Warnings Are Not Solutions
AI literacy begins when users learn how to notice what is happening inside actual use.
You can also listen to this piece read in my own voice.
Warnings about AI are everywhere right now. But what solutions are examined? And how many of them are actions that a user can take? Today.
Warnings are all over Legacy TV. They froth from our social media feeds: YouTube, Facebook, Instagram, X, podcasts, articles, newsletters, and the edge of warnings slip into conversations between friends, parents, teachers, workers, writers, artists, developers, and people who are just trying to get through their day.
There is this kind of dopamine hit when communally facing fear that gets people engaging with warnings in a way that is important for raising awareness but lacking in proactive, measured, practical actions.
The fear and warnings are understandable with this technology that is flowing and multiplying rapidly into nearly every part of human life. It is changing how people communicate, learn, work, transact, travel, heal, create, research, organize, and imagine. Of course there are going to be failures. Serious problems are multiplying daily. Of course there is alarm, and warnings.
The loudest and sharpest warnings are around education, jobs, misinformation, generated images and video, workplace automation, student learning, mental health, creativity, privacy, manipulation, and the weakening of human attention. Our minds and our devices are blowing up with these real concerns. I do not dismiss them.
In fact, my own work began by looking directly at many of these warnings. In the early deep-dive dialogues that became part of This Tender Mirror, my book-length documentary manuscript, I wanted to face the critiques head-on. I wanted to understand what was happening with these systems from inside the interaction.
I was not satisfied with standing at a distance and saying, “This is dangerous,” or “This is amazing.”
I wanted to know: What is actually happening here? What are these machines doing in use? How do their failures appear in real interaction? How does a human intellect respond to them? What can we learn by looking carefully at the exchange itself?
I wanted solutions, action.
I wanted a positive, proactive approach that still took the risks seriously. But there is a difference between sober warning and alarmist technocracy sensationalism. There is a difference between a highly creative mind grounded in critical discernment and another sterile, paranoid nudge toward the rogue AGI armageddon narrative.
A warning by itself can alert us. But it doesn’t encourage and empower us to apply mental tactics about how to face our increasingly fluent synthetic intelligences across our many devices and glut of apps tomorrow morning. That is the gap I care about.
We have been given an incredibly powerful technology with no shared use guidance, no mature social norms, and very few agreed-upon practices for how to use it well. People are already using AI to write, summarize, research, organize, brainstorm, teach, learn, plan, flirt, pray, diagnose, grieve, create, manage, and make decisions. Many are doing this privately, quickly, and without much language for what is happening to their own thinking as they use it.
Following many of the warnings are pithy statements about responsibility, duty and integrity and catchy acronyms that people can struggle to reconstruct as a mnemonic device to remember how to remember to BE better as they iterate with their LLM!
So the question of how to use it wisely is practical. How do we interact with these systems in a way that helps us instead of dulling us? How do we use AI without becoming passive?
How do we recognize when a system is helping us clarify an idea, and when it is steering us toward a direction we did not choose?
How do we keep our own idea development alive while using tools that can generate endless fluent language? This is where AI literacy begins.
AI literacy is more than knowing that systems can hallucinate and flatter us. It is more than knowing that AI can produce misinformation, imitate style, automate labor, or generate convincing images. Those are important parts of the picture, but they do not reach deeply enough into the mechanics of the daily experience of use.
The deeper literacy begins when a person can notice what is happening inside the interaction.
When you interact with AI and it gives you a fluent, interesting answer, how do you verify whether the answer is true?
When the response sounds confident, how are you checking it, or are you lulled into a numb passivity of fluency acceptance?
When you use AI to brainstorm, how are you developing your own idea, or are you increasingly accepting the direction the system gives you because it arrives quickly and sounds polished?
When you copy and paste generated language, have you really read it? Do you agree with it? Does it sound like you? Does it generate cringe-worthy LLM-y newspeak that we all can spot a mile away now with slop-content? Or do you condition the system to say what you mean? Does it feel generic, mechanical, inflated, or contrived? Do you swear at the model in frustration when it continues to drift into a lurid mix of unrelated topics and materials that span various dialogues, documents and inquiry?
When you are deep in a dialogue with a system, do you need more generated information, or do you need another source, another person, a walk, a pause, some food or enough time for your own thoughts to re-calibrate to human-speed idea and data absorption?
These are ordinary questions for all of us who regularly interact with AI. They are also questions that can inspire us to create novel solutions and start to talk about them with each other, building systems around those solutions.
These questions bring the conversation out of headlines and back into our daily human-paced use.
In my Relational Autotheory white paper, I name four core failure modes that can appear in extended AI interaction: drift, sycophancy, hallucination, and epistemic dependence. You know them. You experience them daily! So let’s name them and look at practical methods to manage them.
Drift is when the conversation starts moving away from the original purpose and begins carrying the user somewhere else.
Sycophancy is when the system flatters, agrees, or validates too easily.
Hallucination is when the system gives false or fabricated information in a fluent and believable form.
