How the Structure of Thought Is Changing

The debate about artificial intelligence is lagging behind.

Public discussion, by inertia, focuses on the question of when machines will finally surpass humans in intellectual activity. For practice, this question is no longer the main one. The turning point occurs earlier: when a significant portion of intellectual operations moves beyond individual consciousness and becomes an external computational resource available on demand.

After that, the very procedure of solving a problem changes: first, an external reasoning loop is invoked, then the person checks, selects, corrects, and assembles the final position.

The threshold has already been crossed.

Intelligence begins to function as infrastructure. This change affects knowledge production, the organization of intellectual labor, and the criteria of professional value.

A good language for describing this shift is provided by an article by Stephen Shaw and Gideon Nave from Wharton. The authors expand the classic two-system model of thinking and introduce a third loop — external artificial reasoning. In this scheme, artificial intelligence acts as a participant in the judgment process: it offers options, forms a primary interpretation, sets the direction for finding a solution, and sometimes takes the place of internal rational work.

The central concept of the article is "cognitive capitulation".

This is what the authors call a situation in which a person accepts the model's output with minimal critical processing and adopts it as their own decision.

Behind this is a series of three studies: 1,372 participants and 9,593 observations. Participants turned to the assistant in more than half of the cases. In the first study, access to an accurate assistant increased answer accuracy by 25 percentage points compared to the no-assistance mode. Access to an error-prone assistant reduced accuracy by 15 points. In a pooled analysis of the three studies, the probability of a correct answer was more than 16 times higher when the external loop gave a correct answer than when it gave an incorrect one.

In all three studies, using the external loop increased participants' subjective confidence, including when the assistant was wrong. This means that not only the quality of the solution changed. The connection between the truth of an answer and the feeling of intellectual reliability was restructured.

The authors tested whether this effect could be mitigated. Time pressure reduced baseline accuracy by 13.5 percentage points, but dependence on the quality of the external assistant remained. Monetary rewards for accuracy and immediate feedback improved results but did not eliminate the problem. Among active users of the assistant, accuracy with correct prompts increased from 77.2% to 84.8%, and with incorrect prompts from 26.8% to 40.6%.

There are also differences between people. Higher trust in artificial intelligence increased the tendency to follow its answers. A propensity for analytical thinking and higher fluid intelligence acted as protection. When the external loop gave an incorrect answer, on average 73.2% of such episodes ended in cognitive capitulation.

This leads to a broader conclusion about labor. The cost of intellectual work itself as a factor of production is changing. If formalization, synthesis, primary interpretation, argument construction, and rough design become cheap services, individual intellectual power in the old sense ceases to be rare.

For professions of mental labor, this means a structural reduction in demand for a significant portion of human involvement.

Other abilities rise higher: problem formulation, discipline of verification, context retention, distinguishing the essential from the non-essential, and responsibility for the final decision. Therefore, the central question becomes the autonomy of the subject in conditions where reasoning increasingly unfolds in an external loop.


❗️❗️❗️❗️❗️❗️❗️❗️ / Not banned in the Russian Federation