When Intelligence Is Not Human
Talking about alien intelligence does not necessarily mean thinking about extraterrestrial beings. The word alien can be used to refer to a form of intelligence of non-human origin, with a learning history, a body (or the absence of one), and constraints that are radically different from our own. It is an intelligence that does not share our biology or our developmental trajectory, and therefore we cannot assume that it thinks, represents the world, or reasons in the same way humans do.
There are multiple examples of “alien” intelligences in this sense. For instance, octopus cognition does not rely on our traditional categories: it is based on distributed nervous systems, with a strong capacity for problem solving and learning without a centralized brain like ours. Another example is the collective intelligence of ants or bees, which also constitutes an alien case: intelligence does not reside in an individual, but in the dynamics of the system. Even young children, from the perspective of developmental psychology, can be considered a partially alien form of intelligence: they possess sophisticated abilities, but organized in ways that do not coincide with the adult mind.
Get Sergio Gevatschnaider’s stories in your inbox
Join Medium for free to get updates from this writer.
These same ideas can be applied directly to large language models (LLMs). Although they produce fluent human language and seemingly intelligent responses, their way of learning and representing information is profoundly different. They have no body, emotions, intentions, or direct experience of the world; they learn from enormous volumes of textual data and optimize statistical predictions. The risk, then, is to confuse behavioral similarity with cognitive similarity: the fact that a model responds correctly does not imply that it does so for the same reasons as a person.
This approach is developed explicitly in the work “Evaluating the capacities of large language models”, authored by Michael C. Frank, a professor in the Department of Psychology at Stanford University. In this article, Frank proposes approaching LLMs as a form of alien intelligence and using tools from developmental and comparative psychology to study what kinds of representations underlie their behavior.
Rather than limiting evaluation to performance, the author suggests designing tests that assess generalization to genuinely novel situations, the use of artificial or simplified stimuli that avoid superficial shortcuts, strict control conditions, and converging evidence across multiple tasks. The goal is not to decide whether LLMs “think like humans,” but to understand what kind of intelligence they are, which abstractions they can acquire and which they cannot, and under what conditions those capacities emerge.
This idea, which may sound like science fiction — considering AI as an alien intelligence — is in fact a scientific strategy that forces us to abandon anthropocentrism and leads us to reflect on something fundamental about the nature of learning, both artificial and human.
