Why AI cannot ‘think’ like humans do

The article explores the distinction between artificial intelligence and human cognition, highlighting that while modern large language models produce outputs resembling human thought, their underlying processes differ significantly. Human intelligence relies on predictive processing, where we constantly anticipate physical outcomes based on sensory feedback and prior experiences. In contrast, current AI primarily predicts textual patterns from vast datasets, allowing it to describe plausible scenarios without genuinely understanding the causal mechanisms or physical realities behind them. To bridge this gap, emerging "world model" technologies aim to teach AI to predict how situations evolve through interaction with the physical world. By incorporating actions, motion, and sensory data, these systems can learn to anticipate consequences similar to humans, such as a cup tipping over. This approach moves AI beyond passive data processing toward active engagement, enabling robots to learn from trial, error, and real-time feedback rather than relying solely on static information. However, the article concludes that achieving similar predictive results does not equate to human-like understanding or perception. While AI can mimic the ability to foresee outcomes, it lacks the contextual, embodied knowledge that humans develop through early childhood exploration and social interaction. This distinction is crucial for open data and AI ethics, as it underscores the necessity of scrutinizing data sources and model architectures to ensure transparency. Recognizing that performance does not imply comprehension helps developers and policymakers address the risks of overestimating AI capabilities and ensures more responsible integration of these technologies into society.

Source: theconversation.com
Published on 2026-11-06

Why AI cannot ‘think’ like humans do

Related news