As artificial intelligence (AI) systems continue to outperform humans in an increasing range of specialized tasks—from playing complex games to generating coherent text and driving vehicles—a fundamental question emerges at the intersection of philosophy, cognitive science, and engineering: should we aim to build AIs that think like humans, or should we embrace non-humanlike architectures that may be more efficient or powerful, even if they diverge radically from biological intelligence?
This paper draws on a compelling analogy from the history of aviation: the fact that airplanes, while inspired by birds, do not fly like birds. Instead of flapping wings or mimicking avian anatomy, engineers developed fixed-wing aircraft governed by aerodynamic principles that enabled superior performance. This decoupling of function from biological form invites us to ask whether intelligence, like flight, can be achieved without replicating the mechanisms of the human mind.
We explore this analogy through three main lenses. First, we consider the philosophical implications: What does it mean for an entity to be intelligent if it does not share our cognitive processes? Can we meaningfully compare different forms of intelligence across radically different substrates? Second, we examine engineering trade-offs in building AIs modeled on human cognition (e.g., through neural-symbolic systems or cognitive architectures) versus those designed for performance alone (e.g., deep learning models). Finally, we explore the ethical consequences of diverging from human-like thinking in AI systems. If AIs do not think like us, how can we ensure alignment, predictability, and shared moral frameworks?
By critically evaluating these questions, the paper advocates for a pragmatic and pluralistic approach to AI design—one that values human-like understanding where it is useful (e.g., for interpretability or human-AI interaction), but also recognizes the potential of novel architectures unconstrained by biological precedent. Intelligence, we argue, may ultimately be a broader concept than the human example suggests, and embracing this plurality may be key to building robust, beneficial AI systems.
