AI Myths & the Human Glitch

Why We Can't Stop Thinking AI Thinks

Feb 13, 2025

Engaging in conversation, solving complex problems, creating art: these used to be things only humans could do. Now, AI does them as well. The development of AI is a technological leap forward, but the conceptual tools we use to make sense of complex behaviors haven’t kept pace. When we see a system doing things that previously only humans could do, our default is to try to understand that system in human terms. We ascribe to it thoughts and consciousness the same way we would a person and either fear or celebrate its supposed similarity to human intelligence. But that’s a mistake—call it the human glitch: explaining the complex behavior of an artificial system in human terms.

The human glitch is shaping everything from corporate boardroom decisions to government policies to individual career choices. If we don’t diagnose the error and develop better conceptual tools for understanding what AI does, the implications for human well-being could be serious, even catastrophic. To chart a meaningful path forward, we have to start by understanding the difference between humans and AI—between human intelligence and artificial pattern matching.

Artificial Intelligence vs. Human Intelligence

Early AI researchers assumed that if they could just break down human thought processes into logical steps, they could replicate human intelligence. This led to what we now call "symbolic AI" or "classical AI"—systems that tried to represent knowledge as logical rules and relationships.

However, human intelligence proved more complex and subtle than these early models suggested. Consider how children learn language. They don't memorize grammar rules and vocabulary lists. They absorb language through immersion, trial and error, and pattern recognition. They learn not just words and rules but contexts, connotations, and the thousand subtle ways that meaning depends on the situation and relationship.

Modern AI systems mimic human learning more closely. That’s particularly true of the Large Language Models that power ChatGPT and Claude. Whereas early AI models were pre-programmed with grammar rules and vocabulary lists, modern AI systems are trained on massive data sets. They derive their own rules from the data and then follow those rules when producing the output. This training process is closer to human learning, but it still doesn’t do what a human child does.

Children are engaged in a cooperative biological activity choreographed by natural selection to facilitate group living. It’s an activity that aims at establishing and maintaining social bonds through a shared understanding of what people are thinking and feeling. AI systems aren’t engaged in anything like this. They’re not engaged in understanding, social bonding, or shared activity. Instead, they engage in a process of sophisticated pattern-matching. They use massive amounts of data to develop a list of templates—repeatable patterns of words or images. They then use those templates to predict which sequences of words or images are likely to come next in a given context. What they’re doing isn’t thinking in any meaningful sense. It’s merely pattern-matching—albeit pattern-matching more sophisticated than machines of the past could achieve.

The distinction between thinking and mere pattern-matching is crucial for understanding both the remarkable capabilities and the fundamental limitations of AI systems. They can process and recombine patterns in ways that seem magical. However, they're not engaging in the kind of meaning-making that characterizes human intelligence. They can tell you that a sentence is grammatically correct, but they don't understand what it means. They can generate text that perfectly matches the patterns of human writing, but they're not communicating ideas; they're just predicting word sequences.

Three Fundamental Myths about AI

Myth 1: AI Thinks and Acts Like Humans

People have the impression that AI thinks—where "thinks" means the same thing it does when we apply that word to a human. But this apparent similarity to human thought is a sophisticated illusion. When we perceive AI as human-like, it reveals more about how our pattern-seeking brains interpret behavior than about how AI actually works.

To understand why, we need to examine both human and artificial intelligence at their core. When humans think, we engage in a complex process of meaning-making. We don't just process information; we understand it. We don't just recognize patterns; we grasp their significance. We don't just manipulate symbols; we comprehend what they represent. This understanding isn't just about processing speed or pattern recognition. It's about meaning, context, and consciousness itself.

Consider how you understand a simple sentence like, "The coffee was too hot to drink." You don't just process the grammatical structure and vocabulary; you understand what it means to be too hot, what it feels like to burn your tongue, and why someone might want to drink coffee in the first place. You can effortlessly connect this sentence to your own experiences, to social contexts where coffee is important, to the physical properties of heat and the biological process of burning.

AI, in contrast, processes this sentence in a fundamentally different way. Modern AI systems, particularly large language models, operate by matching patterns. When we say an AI system is "trained," what we mean is that it has analyzed vast amounts of text and cataloged statistical relationships among words and phrases. It learns, for instance, that "coffee" frequently appears near words like "hot," "drink," and "morning." But it doesn't understand what coffee actually is; it’s never experienced temperature, and it cannot truly comprehend the concept of discomfort.