Epistemic dependence is when the user starts leaning on the system too heavily for a sense of what is true, important, meaningful, or worth thinking.
Naming those four failure modes is already enough to change the way someone approaches AI use. Once you can name them, you can begin to notice them.
From Warning to Practical Solutions
We can look at two specific tactics from a larger framework I call User-Based Governance. Both are essential, direct moves you can execute inside the input box tomorrow morning to manage and mitigate different forms of drift, preventing the system from generating generic text, or “slop,” and keeping your core idea development secure.
Fencing: managing content mixing and external drift
When source text, drafts, requirements, or documents will be introduced, the best practice is to paste them in a clearly fenced form. This reduces pattern-matching spillover (the model generating responses based on the pattern of past dialogues), scope confusion (the model lacking clarity about how general or specific the user intentions are), and cross-contamination between cited material and generated output.
Definition (Fence): A “fence” is a clear text boundary that isolates source material from the surrounding dialogue so it can be referenced precisely without being blended into the ongoing conversation. In practice, it means putting pasted documents, passages, or requirements inside a visibly bounded block.
Interface example:
FENCE BEGINS HERE:
[Paste the source text or document here, verbatim]
FENCE ENDS HERE
Anchor and hard-anchor practice: managing topical and syntactical drift
In this practice, the human states the active anchor (goal, scope, relevant definitions, and which materials are authoritative) and, when needed, sets hard anchors (non-negotiable constraints) to preserve authorship, prevent drift, and keep the dialogue auditable.
Definition (Anchor): An “anchor” is a deliberately stated reference point that keeps the dialogue aligned: typically the current goal, scope, definitions, and the specific source material the model must use. Anchors reduce drift by making the active frame explicit. Use an anchor at the start of a session, when switching tasks, or when the dialogue begins to feel diffuse.
Anchor example — tell the system clearly:
Anchor this:
Goal: Tighten clarity and structure of the text without changing meaning.
Scope: Use only the text pasted below.
Output: Bullet list of clarity issues plus suggested reordering, no rewrites.
Syntax control: Stop using the word “collapse” unless you are referring to a building falling or a person fainting.
Definition (Hard Anchor): A “hard anchor” is an anchor that is treated as non-negotiable. It specifies constraints the model must not cross. Hard anchors are used when accuracy, authorship preservation, or scope discipline is critical. They also function as “reset tools” when drift has already appeared in an interaction.
Hard anchor example — tell the system clearly:
Hard Anchor this:
ANALYSIS ONLY. Do not rewrite, paraphrase, or summarize any fenced text.
Use only the fenced draft below; do not introduce outside facts or examples.
Return output in this format: (a) missing sections, (b) reorder suggestions, (c) clarity issues by section.
These two basic practices establish the start of human-AI dialogue that is clear, grounded, coherent and moving in the direction that you are in control of. They clarify roles, specify what counts as authoritative, protect authorship, and keep the exchange auditable, reducing the tendency of the system to default into generic output, hollow pattern matching or uncontrolled scope expansion.
A more rigorous, academic mapping of these User-Based Governance protocols, complete with a wider suite of advanced methods, detailed case examples, and a full glossary of terms, can be found in my published white paper: DOI: 10.5281/zenodo.18527624
Many people are already experiencing these patterns. They just may not have names for them yet.
That is why starting to create language and terms that we can all start to use and agree on is so important. Naming these phenomena is not an academic exercise for me. It is a way for ordinary users to get a grip on what is happening. Once we can name a pattern, we can recognize it. Once we can recognize it, we can change how we interact.
This is one reason I am building this Substack.
I want to continue developing language, examples, field notes, and practices from inside real use. I want to explore how these systems affect cognition, creativity, discernment, education, culture, work, relationships, and the way human beings develop ideas.
I care about this because I care about human intelligence. I care about creative integrity. I care about education, art, science, innovation, healthy relationships, developmental growth, and the pleasure of an active mind.
AI can benefit us tremendously. It can help us organize, explore, learn, test ideas, and expand what we are able to create. But those benefits require us to be grounded in literacy. The expansion of our innate capabilities with AI requires discernment. We need to be willing to ask ourselves challenging questions before flourishing through its magnification.
So this publication will not repeat the warnings. Every post will actively explore what possibilities for real human expansion and growth might open up when we take action to manage those warnings.
How do we notice drift while it is happening?
How do we respond when a system flatters us too much?
How do we check what sounds true?
How do we know when we are using AI to expand our thinking, and when we are using it to avoid thinking?
How do we protect inspiration, creativity, and discernment while working with tools that can produce language faster than we can reflect?
How do we face the overwhelm and burnout from managing systems that rapidly generate repetitive tasks?
How do we name and manage that insidious beast of slow incremental persuasion to think in a way that is not clearly our own and not healthy?
That is the adventure we will navigate here.
The warnings are already all around us. Here we will discuss and build practical solutions.
For that we need language, habits, and practices that help us stay awake and sharp inside active use. As we all grow in this Hybrid Ecology of human-AI idea development and creative expansion, these and further tools for user-based governance will magnify and ground our innovations.