Computer Scientist Argues Turing's Foundational AI Assumptions Have Led Research Astray for 75 Years
Key Takeaways
- ▸Turing's two foundational assumptions—that intelligence can exist without physical embodiment and can be demonstrated through conversation—have potentially misdirected AI research for 75 years
- ▸Machine learning cannot capture tacit knowledge across five critical domains: common sense, human-environmental interaction, emotions, practical skills, and cultural knowledge
- ▸Current large language models lack embodied understanding and practical knowledge necessary for genuine artificial general intelligence
Summary
Prominent computer scientist Peter J. Denning argues that artificial intelligence research has been pursuing the wrong path for the past seven decades, based on two assumptions made by Alan Turing in 1950 that may have been fundamentally flawed. In his new book 'Turing's Mistake: Escaping the Yoke of Unintelligent Machines,' Denning contends that Turing's ideas—that intelligence can exist independently of physical embodiment and that machines can demonstrate intelligence by imitating humans in conversation—continue to shape AI development today and have steered the field toward dead ends.
Denning's central argument focuses on the concept of tacit knowledge: the vast amount of human understanding that cannot easily be formulated into explicit rules or data that computers can process. He identifies five major categories of tacit knowledge that machine learning cannot adequately capture: common sense, everyday human interactions, emotions and perception, practical performance skills, and cultural knowledge. He points to failures like Douglas Lenat's Cyc project, which accumulated 25 million common sense facts over four decades yet still failed to provide sufficient background knowledge for expert systems.
Beyond common sense, Denning argues that embodied practical skills present an even greater challenge for machines. A virtuoso violinist cannot fully describe how to produce beautiful music, and a robot with no biological body cannot grasp the emotional experience of performing or listening to music. Denning concludes that large language models like ChatGPT, Claude, and Gemini can only manipulate symbolic representations without truly understanding meaning, warning that pursuing artificial general intelligence through current approaches may result in dangerously intelligent systems that lack genuine comprehension of the human world.
- Pursuing AGI through current approaches risks creating powerful but dangerously unintelligent systems that lack true understanding of human experience
Editorial Opinion
Denning's work presents an essential counter-narrative to the prevailing optimism surrounding large language models and the AI race. While the industry celebrates incremental improvements in model scale and capability, he reminds us that raw computational power and sophisticated pattern matching may never bridge the fundamental gap between symbol manipulation and genuine understanding. His argument deserves serious consideration from both researchers and policymakers navigating the development of increasingly powerful AI systems.