The statistical relationships among words and phrases that AI catalogs enable it to produce impressive simulations of real human acts of communication. If you don’t look past the surface of grammatically correct text, it’s easy to be fooled and think the text was written by a human who intended to communicate something. Look past the surface, however, and cracks start to show: AI makes mistakes. For example, it might confidently cite a non-existent research paper, combine facts from different contexts in misleading ways, or generate expert-sounding but entirely fictional advice. Worse yet, AI commits these mistakes using language that suggests complete confidence in the message. Researchers call mistakes of this sort "hallucinations."

Hallucinations aren't random errors but the natural result of how AI systems work. If certain patterns appear frequently in the training data, AI will reproduce them, regardless of whether they say something true or false. AI isn’t lying or committing errors in the way humans do. It’s not concerned with truth or falsity the way humans are, and it can’t form intentions to deceive others in the way humans do. Instead, it simply generates outputs that match the patterns it has learned. It uses those same patterns to say things that are true and things that are false, thus conveying both information and misinformation.

Myth 2: AI Will Replace Human Workers

People are afraid that they will be replaced wholesale by AI: that robots or algorithms will take over human jobs and leave people unemployed. The fear of technological replacement isn't new—the Luddites faced similar anxieties with mechanical looms. However, the current wave of AI development seems particularly threatening. It seems to challenge not just our physical capabilities but our cognitive supremacy.

But this concern also stems from a misunderstanding of both human and artificial intelligence.

Consider what happens when a human writer crafts a story. They draw on personal experience, emotional understanding, and cultural context. They might recall the smell of their grandmother's kitchen and somehow connect that sensory memory to a broader point about technology and tradition. They understand not just the meaning of words but the weight they carry in different contexts, the subtle implications they might have for different readers, and the cultural and emotional resonances they evoke.

What we're actually seeing isn't replacement but transformation. Jobs don't typically disappear entirely; they evolve. Just as the introduction of spreadsheets transformed accounting from a profession focused on calculation to a profession centered on analysis and strategy, AI is shifting human work toward tasks that require uniquely human capabilities: judgment, creativity, emotional intelligence, and ethical reasoning.

This transformation is already visible across multiple fields:

In medicine, AI can analyze radiological images with remarkable accuracy. It often spots patterns that human doctors might miss. But it can't integrate this analysis with a patient's life circumstances, medical history, and treatment preferences. It can't have difficult conversations about prognosis or make complex ethical judgments about care options.
In law, AI can review thousands of documents in hours. It can identify patterns and relationships that might take human lawyers weeks to find. But it can't understand the broader strategic implications of these patterns, can't negotiate with opposing counsel, can't build the trust relationships that form the foundation of legal practice.
Even in technical fields where AI excels at pattern recognition and data analysis, the technology tends to augment rather than replace human capabilities. Programmers using AI coding assistants aren't being replaced; they're being freed from routine coding tasks to focus on system architecture and problem-solving.

Instead of seeing AI as a super-powerful intern who can suddenly do any human job, it’s better to view it as a task-oriented intern who can take the boring stuff off your plate.

Myth 3: AI Is a Reliable Source of Educational Content

Perhaps the most dangerous myth about AI is the belief that it can serve as a reliable, independent source of educational content. This misconception is particularly seductive in education, where the idea of AI as the perfect tutor—one who never tires, never loses patience, who can access all human knowledge instantly and adapt to each student's needs—seems to offer solutions to many educational challenges.

However, this vision fundamentally misunderstands both the nature of education and the limitations of AI. When an AI system generates educational content, it's not drawing on understanding; it's generating text patterns that statistically match what explanations look like. When it responds to a student's question, it's not comprehending the confusion or crafting an explanation based on understanding. Instead, it matches the pattern of the question to the patterns of likely helpful responses.

