Can artificial intelligence think like a human?

Artificial intelligence based on large language models, such as ChatGPT, Claude, and Gemini, can produce text and ideas resembling human ones by drawing inspiration from the structure of the human brain through artificial neural networks. The article explores the distinction between similarity in results and similarity in underlying processes between AI and human thought, citing Pierre-Luc de Saint-Saul. A key concept introduced is "predictive processing," which suggests that perception involves creating expectations, comparing them with incoming information, and making adjustments. This process is illustrated by the example of predicting the outcome of reaching for an empty glass. While chatbots predict the next word based on learned patterns from data, this differs from human cognition, as the text argues AI does not perceive or understand the world in the same way humans do. The concept of "world models" is emerging technology aiming to bring AI closer to human thought by enabling systems to predict state evolution and act upon those predictions, which could eventually lead to a resemblance to human understanding, though this is not proven.

The core tension in this discussion lies between performance and process: the ability of AI to reproduce human-like results versus its actual internal mechanism for achieving those results. The analogy of predictive processing—where expectations shape action and experience updates expectations—provides a framework for analyzing how world models might bridge this gap. Humans develop complex understanding through embodied experience, where learning involves connecting sensory input (feeling weight, seeing pouring) with potential actions, leading to mental predictions without necessarily performing the physical act itself. Conversely, current AI prediction operates on statistical pattern matching from vast datasets; it can describe outcomes but does not inherently possess the subjective, embodied understanding that shapes human cognition. The implication is that achieving functional equivalence through simulation does not automatically equate to possessing genuine perception or understanding. The development of world models in AI suggests a potential path where experiential learning (incorporating actions and sensory feedback) could force systems to develop internal representations similar to human knowledge structures. The critical unresolved question remains how the nature of embodied experience translates into the statistical patterns that constitute true understanding. What is the cost or benefit of an AI system successfully simulating cognition without actually inhabiting it?