Why AI May Never Reach Human Intelligence - SciTechDaily

What the report says

SciTechDaily reported on a new analysis by computer scientist Peter J. Denning arguing that artificial intelligence may not achieve human-level intelligence because key parts of human understanding cannot be reduced to programmable instructions. Denning’s essay, titled “Turing’s Mistake: Escaping the Yoke of Unintelligent Machines,” revisits ideas associated with Alan Turing’s 1950 imitation game and the view that intelligence could be separated from the body and recreated in software.

Denning’s central claim is that AI systems lack tacit knowledge: the unspoken background that supports common sense, social judgment, intuition, practical skill, perception, and culture. As described by SciTechDaily, he cites decades-long efforts such as Douglas Lenat’s Cyc project, which accumulated millions of common-sense entries but still did not capture the deeper human context needed for expert reasoning. Denning also distinguishes between describing an outcome and possessing embodied skill, using musicianship as an example of knowledge that is difficult to express as rules.

The analysis says this limitation creates a “representation problem” for machines that process symbols and data without sharing human meanings, bodies, histories, or cultural experience. Denning argues that larger language models alone will not solve this gap, because words depend on context that extends through prior conversations, social norms, and lived experience.

Why it matters: Denning’s warning is not mainly about superintelligent machines overtaking humans. According to the report, he is more concerned that automated, networked AI systems could become powerful and hard to align with human goals while still lacking human understanding. The argument adds to ongoing debate over whether current AI scaling can lead to artificial general intelligence or whether fundamentally different approaches are needed.