Levels Of Decision Making And Certain Problems Of Artificial Intellect

TL;DR

This paper delves into the psychological intricacies of human decision-making, particularly at perceptual and speech-thinking levels. It argues that human decision-making is a dynamic, constructive process, not merely a logical deduction, and proposes that understanding these psychological peculiarities is crucial for advancing artificial intelligence.

What Problem Does This Paper Address?

The paper tackles the limitations of traditional artificial intelligence approaches, specifically heuristic programming, which often fail to capture the essence of human decision-making. While heuristic programming aims to formalize human thought processes, it overlooks key characteristics like the reflective and active nature of human cognition. The authors highlight that human decision-making isn't just about processing symbols but about actively constructing meaning and understanding from the world. They aim to bridge this gap by analyzing the psychological underpinnings of how humans make decisions, from recognizing objects to planning actions.

Methods

The paper employs a combination of experimental research and theoretical analysis.

  • Perceptual Decision Making: Experiments involved subjects identifying visual stimuli (figures, noisy images) under varying conditions. This allowed researchers to analyze the stages of image formation, the role of search operations, and the characteristics of perception and identification processes.
  • Speech-Thinking Level Decision Making: This involved experimental tasks where subjects had to construct an action mode based on logical conditions and object relationships. The researchers analyzed the "action principle" (reflective component) and the "operational component" (action program) of decision-making.
  • Comparative Analysis: The study compares different levels of decision-making (perceptual vs. speech-thinking) and contrasts human approaches with those of heuristic programming.

Datasets

Not discussed in the paper. The experiments described appear to use custom-generated stimuli and tasks rather than established datasets.

Key Insights and Ideas

  • Decision-Making is Constructive, Not Just Deductive: Human decision-making, especially at the perceptual level, is an active process of "bailing" new content from objects by integrating them into new relational systems. It's not just about recognizing what's already known.
  • The "Sliding" Nature of Standards: In identification tasks, the standards or criteria used for comparison are not fixed but "slide" and adapt based on the dynamics of image formation and the phase of the process.
  • Action Principle and Operational Component: Human decision-making at the speech-thinking level involves an "action principle" (reflective understanding of the situation) and an "operational component" (the actual action program). These two are interconnected but not always perfectly synchronized, with the operational component often lagging behind the action principle.
  • Levels of Generalization: Both perceptual and operational decision-making involve different levels of generalization, moving from basic perceptual cues to more abstract semantic meanings and action principles.
  • Heuristic Programming's Blind Spot: Traditional AI approaches often miss the reflective and active nature of human cognition, focusing too much on symbolic manipulation.

Key Results

  • Perceptual Image Formation: This process occurs in distinct phases, starting with basic spatial and color information and progressing to detailed feature differentiation.
  • Search Strategies: In complex visual search tasks, humans employ various strategies, including chaotic search, directed search, and semantic analysis, depending on the task and context.
  • Identification Efficiency: Simple object identification is a more curtailed process than general perception, relying on a shortened set of familiarizing components.
  • Action Mode Construction: Experimental tasks showed that the construction of an action mode is significantly influenced by the underlying "action principle" and tends to involve increasing levels of generalization. The operational scheme often lags behind the action principle.

Limitations and Future Work

  • Simulation for AI: The authors suggest that understanding and simulating these psychological peculiarities of human decision-making could be a promising avenue for developing more effective artificial intelligence.
  • Further Research: The paper implies a need for continued research into the interplay between cognitive and operational components of decision-making across different levels. The authors also hint at the importance of understanding how "standards" are formed and utilized in identification.

Summary updated

BibTeX
@inproceedings{1975levels,
  title = {{Levels Of Decision Making And Certain Problems Of Artificial Intellect}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {1975},
}
Figure