Alan Turing’s famous ideas about artificial intelligence may have led AI research down the wrong path for the past 75 years, according to renowned computer scientist Peter J. Denning.
In his new book, Turing’s mistake: escaping the yoke of mindless machinesDenning argues that two fundamental assumptions made by Turing in 1950 continue to shape AI research today. The first is that intelligence can exist independently of the physical body and can therefore be recreated with computer software. The second was that machines could demonstrate intelligence by successfully imitating human speech, an idea that later became known as the Turing Test.
“These two arguments have shaped much of AI research and development,” Denning writes. “My premise is that our acquiescence to these claims led to the AI mess we’re in today.”
Denning argues that the pursuit of artificial general intelligence (AGI), or machines with human-level intelligence, is unlikely to succeed. Instead, he warns, the technologies society is building could pose new and significant risks.
The problem of tacit knowledge
Central to Denning’s argument is the idea of tacit knowledge. Tacit knowledge refers to the vast amount of human understanding that cannot be easily put into words or expressed in a form that a computer can process.
He says machine learning cannot capture five major categories of tacit knowledge: common sense, everyday interactions with people and the environment, emotions and cognition, practical performance skills, and culturally embedded social and historical knowledge.
Researchers have long tried to organize common knowledge into databases. One of the best-known efforts is Douglas Lenat’s Cyc project, which began in the 1980s with the aim of creating an extensive collection of common sense facts. After 40 years of work, the project included approximately 25 million entries.
“But even this Treasury has not been able to build up enough common sense background to make expert systems smart enough to be experts,” Denning said. “Sikes verified that much of the knowledge that makes people experts cannot be articulated as propositions.”
Denning believes there is an even bigger challenge with practical skills.
“Our performance skills in thousands of domains are not transferable to machines,” he explains. “Descriptions of skilled outcomes (‘what we know’) can often be expressed as bits and stored on machines, but we do not know how to encode the embodied knowledge (‘know-how’) for achieving skilled performance. ”
He uses accomplished musicians as an example.
“A master violinist can play beautiful music, but he cannot explain to an acolyte how to produce that music.
“Even if a robot could observe and imitate a skilled human, a robot without a biological body would not be able to understand how a musician feels when he plays beautiful music, or how an audience feels when he hears it.”
Denning also includes intuition, intuition, imagination, and spontaneous creativity as forms of tacit knowledge beyond the reach of machines.
Why human knowledge resists encoding
Denning argues that all of these limitations stem from what he calls the “problem of representation.”
Computers can only perform computations using data and instructions encoded in a physical format that it can understand and process. However, tacit knowledge does not naturally fit into that framework.
“Behind every word is a deep well of tacit knowledge that gives it meaning,” Denning says. “Words are only symbolic representations of meaning, not meaning itself. Commonly used large-scale language models such as ChatGPT, Claude, and Gemini only manipulate words; they cannot know or understand what the words mean.”
According to Denning, this creates a fundamental rift. Scientists still cannot fully explain how tacit knowledge works in humans, and therefore cannot translate it into a form that machines can use.
“How we host tacit knowledge is largely a mystery,” Denning admits. “All we know is that it is embodied. We have no idea what we can observe and measure in our bodies to reveal that.”
Context and culture shape intelligence
Denning also argues that intelligence is highly dependent on context, the surrounding circumstances that give meaning to words, actions, and decisions.
Depending on the context, people can recognize sarcasm, humor, sincerity, and emotion. It helps us decide when to be diplomatic, when to joke, and how to interpret myriad social cues.
“If you investigate where the assumptions in the current context come from, you’ll find that they are based on previous conversations from previous contexts. Each conversation is further based on previous conversations and their contexts. This pattern is endless and fractal,” Denning explains.
Culture is also a big hurdle for AI.
Denning explains that culture includes values, norms, judgments, history, community, atmosphere, and even relationships of power and consideration.
“Human speech incorporates background assumptions that give meaning and relevance to the words being used,” Denning explains.
“Scaling up LLMs with ever-larger neural networks does not allow LLMs to capture the embodied human knowledge known as culture. LLMs fail to achieve the Turing Test’s goal of demonstrating machine thinking indistinguishable from human thinking.”
AI safety and the limits of human understanding
Denning concludes that humans and AI systems may eventually develop different forms of tacit knowledge that neither can fully understand.
“Machines cannot read our tacit knowledge, and we cannot read their tacit knowledge,” he writes. “We are aliens who have crossed an insurmountable chasm.”
This gap, he argues, raises serious concerns about the safety of AI. If machines cannot interpret the implicit context behind human intentions, it may be impossible to reliably align advanced AI systems with human goals.
“Through AI automation, machine agent networks are likely to develop their own machine intelligence that does not reach the level of general human intelligence, but still has the potential to cause serious problems for humans. This threat is greater than a takeover by superintelligent machines,” he explains.
“Machine intelligence has different concerns than us and doesn’t seem to care about us. Its ways of thinking and problem-solving seem foreign to us. We don’t yet know how to live safely with these machines.
“Retreating from the singularity of AI automation will ask us to do a lot. We start by accepting that our familiar culture is disappearing and that we don’t know what will happen as intelligent machines emerge in our society. We refuse to think and be subordinate to machines. We refuse to submit to the yoke imposed by machines of low intelligence. Most importantly, we reaffirm our humanity, re-declare what makes us different from machines, and celebrate those differences.”
