The Sciences of the Artificial - 1

Interdisciplinary research gave me a toolbox to disassemble a host of theories across subjects such as statistics, psychology, physics, finance, and economics. Among the thinkers who continue to evoke awe in me is Herbert Simon, simply because of the depth of his philosophy and vision. Simon, often referred to as the father of Artificial Intelligence and behavioral finance, authored 30 books and over 1,000 research papers. He was awarded the Turing Award in 1975 and the Nobel Prize in 1978. Simon uniquely bridged psychology and intelligence, a core area of my research. While no single work in research can be considered purely seminal, Simon's, "The Sciences of the Artificial", first published in 1969, stands as a masterpiece for anyone passionate about modern science. Over the next seven features, I will revisit this work to contrast the AI of today with the AI of tomorrow.

Talking About Intelligence

Understanding intelligence begins with understanding nature. Artificial systems, or artifacts, are extensions of their environments. They show meaningful behaviour only when their design integrates various components as interconnected systems. These systems perform best when modelled with a focus on commonality and context rather than on overly specific details. Intelligence, in this view, lies in identifying patterns and extracting knowledge from natural systems.

Nature

Simon emphasizes natural science as a foundation for understanding complexity. Natural science seeks to uncover simplicity within what appears chaotic, finding patterns that make the complex comprehensible. Simon's phrase, "Wonderful but not incomprehensible" reflects this idea. Natural phenomena, when studied systematically, reveal logical structures that not only make them understandable but also deepen our appreciation of their beauty and order.

Artifice

Artificial systems are human-made constructs created to serve specific purposes. These artifacts embody both natural laws and human goals. For instance, while airplanes must obey the laws of gravity and aerodynamics, their design reflects the human purpose of transportation. The laws governing artificial systems arise from collective human effort e.g. economic systems, engineering principles, and organizational structures which Simon refers to as artifice. This dual adherence to natural laws and human purposes makes artifacts a unique and valuable domain of study.

Environment

The environment plays a critical role in shaping artifacts. Artifacts serve as interfaces between their internal mechanisms and the external environment. For example, a clock's design must adapt to whether it operates on land, at sea, or in space. Often, understanding an artifact's interaction with its environment provides sufficient insights into its behavior, reducing the need for detailed knowledge of its inner workings. Adaptive mechanisms, such as a ship's chronometer, ensure stable performance even under varying conditions.

Extension

Simon highlights that achieving goals or adapting to challenges depends largely on a few key environmental characteristics. Using the concept of homeostasis, he explains how systems maintain stability despite external changes. A well-designed artifact aligns its inner workings with its environmental demands, ensuring reliable performance. Recognizing and working within the limits and boundaries of an artifact's performance is essential for effective design.

Imitation

Artifacts often exhibit "resemblance from without rather than within," meaning they can mimic behavior without replicating internal structures. This explains why distinct physical systems, such as clocks or computers, can exhibit similar behaviors. Simulation is a powerful tool for understanding and testing these systems, much like a thought experiment that explores possible outcomes. Even when we know a system's premises, the implications of its interactions can be hard to predict, making simulation indispensable for discovery.

Design

Design challenges often revolve around predicting how different components within a system will interact. Understanding these assemblages is critical, as their organization determines the system's behavior. Abstraction helps simplify this process, allowing designers to focus on broader patterns without getting lost in unnecessary details. Simon underscores the importance of building knowledge incrementally, layer by layer, using abstraction and approximation.

Model

Simplified models capture a system's essential functions without requiring detailed knowledge of its components. For example, whether a structure is made of wood or iron may be irrelevant as long as it meets strength requirements. Computers, which Simon describes as truly protean, epitomize this concept. Their behavior depends on their organization rather than their physical makeup. Reliable systems can even be built from unreliable components through careful organization, ensuring consistent functionality despite individual failures.

Unreliable

Simon argues that unreliable components do not prevent the creation of reliable systems. By organizing components effectively, systems can achieve dependable performance. According to Simon, systems should be described based on how they perform the same functions despite having different structures. This perspective shows that systems do not need complete micro-theories of their components to be functional.

Abstraction

Abstraction is key to understanding broad functional characteristics of systems. Heuristic methods simplify problem-solving by focusing on key patterns rather than exhaustive detail. Simon notes that the complexity of computers often reflects the complexity of their environments rather than their physical properties. For Simon, the organization of components is more critical than their material composition. Symbol systems, such as computers and the human brain, adapt to their environments and manipulate information, making them central to the study of intelligence.

The Limits of Today's AI

Today's AI systems, including large language models (LLMs) and machine learning algorithms, diverge significantly from Simon's vision. LLMs rely heavily on statistical correlations and vast datasets, offering specificity without true environmental or adaptive integration. Machine learning systems prioritize optimization over contextual flexibility, while brain-inspired neural networks mimic human cognition superficially without embracing Simon's emphasis on organizing components to achieve intelligence. These approaches reduce intelligence to pattern recognition, failing to address the deeper challenges of understanding complex systems and their interactions with nature.

Tomorrow's AI

The future of AI must be contextual. We cannot endlessly generate data and rely on brute computational force to extract knowledge. This approach is inefficient and inherently unintelligent. Current methods treat intelligence like a juicing machine's throwing in massive amounts of data to produce outputs. While this may pass the Turing test, it cannot grapple with deeper challenges, such as imposing order on disorder or resolving contradictions in logic. True intelligence must simplify, compress, and abstract information, enriching knowledge while using less data. In this future, the power will rest with organizations that understand the DNA of data and prioritize contextual, adaptive designs.

Conclusion

Understanding intelligence is like decoding a map where only the landmarks are visible. Each landmark—whether a concept, a model, or a system—connects to others, forming a coherent whole that guides us through the complex terrain of nature and design. Just as a tree grows by balancing its roots and branches, the future of AI must balance its foundational structures with its adaptive capabilities. The art of intelligence lies not in brute force or endless data but in finding harmony between simplicity and complexity—turning the incomprehensible into something both wonderful and profoundly useful.