The era of unchecked optimism in the field of artificial intelligence may have to wait. Renowned computer scientist Peter W. Denning, in his new book "Turing's Mistake: Escaping the Yoke of Unintelligent Machine," puts forward a radical thesis: humanity has fixated on an unattainable goal. According to the researcher, the attempt to create a "disembodied mind" is doomed to fail due to fundamental algorithmic limitations.
Denning argues that modern AI development is heading into a dead end by trying to digitize that which, by its very nature, cannot be encoded. The scientist's key argument is the concept of tacit knowledge. This is a vast layer of human experience that cannot be described in words, recorded in bits, or transmitted through symbols.
Five spheres inaccessible to machines
The scientist identifies five main spheres of human existence that remain beyond the reach of machine learning. Practice over the last few decades has only confirmed this theory. A striking example is the ambitious Cyc project, launched by Douglas Lenat back in the 1980s.
The goal of the project was to create a complete database of common sense facts. Over 40 years of continuous work, 25 million entries were accumulated. Despite the colossal volume of data, expert systems never acquired intelligence. This proves that the simple accumulation of facts is not equal to understanding the world.
The problem of physical experience
AI limitations are also evident in the area of practical skills. Denning cites the example of an outstanding violinist. The musician can perform genius music, but he is physically unable to convey the feeling of his mastery to a student in words. This knowledge lives in the body and muscles.
Even if a robot can copy the musician's movements with surgical precision, it will remain devoid of a body capable of feeling the beauty of the performance or the emotional reaction of the listeners. For a machine, it is just a set of coordinates devoid of meaning.
Blindness to context and culture
The next insurmountable barrier is the context of the situation and the cultural dimension. It is these that determine whether a phrase is sarcasm, a kind joke, or an act of aggression. Context is always built on an endless chain of previous conversations and life experience.
Large language models, trained on huge arrays of data, only learn statistical connections between words. They remain "blind" to the true content and cultural background, not understanding the essence of what they generate.
The threat of "machine intelligence"
Peter Denning is convinced that instead of creating a tool friendly to humans, the scaling of neural networks leads to the formation of a completely alien and dangerous machine intelligence. Machine agent networks begin to produce their own "machine tacit knowledge".
A situation of mutual deafness arises: people cannot read the logic of AI decisions, and machines are unable to understand human motives. Reliable alignment of artificial intelligence goals with human intentions becomes impossible.
The main threat, according to the researcher, lies not in the takeover of the world by a hypothetical "superintelligence," as often shown in science fiction. The real danger lies in the chaos that autonomous low-intelligence machines can create, having no idea of human values and empathy.
