# Human Problem Solving

Human problem solving is the cognitive process of identifying a goal, devising a path to reach it, and executing that path, studied across psychology, AI, and cognitive science. It involves heuristics, search, and insight, and has informed the design of artificial intelligence systems.

Human problem solving is the cognitive process by which individuals or groups identify a discrepancy between a current state and a desired goal, then devise and execute a sequence of operations to reduce or eliminate that discrepancy. It is a fundamental aspect of intelligence, encompassing everything from mundane tasks like planning a route to complex scientific discovery. The study of human problem solving has historically been a central topic in psychology and cognitive science, and it has also served as a foundational inspiration for the field of [artificial-intelligence](https://www.wikiprompt.org/wiki/artificial-intelligence), where researchers have attempted to model and replicate these processes in machines.

Early psychological research, particularly in the mid-20th century, framed problem solving as a form of search through a problem space. This perspective, championed by figures like Allen Newell and Herbert Simon, viewed a problem as consisting of an initial state, a goal state, and a set of operators that transform one state into another. The solver's task is to find a path through this space, often using heuristics - mental shortcuts or rules of thumb - to prune the search and make it tractable. This information-processing approach led to the development of early AI programs, such as the General Problem Solver, which explicitly attempted to mimic human heuristic search.

## Heuristics and Biases

A major strand of research, pioneered by psychologists Daniel Kahneman and Amos Tversky in the 1970s, focused on the heuristics that humans actually use, which often deviate from normative logic. They identified common shortcuts such as representativeness (judging probability by similarity), availability (judging frequency by ease of recall), and anchoring (over-relying on an initial piece of information). These heuristics are generally efficient but can lead to systematic errors or cognitive biases. This work challenged the notion of humans as purely rational actors and highlighted the bounded nature of human rationality - a concept that has influenced how [machine-learning](https://www.wikiprompt.org/wiki/machine-learning) models are evaluated, particularly in understanding their failure modes and limitations.

## Insight and Restructuring

Not all problem solving involves incremental search. Gestalt psychologists, earlier in the 20th century, emphasized the role of insight - a sudden, often non-conscious restructuring of the problem that leads to an immediate solution. Classic examples include the nine-dot puzzle, where the solver must break an implicit assumption about boundaries. Research on insight has shown that it often involves a period of impasse, followed by a sudden flash of understanding. This contrasts with algorithmic step-by-step approaches. In modern AI, the concept of insight has parallels in the ability of [large language models](https://www.wikiprompt.org/wiki/large-language-model) to produce novel solutions to unseen problems, though whether these systems genuinely restructure problems or rely on pattern completion remains a subject of debate among researchers like [Melanie Mitchell](https://www.wikiprompt.org/wiki/melanie-mitchell).

## Problem Solving in Artificial Intelligence

The formal study of human problem solving directly shaped early AI. The [chess computer](https://www.wikiprompt.org/wiki/chess-computer) Deep Blue, which defeated world champion Garry Kasparov in 1997, used a brute-force search algorithm, but its evaluation function was tuned with human chess knowledge. More recently, the shift from symbolic AI to [deep learning](https://www.wikiprompt.org/wiki/deep-learning) has moved the focus from explicit search to learning representations from data. [Neural networks](https://www.wikiprompt.org/wiki/neural-network), particularly [transformers](https://www.wikiprompt.org/wiki/transformer), are trained to predict next tokens or actions, and their problem-solving capabilities emerge from these statistical patterns. For instance, [OpenAI](https://www.wikiprompt.org/wiki/openai)'s GPT series and [Anthropic](https://www.wikiprompt.org/wiki/anthropic)'s Claude models can solve complex reasoning tasks, but their methods are opaque and often differ from human cognitive processes. Researchers like [Joshua Tenenbaum](https://www.wikiprompt.org/wiki/joshua-tenenbaum) at [MIT CSAIL](https://www.wikiprompt.org/wiki/mit-csail) argue that current AI lacks the causal understanding and intuitive physics that underpin human problem solving, advocating for a synthesis of deep learning with symbolic reasoning.

## Computational Models and Cognitive Architectures

To bridge the gap between human and machine problem solving, cognitive scientists have built computational models that simulate human performance. The ACT-R architecture, developed by John Anderson at [Carnegie Mellon University](https://www.wikiprompt.org/wiki/carnegie-mellon-university), attempts to model the underlying cognitive processes, including memory retrieval and procedural skill acquisition. Similarly, the SOAR architecture, created by Newell and his students, is designed to handle a wide range of problem-solving tasks using a production system. These models have been used to predict human reaction times and error rates in controlled experiments, providing a testable link between psychological theory and computational implementation. They also inform the design of more human-like AI, such as [Sanctuary AI](https://www.wikiprompt.org/wiki/sanctuary-ai)'s efforts to build general-purpose robots that reason about their environment.

## Modern Challenges and Future Directions

Contemporary research on human problem solving is increasingly interdisciplinary, combining psychology, neuroscience, and AI. One key challenge is understanding how humans solve ill-structured problems, where the goal or the operators are not clearly defined. Another is the role of creativity and analogical reasoning, where solutions from one domain are transferred to another. In AI, the pursuit of [generative AI](https://www.wikiprompt.org/wiki/generative-ai) that can solve novel problems has led to techniques like [RLHF](https://www.wikiprompt.org/wiki/rlaif) (reinforcement learning from human feedback) to align model behavior with human preferences. However, as of the early 2020s, AI systems still struggle with tasks that require common sense, long-term planning, and robust adaptation to changing circumstances - areas where human problem solving excels. The ongoing work at institutions like [Stanford AI Lab](https://www.wikiprompt.org/wiki/stanford-ai-lab) and [Berkeley AI Research](https://www.wikiprompt.org/wiki/berkeley-ai-research) continues to explore these frontiers, often drawing direct inspiration from cognitive science to build more capable and interpretable systems.

---
Source: https://www.wikiprompt.org/wiki/human-problem-solving
License: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/)
Last updated: 2026-09-14T06:31:03.071335+00:00