The way artificial intelligence is reshaping our understanding of human psychology is profound, but before we can truly grasp how machines might mimic or understand us, we need to ask a foundational question: what exactly makes human language unique? A recent exploration into animal cognition reveals three key features that distinguish true language from mere word use—and these insights have major implications for AI development.
What Is AI in Psychology?
Artificial intelligence in psychology refers to computational systems designed to model, simulate, or augment human cognitive processes. From natural language processing to affective computing, AI tools are increasingly used to study everything from decision-making patterns to emotional responses. But as we build systems that can generate coherent text and even mimic conversational nuance, a critical question emerges: can AI truly understand human psychology, or is it simply operating with sophisticated pattern matching?
The distinction between words and language matters here. Many animals—including chimpanzees, dogs, and birds—demonstrate remarkable abilities to understand and use individual words. Chaser the border collie understood over 1,000 spoken English words. Kanzi the chimpanzee distinguished between "Put the cup on the dog" and "Put the dog on the cup." Yet these achievements don't constitute true language because they lack three essential features that humans possess.
The Three Features That Define Human Language
Rule-Based Grammar
Grammar provides structure to our communication. When Kanzi separated toys based on word order, researchers celebrated because he appeared to understand that "Put the cup on the dog" means something fundamentally different from "Put the dog on the cup." This isn't just memorizing phrases—it's grasping relationships between words and how their sequence creates meaning.
In human language, the difference between "Ben saw Miranda" and "Miranda saw Ben" is deep and fundamental. We don't need to relearn these; our brains automatically process the grammatical structure that tells us who did what to whom. AI systems struggle with this kind of contextual understanding, often producing outputs where the grammar is technically correct but the meaning falls apart, evidence that pattern recognition alone doesn't equal comprehension.
Generativity
Generativity refers to the ability to produce new combinations that make sense, even if they've never been heard before. Washoe, a chimpanzee who used sign language, once spontaneously signed "tiger-horse" when she saw a zebra in a book. This was impressive, but human children do this constantly and more creatively.
"I goed to the park" or "We'll meet nexterday", these aren't just errors; they're generative attempts at creating new linguistic forms that fit grammatical rules, even if imperfectly. Children make these mistakes precisely because they're internalizing the rules rather than memorizing phrases. AI can generate novel sentences, but it lacks the intuitive grasp of why "I goed" feels wrong in a way that reveals an underlying understanding of correct grammar.
Arbitrariness of Word Assignment
Perhaps the most counterintuitive feature is this: there's no inherent connection between a word and what it represents. A turkey doesn't look like the word "turkey," nor does anything about its appearance suggest that name. The connection is arbitrary, a random assignment that humans collectively agree upon.
German marks plurals with "-e" or "-n"; English uses "-s." Both work equally well; there's nothing logical about either system. Human languages occasionally use onomatopoeia, words like "boing" or "trickle" that sound like what they represent, but these are rare exceptions, not the rule.
This arbitrariness is where AI shows its limitations. Word embedding models can learn associations (like "king" minus "man" plus "woman" equals "queen") by detecting statistical patterns across massive text corpora. But there's no true understanding of why those associations matter, the system is tracking correlations, not grasping meaning in the way a human does when they encounter an entirely new word and figure it out from context.
Errors as Insight
In graduate school, I learned about a four-year-old who encountered a pitchfork for the first time. Looking at its tines, she declared: "At my house we have forks, not threeks." The child was conducting what amounted to linguistic field research, collecting data on how words map to objects in her world.
This is precisely how children learn language: through active hypothesis testing and theory building. They don't just absorb vocabulary; they discover the arbitrary nature of word-object associations. Adults often miss this because we've thoroughly internalized arbitrariness, we don't notice that "fork" could have been any random sound as long as everyone agrees on it.
AI systems, by contrast, never make these kinds of errors. They don't say "threeks" or create their own linguistic rules, they either learn the associations correctly (or not) based on statistical probability, or they fail. This absence of generative error is telling: it suggests that AI language models operate fundamentally differently from human cognition, even when they produce outputs that sound remarkably natural.
Beyond Words: What Animals Can Teach Us About Language and Psychology
Many animals communicate with sophisticated systems. Lions use about ten distinct sounds to coordinate among themselves; mockingbirds can produce up to 200 different vocalizations, most of which are mimicry. Dogs understand hundreds of words and gestures. Yet none of these systems demonstrate the combination of grammar, generativity, and arbitrariness that characterizes human language.
This matters for AI psychology because it highlights what might be uniquely human about our cognitive abilities: not just intelligence or pattern recognition, but the capacity to create an open-ended system of symbolic representation that allows us to think about thinking itself. A fourth emerging feature may be meta-linguistic awareness, the ability to reflect on language as a system, to question why we say things the way we do, and to understand how meaning is constructed socially.
Can AI Understand Human Psychology?
The answer lies in understanding what "understand" means. If we define it as statistical prediction, anticipating what comes next in text or conversation, then yes, AI can achieve impressive results. Large language models can generate psychologically insightful content and pass tests designed to measure human-like reasoning.
But if we define understanding as the kind of generative, rule-based, arbitrary symbol manipulation that characterizes human language, and which appears to be rooted in the neural architecture itself rather than learned through data, then AI falls short. It doesn't have an internal model of how words work together; it has a probability distribution over token sequences.
This distinction matters for psychology because understanding human cognition requires more than predicting responses, it requires modeling how minds actually construct meaning, make errors that reveal underlying rules, and creatively extend their linguistic systems in unexpected ways. As AI continues to advance, the insights from animal cognition research remind us that there's still a lot we don't know about what makes human language, and by extension, human psychology, truly unique.
This article is based on research published in Psychology Today by Dr. Sarah Dunphy-Lelii, Associate Professor of Psychology at Bard College, specializing in animal cognition and child development.
