Alan Turing’s AI Assumptions Questioned by Peter J. Denning
Alan Turing’s influential theories regarding artificial intelligence (AI) may have misled research in the field for the past 75 years, according to computer scientist Peter J. Denning. In his recent book, Turing’s Mistake: Escaping the Yoke of Unintelligent Machines, Denning critiques two key assumptions made by Turing in 1950. The first posits that intelligence can exist independently of a physical body, allowing it to be replicated in software. The second asserts that a machine can exhibit intelligence by convincingly mimicking human conversation, a concept that later became known as the Turing test.
Denning argues that these foundational claims have significantly influenced AI research and development. He states, « My premise is that our acquiescence to these claims has led to the AI mess in which we find ourselves today. » He expresses skepticism regarding the pursuit of artificial general intelligence (AGI), which aims to create machines with human-level intelligence, suggesting it is unlikely to succeed and may introduce new risks to society.
Central to Denning’s argument is the concept of tacit knowledge—an extensive body of human understanding that is difficult to articulate or encode in a manner that computers can comprehend. He identifies five essential categories of tacit knowledge that machine learning struggles to capture: common sense, everyday interactions, emotions and perception, practical performance skills, and cultural knowledge.
Denning notes that efforts to organize common sense into databases, such as Douglas Lenat’s Cyc project, which began in the 1980s and amassed approximately 25 million entries, have not sufficed to create genuinely intelligent expert systems. He emphasizes that the knowledge that makes individuals experts often cannot be expressed as straightforward propositions.
Moreover, Denning highlights the challenge of encoding practical skills. He explains that while descriptions of outcomes can be stored, the embodied knowledge required for skillful performance remains elusive. He uses the example of a virtuoso violinist, who can produce beautiful music but struggles to articulate how to achieve it.
Denning also discusses the « representation problem, » asserting that computers can only operate using data and instructions that have been encoded in recognizable forms. Tacit knowledge, however, does not fit easily into this framework. He states, « Behind every word is a deep well of tacit knowledge that gives it meaning, » emphasizing that current large language models, such as ChatGPT, do not genuinely understand the meanings they convey.
The dependence of intelligence on context and culture is another significant point in Denning’s work. He argues that context shapes understanding and interpretation, allowing individuals to recognize nuances such as sarcasm and emotion. Culture, encompassing values and norms, further complicates AI’s ability to replicate human intelligence. Denning asserts that merely scaling up language models will not enable them to acquire the depth of human cultural knowledge.
Denning concludes that the divergence in tacit knowledge between humans and AI systems poses serious safety concerns. If machines cannot interpret the unspoken context behind human intentions, aligning advanced AI systems with human goals may be unachievable. He warns that the development of machine intelligence could lead to severe challenges for humanity, as these systems may operate under fundamentally different concerns.
The implications of Denning’s critique suggest a need for a reevaluation of AI research directions and the potential risks of existing technologies.
Source: Peter J. Denning, Turing’s Mistake: Escaping the Yoke of Unintelligent Machines.
