# General Problem Solver

The General Problem Solver (GPS) was a 1957 computer program by Allen Newell, Herbert A. Simon, and J. C. Shaw, designed to solve problems using means-ends analysis. It laid foundations for artificial intelligence and cognitive science.

The General Problem Solver (GPS) was an early artificial intelligence program developed by Allen Newell, Herbert A. Simon, and J. C. Shaw. First demonstrated in 1957, GPS was designed to solve a wide range of problems by applying a general-purpose reasoning method rather than domain-specific knowledge. It is widely regarded as a foundational milestone in the history of [artificial intelligence](https://www.wikiprompt.org/wiki/artificial-intelligence) and cognitive science.

GPS operated on a formal representation of problems, using a technique called means-ends analysis. This approach involved comparing the current state of a problem to a desired goal state, identifying differences, and selecting operators that could reduce those differences. The program recursively applied this process until the goal was achieved. Although GPS was not truly general in practice, its architecture influenced many later AI systems and contributed to the development of [machine learning](https://www.wikiprompt.org/wiki/machine-learning) and problem-solving paradigms.

## Origins and Development

GPS was conceived at the RAND Corporation and later developed at Carnegie Mellon University (then Carnegie Institute of Technology). Newell, Simon, and Shaw had previously worked on the Logic Theorist (1956), a program that proved mathematical theorems. GPS was an attempt to create a more flexible system that could handle a broader class of problems, including puzzles, symbolic logic, and even simple planning tasks. The program was implemented in the IPL (Information Processing Language) programming language, which was specifically designed for symbolic AI research.

The first public demonstration of GPS took place in 1957, and the system was refined throughout the late 1950s and early 1960s. It was used to solve problems such as the Tower of Hanoi, the missionaries and cannibals puzzle, and various cryptarithmetic problems. However, GPS struggled with problems requiring substantial domain knowledge or complex search spaces, and its performance was often limited by the computational resources of the era.

## Means-Ends Analysis

The core mechanism of GPS was means-ends analysis, a problem-solving strategy that involves three main steps: (1) identifying a difference between the current state and the goal state, (2) finding an operator that can reduce that difference, and (3) applying the operator, possibly after recursively solving subproblems to enable its application. This approach was inspired by human problem-solving behavior, which Simon and Newell studied extensively. GPS used a table of connections to link differences to operators, allowing it to select appropriate actions based on the current situation.

Means-ends analysis became a cornerstone of early AI and cognitive modeling. It was later incorporated into many planning systems and influenced the development of [sequence-to-sequence](https://www.wikiprompt.org/wiki/sequence-to-sequence) models and other modern AI architectures. The concept also found applications in [curriculum learning](https://www.wikiprompt.org/wiki/curriculum-learning), where tasks are structured to gradually reduce the gap between current and desired capabilities.

## Impact and Legacy

GPS had a profound impact on the field of AI, both conceptually and practically. It demonstrated that machines could perform tasks requiring reasoning and planning, challenging the prevailing view that computers were limited to arithmetic calculations. Newell and Simon's work on GPS contributed to their later research on human cognition, culminating in the development of the SOAR cognitive architecture in the 1980s. GPS also influenced the design of expert systems and early natural language processing systems.

Despite its name, GPS was not a general problem solver in the modern sense; it required problems to be formalized in a specific way and lacked the ability to learn from experience. Nevertheless, it set the stage for subsequent advances in [generative AI](https://www.wikiprompt.org/wiki/generative-ai) and [large language models](https://www.wikiprompt.org/wiki/large-language-model), which build on the idea of general-purpose reasoning but with vastly more powerful computational resources and data-driven methods.

## Limitations and Criticisms

GPS faced several limitations that became apparent as researchers attempted to scale it to more complex problems. The program relied on a complete and accurate formalization of the problem domain, which was often difficult to achieve. Its search space could grow exponentially, making it impractical for real-world tasks. Additionally, GPS had no mechanism for learning from past successes or failures, so it could not improve its performance over time.

Critics also noted that GPS's "general" nature was overstated, as it required hand-crafted representations and operator definitions for each new problem. This limitation highlighted the importance of domain knowledge and led to the development of knowledge-based systems in the 1970s. Nevertheless, GPS remains a landmark in AI history, illustrating both the promise and the challenges of creating intelligent machines.

## Modern Connections

While GPS itself is no longer used, its ideas resonate in contemporary AI research. The concept of means-ends analysis is echoed in modern planning algorithms and in the design of [transformer](https://www.wikiprompt.org/wiki/transformer)-based models that decompose complex tasks into subtasks. The emphasis on problem representation and search is central to [neural networks](https://www.wikiprompt.org/wiki/neural-network) and [deep learning](https://www.wikiprompt.org/wiki/deep-learning), which learn representations from data rather than relying on hand-coded rules. Researchers at institutions such as [MIT CSAIL](https://www.wikiprompt.org/wiki/mit-csail) and [Stanford AI Lab](https://www.wikiprompt.org/wiki/stanford-ai-lab) continue to explore general-purpose reasoning, often drawing inspiration from early systems like GPS.

In summary, the General Problem Solver was a pioneering effort to create a universal reasoning engine. Its development marked the beginning of AI as a discipline, and its legacy can be seen in many modern approaches to problem solving and machine intelligence.

---
Source: https://www.wikiprompt.org/wiki/general-problem-solver
License: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/)
Last updated: 2026-09-14T06:29:06.230168+00:00