Beyond Large Language Models: Are We Building Intelligence or an Illusion?
Based on a conversation in the InfoFire series between computer scientist Ricardo Baeza-Yates and information scientist Shalini R. Urs, this article explores the philosophical, cognitive, and societal shifts that have accompanied generative AI’s evolution from an external tool to an internal “cognitive companion.”
It essentially outlines five core arguments regarding how AI is reshaping human thought:
• The Great Illusion (Imitation vs. Comprehension): This section highlights the “ELIZA Effect,” in which people mistake generative AI’s fluent, polished language for genuine understanding. Baeza-Yates prefers the term “Computational Intelligence” to “Artificial Intelligence,” arguing that large language models generate language statistically. Seduced by linguistic fluency, we increasingly confuse prediction with cognition.
• The Delegation of Mind (Cognitive Offloading): Unlike older technologies that automated physical tasks or basic mathematics, generative AI risks “cognitive substitution” or “cognitive surrender.” By outsourcing intellectual exercises such as writing, coding, and reasoning to AI, humans may lose the productive cognitive struggle required to form original thoughts and develop critical judgment.
• Truth in the Age of Prediction: Information retrieval, such as through traditional search engines, preserves human agency and verification, whereas AI generation encourages immediate acceptance. LLMs do not “hallucinate” out of malfunction; they “confabulate” because their objective is statistical probability rather than factual truth. Detaching credibility from reality creates an “era of algorithmic plausibility.”
• Intelligence without Experience: AI systems process language but lack physical embodiment, causal reasoning, common sense, and “tacit knowledge” gained from living in the physical world. Furthermore, because training data relies heavily on dominant digital structures, such as English, it creates “AI colonialism”—a quiet standardization that compresses global cultural and intellectual diversity into a single dominant narrative.
• Education in the AI Era: Because AI can immediately perform lower-order cognitive tasks, such as summarizing and basic programming, education must shift from testing memory to cultivating human judgment and metacognition. Baeza-Yates suggests pedagogical changes such as requiring students to actively critique and debug AI-generated answers or bringing back oral examinations.
Ultimately, the article serves as a warning that AI should be used to amplify human intellect, not atrophy it; the greatest danger is that humanity might gradually surrender the very cognitive capacities that define its existence.
