liveBook · Manning

Thinking

As explained in Machines that Think

The mental process of forming thoughts, reasoning, or engaging in conscious cognition.

Thinking involves forming thoughts and reasoning, but it cannot be observed directly. We infer it from signs such as coherent expression, learning, adaptation, and problem solving. These signs matter, yet a correct or fluent response does not by itself reveal the process behind it or prove understanding. section 1.5 section 9.1.2

Language is both a means of expressing thought and a tool for organizing it. Naming helps distinguish and compare elements of experience, while syntax connects ideas and supports reasoning beyond immediate perception. This close association also makes fluent machine language tempting to treat as evidence of a mind, even though generating language does not settle whether a system thinks. section 2.1.1 section 2.1

Language and attention

Guugu Yimithirr speakers use cardinal directions rather than terms such as left and right, directing attention to orientation. Russian distinguishes light and dark blue as separate basic colors, an example of languages organizing perceptual categories differently. section 2.1.1

Figure 9.2 Intelligence inferred from observable signs. Standard definitions describe intelligence through capacities such as deriving information, learning from experience, adapting to the environment, and using reason. These capacities are recognized through performance, while the underlying mechanism remains partly hidden.

The figure distinguishes observable signs of intelligence from the partly hidden mechanisms behind them. section 9.1.2

Machine performance calls for a two-sided judgment. Hallucinations, bias, brittleness, and lack of grounding are serious limitations, but errors alone do not establish an absence of intelligence. Conversely, fluent explanations and successful problem solving do not prove human-like understanding: similar outputs can arise from different processes. section 9.0

Current AI systems show capacities associated with thought, including adaptation and problem solving, while lacking features central to human mental life, such as an embodied perspective and a continuous lived history. Their cognition may therefore be neither human thought reproduced nor empty imitation, but a form of intelligence with unfamiliar strengths and missing pieces. section 9.5.6

The boundary of thinking also shifts as machines master tasks once treated as distinctively intellectual. Arithmetic and chess became less persuasive as measures of thought when machines performed them, leaving open the question of which capacities should count and why. section 1.6.2

How Machines that Think covers it

How deep it goes
Develops an analytical comparison of observable capacities and hidden mechanisms, then qualifies claims about machine cognition without settling whether AI thinks.
Examples
Uses cardinal directions in Guugu Yimithirr, Russian color categories, calculators, and chess programs to illustrate language and changing standards of thought.
What you can do after
Evaluate claims about machine thinking by weighing observable performance, underlying mechanisms, and differences from human cognition.
How it connects
Connects thinking with language, intelligence, reasoning, understanding, grounding, and problem solving.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Machines that Think. The section links point to the parts of the book it draws on.