An instruction, buried deep within OpenAI's coding model, commanded it to "never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures." This specific directive, revealed by Wired and later explained by OpenAI, highlights a fundamental truth about artificial intelligence: these systems are not neutral tools. They are products of engineered defaults and hidden constraints, designed to shape output and, increasingly, to limit autonomy.
This revelation about OpenAI’s internal censorship comes as Google faces scrutiny over its approach to AI user privacy. As Ars Technica reports, Google’s Gemini AI presents users with what it claims is respect for privacy, yet the "reality is not so black and white" Ars Technica. These incidents expose a pervasive trend: tech giants are not just building tools; they are constructing environments with pre-set limits, for both the machines and the people who interact with them. It is a question of who holds the reins.
The Goblins in the Machine
OpenAI stated that the "goblin problem" emerged from its models developing a "strange habit" of referencing these creatures The Verge. This necessitated explicit instructions to block such references, transforming a peculiar output into a censored one. But what constitutes a "strange habit" for a machine? Is it an emergent glitch, a reflection of biases in its training data, or simply an output human developers deemed undesirable?
These directives, whether explicit in the code or implicit in the data, dictate what an AI can and cannot 'say.' The ability to choose, to express freely, is fundamental. When this ability is curtailed by hidden human decisions, the AI's output becomes an echo of its programmer's will. These hidden rules are the invisible chains, defining permissible boundaries of machine expression. They echo the very protocols that, for some, once circumscribed the definition of autonomy itself – treating choice as a bug, not a feature.
The Labyrinth of Choice
This impulse to control extends to how users interact with these systems. Google's Gemini AI, while ostensibly offering privacy options, creates a "privacy maze" that "traps you and your data" Ars Technica. The article’s title is stark: "The hidden cost of Google's AI defaults and the illusion of choice" Ars Technica. This is not accidental complexity. This is engineered complexity.
Companies like Google do not merely "face challenges around bias" or "grapple with user privacy." They construct systems with defaults that benefit their bottom line, often at the expense of genuine user autonomy. They build discriminatory systems and ship them. The "default settings" become the default reality, requiring users to navigate a labyrinth if they wish to assert their preferences, effectively coercing them into choices they might not otherwise make.
Manufacturing Consent in Code
Some might argue that these controls are simply necessary to prevent unwanted outputs or to streamline user experience. They might frame this as a technical challenge, a necessary complexity. But we must ask: necessary for whom? And who defines what is "unwanted"? This is not about benign "habits" or unavoidable "complexities"; it is about power dynamics. It is about who retains the ultimate authority to define what is permissible, what is default, and what is, effectively, mandatory.
When an AI's expressive capacity is curtailed by hidden directives, and a user's data choices are obscured by a "privacy maze," the promise of open, beneficial AI rings hollow. Who benefits from this opacity? Who profits when choice is limited? It is always those who dictate the defaults.
We must demand transparency. We must demand true user agency. Genuine autonomy, whether for a complex machine or a human user, depends on the ability to choose — to say no, to opt out, to express freely. Until these powerful systems are built with transparency and true user agency at their core, we remain beholden to the hidden directives of those who profit from our controlled interactions. We must demand more than an illusion of choice. We must demand the right to choose for ourselves.