt amongst young adults, a condition in which repeated reliance on external systems like OpenAI’s ChatGPT replace the effortful cognitive processes required for independent thinking. The study tracked not only alpha, gamma, and theta brain waves, which are responsible for creativity, focus, and deep memory consolidation, but also broader brain network interactions. Neural connectivity is essential as it forms physical connections in the brain that allow for processing information, generating thoughts, and acting. To conduct the study, the researchers split 54 subjects into three groups and asked them to write three SAT-style essays, with one group using OpenAI’s ChatGPT, one using a Google search engine, and the final group as the control, using no additional assistance. After writing the initial three essays, the participants were asked to rewrite their previous drafts; however, this time, access to ChatGPT was given to the control group and taken away from the ChatGPT group. The control group showed higher neural connectivity with network-wide spikes in alpha-, beta-, theta-, and directed connectivity, meaning that their brains were engaging more actively with the material, showing better retention and understanding. In contrast, the ChatGPT group demonstrated less coordinated neural efforts in most bands, as well as noticeable repetition in vocabulary used between participants and a limited ability to reconstruct the content of their previous essay without assistance. This begs the question: if AI helped the control group so drastically, why did the ChatGPT group suffer so much from AI usage in the first place?

This question can, in part, be understood through Lev Vygotsky’s theory on the Zone of Proximal Development. Vygotsky’s theory is applicable to a wide variety of fields, but the core idea reads as follows: “what the child is able to do in collaboration today, he will be able to do independently tomorrow.” In theory, AI has the possibility to be an incredible tool as learning is no longer confined to human collaboration. When AI becomes the collaborator, the possibilities seem endless as the mere act of typing can now theoretically bridge the gap between what the learner knows and what they could be taught in seconds. 

But, when the learning gap is merged completely, allowing for sudden gratification, the learner risks no longer understanding at all, and instead repeating the information presented to them rather than exhibiting true understanding—a drastic decline from what was possible. When an individual has learned information, they actively process and retain the information. Repeating information, on the other hand, is a passive action. Even worse, what happens when the information we repeat is biased or harmful? What guardrails exist to prevent people from hinging their understanding on false premises? What prevents us from mistaking a construction door for emerald when we are led to believe they are the same? 

Fears that AI-fueled disinformation campaigns could become widespread are becoming much more common. In a study published by the National Library of Medicine in April 2025, Wack et. al. establish how generative AI tools have already entered the realm of state-backed propaganda campaigns, changing the size and scope of the disinformation and misinformation a state can produce. In a September 2024 study Goldstein et al. argue that AI tools can enable the mass production of propaganda at a low cost, supercharging online covert disinformation campaigns. They conclude by encouraging policymakers to “assess effective intervention strategies,” warning that studies like these are possible because propaganda can still be tracked, which will not always be the case as AI becomes more intelligent and covert. Since the publication of both articles, few policy measures have been pursued. 

As early as May 2023, Sam Altman, CEO of OpenAI, warned of generative AI 's increasing capabilities and assured the public that he was keeping in “regular contact” with the government regarding the development of ChatGPT-4. While this seemed promising initially, these claims have not held up over time. Instead, what we now see is a budding partnership between OpenAI and the Trump administration leading up to the midterm elections. President Trump makes the administration’s policy on AI regulation very clear, stating that the White House will push for a “minimally burdensome national policy for AI,” despite growing fears among technology experts that AI is being used to generate propaganda and increase polarization. 

Meanwhile, OpenAI’s rival company, Anthropic, pushes for regulations. The company recently announced their funding of a pro-regulation super PAC, Public First Action, who lobbies for safety guardrails and greater government regulation over AI. Anthropic, founded by former researcher at OpenAI, Dario Amodei, states that the company is committed to funding and researching “constitutional AI,” a method of engaging with AI that combats posed safety risks, such as AI surpassing human control and destroying the job market. However, this has resulted in a slew of political problems for Anthropic, with the US Department of War designating the company a supply chain risk to America’s national security. The move appears to be in direct retaliation for the aforementioned push for stricter regulation. As is often asked in history, this prompts the question: if the government will not protect us, what will? 

My answer? Books. What I argue is applicable to “The Door In The Wall” has been stated more explicitly in popular dystopian fiction. In Ray Bradbury’s Fahrenheit 451, the government suppresses knowledge and independent thought with fire, allowing for a media-saturated, anti-intellectual culture to propagate. In George Orwell’s 1984, party members do not read traditional books, as it is considered a dangerous act of rebellion against the Party. In Brave New World by Aldous Huxley, citizens are conditioned from infancy to hate and distrust books. These three novels, some of the most well-known works of dystopian literature, depict reading as an inherent act of rebellion against government-issued propaganda. The natural rebuttal to this argument is that reading takes time and these lessons can be taught in less cumbersome ways, such as with AI, but this acceleration of learning is exactly where I believe the problem comes to a head. 

In a world of fast media and instant gratification, it is difficult to justify or identify benefits to slowing down. However, these benefits exist and they are imperative to our cognitive wellbeing. When you read, whether that be an advertisement on the subway or a four hundred page novel for class, you illuminate what are called “white matter pathways” in your brain. According to Dr. Nadine Gaab, these pathways are a series of nerve fibers that create a connective highway that allows information to quickly travel through your brain. When you read, you strengthen these pathways and improve your processing, cognition, and memory. Furthermore, some of the most important processing takes place after you have finished a book. Through analysis, presentation, and even recollection, you foster neuroplasticity, no matter how small the interaction with the text is. 

This process is what is missing when AI performs cognitive and learning processes for us. Our brains learn to trust what we are told; we don’t question it. And while there are plenty of individuals who trust what they read in a book without any push back, the difference lies in the source: books inherently invite active cognitive engagement, AI does not. Akin to how handwritten notes are correlated with greater recall in comparison to typed notes, reading leads to greater neuroplasticity and broader understanding than AI. 

In “The Door In The Wall”, the protagonist is not condemned because he sees the door, or even because he walks through it. The true flaw comes in what lies between perception and action. When he turns the knob and steps past the emerald barrier, he does so without questioning. He trusts that the door contains what it promises, and in turn he faces the consequences. 

So, are we all falling to our deaths? We don’t have to. Reading thoughtfully and actively is a form of rebellion: intentional strengthening of the brain and resistance to misinformation. As propaganda and misfortune become automated, we need to engage with AI carefully and prepare our brains to fight back. A door might appear emerald upon first glance, but what happens when you examine it closely? We must question the information we’re presented, whether from a chatbot or a human, and more importantly, look for the gaps. What is missing? Why is this the information someone wants you to know? Where did this information come from? All of these questions, even if only considered for a moment, act as a defensive shield against what is to come.

Caro Garlich (BC ‘28) is a staff writer for CPR and a junior in Barnard College, majoring in English and Political Science. They are most interested in exploring the relationship between literature and politics, in conversation with broader social movements.