Is AI Actually Intelligent Experts Reveal Why the Term Might Be Misleading

The question of whether artificial intelligence possesses—or will eventually achieve—genuine intelligence remains one of the most debated topics in modern science. By consulting experts across linguistics, computer science, cognitive science, and psychology, it becomes clear that while AI systems are advancing rapidly, the definition of “intelligence” itself is far from settled. If there is a consensus among researchers, it is that our current terminology may be fundamentally inadequate for describing what these machines are actually doing.

Benjamin K. Bergen, a cognitive scientist at the University of California, San Diego, notes that the very concept of intelligence is highly contested, even within psychology. Despite this lack of consensus, experts are increasingly focused on parsing the mechanics behind AI to determine what, if anything, separates these synthetic systems from human cognitive abilities.

The Illusion of Understanding

Emily M. Bender, a professor of linguistics at the University of Washington, argues that we should be skeptical of the term “AI” itself, viewing it more as a marketing label than a coherent technology. When it comes to large language models (LLMs), Bender suggests they are not “thinking” in any human sense.

Linguistics, she explains, teaches us that language is a system where form is paired with meaning. While LLMs excel at manipulating form—the structure and sequence of words—they lack any inherent connection to meaning. Human communication, by contrast, relies on a deep, reflexive process of imagining the state of mind of the speaker. We instinctively project intent onto text, even when none exists. According to the influential 2021 Stochastic Parrots paper, this human predisposition to find coherent meaning in communicative acts makes rigorous evaluation of AI systems exceptionally difficult.

Beyond Cognitive Tasks

Melanie Mitchell, a computer scientist at the Santa Fe Institute, questions the industry trend of compartmentalizing “cognitive intelligence” from the physical and social skills human brains have evolved to handle. While an AI might provide instructions for a home repair, the human user is the one exercising intelligence to interpret those instructions and navigate physical reality.

Mitchell notes that AI remains notably weak in social intelligence—the capacity to coordinate with others to solve problems or navigate social landscapes. Whether this is due to a lack of proper training or an inherent inability to participate in the social world remains an open question in the field.

Intelligence as Adaptation

At Caltech, Anima Anandkumar views intelligence through the lens of data-driven adaptation. For her, the capability to learn from environments and adjust behavior is the core marker of intelligence. She points to AI breakthroughs in scientific modeling, such as weather forecasting, as prime examples of this potential. By processing physical laws at scales ranging from microscopic chemical reactions to planetary climate systems, these models can outperform traditional methods, potentially saving lives through more accurate, early predictions of extreme weather events.

Diverse Capacities of the Mind

Alison Gopnik, a psychologist at UC Berkeley, argues that the quest for “Artificial General Intelligence” is based on the flawed assumption that intelligence is a single, linear scale. Cognitive science, she explains, reveals a mosaic of varied capacities that often serve conflicting functions.

Current AI models excel at “cultural transmission,” utilizing statistical patterns from massive datasets to generate new content. However, this is distinct from the human capacity for active exploration—a hallmark of early childhood development. Children possess a unique ability to build internal models of the world, often through the social “intelligence of care,” where they explore safely while others handle complex cognitive and physical burdens. While current models mirror some human outputs, they remain distant from the holistic, developmental intelligence found in biological life.

The “Clever Hans” Problem

Benjamin K. Bergen highlights the danger of confusing intelligent-seeming behavior with actual intelligence. He uses the historical example of “Clever Hans,” a horse believed to be capable of arithmetic, who was actually just responding to subtle physical cues from its trainer. Similarly, modern AIs may appear to reason logically, but upon closer inspection, they are often simply regurgitating patterns found in their training data.

Ultimately, these experts suggest that while AI demonstrates behaviors consistent with some definitions of intelligence, the mechanisms beneath the surface are fundamentally different from those of human or animal cognition. Whether that difference matters in the long run remains the subject of intense scientific inquiry.