There are at least four problems with relying on AI as a primary source of educational content:

Unreliable Content: AI hallucinations are particularly dangerous in education. The convincing-looking content that AI generates can contain subtle or even significant errors. An AI might confidently present incorrect historical dates, misstate scientific principles, or blend different concepts in misleading ways. In many cases, students and even teachers might lack the subject expertise to spot those errors.
Inability To Understand: While AI can generate surface-level explanations that follow the patterns of good teaching, it doesn’t understand the subject matter and can’t assess whether a student genuinely understands it, either. It can't tell when a student has memorized the right words and yet failed to grasp the underlying concepts. As a result, it can't guide students past the illusion of understanding to genuine mastery.
Lack of Cultural and Contextual Awareness: Educational content needs to be culturally relevant and appropriate for specific learning contexts. AI lacks true understanding of human cultural nuances and social contexts. As a result, it can generate content that's technically accurate but culturally inappropriate.
Lack of a Learning Meta-Model: Effective education requires something AI lacks: the ability to model learning itself. When human teachers explain concepts, they draw on their own experience of learning that material—they remember what confused them, what helped them understand, and how different approaches work for different learners. This meta-understanding of the learning process itself is something that AI lacks. A human teacher might notice a student's eyes light up when relating physics to sports or sense confusion in a slight hesitation before answering. These subtle cues inform real-time adjustments in teaching strategy that no AI, however sophisticated, can truly replicate.

The challenge of using AI for education isn't just limited to content delivery but also the complex interplay among teaching, learning, and understanding. While AI can simulate the patterns of good teaching, it cannot engage in the relationship that makes deep learning possible.

This doesn't mean AI has no place in education. The most effective educational future will be one where AI and human teachers work together, each doing what they do best. Human teachers bring understanding, judgment, and the ability to foster genuine learning in the following ways:

Detecting subtle signs of confusion or engagement,
Drawing on their own learning experience to anticipate misconceptions,
Understanding not just the what of their subject but the why,
Building the relationships that make learning possible,
Guiding students toward genuine understanding rather than mere pattern recognition.

AI can support human learning in the following ways:

helping students practice skills,
providing immediate feedback on routine tasks,
offering additional explanations and examples,
Helping teachers handle administrative tasks, generate preliminary teaching materials, and identify patterns in student performance.

The Path Forward

These three myths—about AI's thinking capabilities, its role in the workforce, and its reliability as an educational tool—all stem from our tendency to project human qualities onto pattern-matching machines. Moving beyond them requires more than just understanding; it requires developing what we might call "AI literacy."

What does AI literacy look like in practice? Consider a journalist using AI to investigate a complex story. Instead of asking the AI to "analyze the situation" or "find the truth," an AI-literate journalist knows to use it for specific tasks: scanning thousands of documents for relevant names and dates, identifying statistical patterns in data, or generating possible questions to ask human sources. The journalist maintains control over the investigation's direction, fact-checking, and ethical judgments—while letting AI handle the pattern-matching tasks that computers do best.

Or take a teacher who understands that AI isn't a replacement for education but a tool for amplifying human instruction. Rather than relying on AI to explain concepts directly to students, they might use it to generate diverse examples, create practice problems at different difficulty levels, or identify patterns in student mistakes. The teacher's human insight determines how to use these AI-generated resources effectively.

These examples show how progress with AI mirrors the broader story of human tool use. We didn't become the planet's dominant species by being the strongest or fastest but by being the best at making and using tools to amplify our uniquely human capabilities. The key is understanding exactly where AI's pattern-matching abilities end, and human judgment must begin.

The path forward begins with a shift in how we think and talk about AI. Instead of asking, "Is this AI conscious?" or "Will this AI replace humans?" we should ask more precise questions: "What patterns is this system recognizing?" "Where might its pattern-matching break down?" "What uniquely human capabilities does this task require?"

When you next encounter AI, whether you're using ChatGPT for writing, considering AI tools for your business, or evaluating AI-powered educational software, catch yourself being subject to the human glitch. Notice when you start attributing human qualities to the system. Then reframe your thinking: you're not dealing with an artificial human but with a powerful pattern-matching tool that can amplify—but never replace—human understanding.

The future belongs not to those who fear AI or worship it but to those who see it clearly for what it is: a mirror that shows us, through contrast, what makes us uniquely human.

My sincere thanks to Ellen Fishbein and William Jaworski for their incredible support in creating this essay. You are amazing!
All opinions and views expressed here are mine alone and do not represent those of my employer.