Forecasting and Assessing the Impact of Artificial Intelligence on Society

Oscar Firschein, Martin A. Fischler, L. Stephen Coles, Jay M. Tenenbaum

Introduction

This 1973 paper serves as a remarkable "time capsule" from the early era of Artificial Intelligence (AI) research. Authored by pioneers at Lockheed and SRI, the study represents a rigorous attempt to move beyond science fiction conjecture. It provides a systematic, expert-driven forecast of how AI would evolve and impact society over the subsequent three decades.

The Core Challenge

By the early 1970s, AI had transitioned out of its infancy, yet its future was obscured by a mix of "overly optimistic" short-term hype and "overly pessimistic" long-term skepticism. The authors argued that because computer systems possess the potential for profound, century-long effects on human life, their trajectory could not be ignored. The paper sought to bridge the gap between technical capability—what can be built—and societal judgment—what should be built and the consequences of doing so.

Methodology: The Delphi Approach

To achieve a structured forecast, the authors employed the Delphi Method, an iterative technique designed to reach a consensus among experts. The process was divided into two primary domains: Technical Analysis and Judgmental Effects on Society.

Figure 1: The overall methodology of the study, balancing technical analysis with judgmental societal effects.

The methodology involved several key steps:

  1. Partitioning: The AI field was divided into functional categories such as Language Understanding, Problem Solving, and Perception.
  2. Product Postulation: Based on these categories, the team derived a list of 21 hypothetical products, such as a "Talking Typewriter" or a "Computer Psychiatrist."
  3. Expert Solicitation: Questionnaires were sent to members of the International Joint Artificial Intelligence Council and the IEEE.
  4. Iteration: Experts provided estimates on prototype dates, likelihood, and desirability. These responses were summarized and returned to the experts for a second round to sharpen the consensus.

Grounding the Forecast in Existing Technology

To provide a baseline for their hypothetical products, the researchers cataloged existing AI-related products and prototypes available in 1973. This included early biometric systems, industrial robots, and voice-operated systems.

Table 1: Examples of early AI-related products and their functions as identified in the 1973 study.

Key Insights and the "Toy World" Trap

A major hurdle identified by the authors was the "Toy World" trap. While AI techniques performed well in "limited domains"—such as the "block world" of polyhedral solids—the researchers questioned whether these methods could scale to complex, real-life databases. This led to the identification of the "Formalization Bottleneck": the realization that AI progress is limited not just by hardware, but by the difficulty of formalizing "common sense," "aesthetics," and "ethics" into algorithmic structures.

Results and Prescience

The expert panel proved remarkably prescient in several areas while remaining overly optimistic in others:

  • Biometrics: They correctly anticipated that "Automatic Identification Systems" (recognizing faces or fingerprints) would see prototypes by the mid-1970s.
  • Industrial Automation: Commercial viability for industrial robots was expected by 1980, a prediction that largely aligned with the rise of factory automation.
  • The Long Game: More complex systems, such as a "General Factotum" (a household servant robot), were pushed out to the year 2000 and beyond.
  • Human-Level Intelligence: The median response for when computers would exhibit "intelligence at adult human level" was roughly 20+ years away (post-1993), though a significant minority believed it would take over half a century.

Societal Impact and Ethical Oversight

The study flagged several products as "potentially dangerous to society," specifically highlighting Automated Intelligence Systems for government monitoring and Robot Tutors due to the potential for state-sponsored indoctrination. Furthermore, they predicted a psychological crisis as machines began to perform "uniquely human" activities, leading humans to question the uniqueness of their own capabilities.

The paper concluded with a forward-thinking recommendation: the creation of an AI Ethical Review Board. Similar to those in the medical profession, such a board would establish safety standards and legislation for autonomous systems—a call to action that remains highly relevant today.

Summary updated

BibTeX
@inproceedings{firschein1973forecasting,
  author = {Oscar Firschein and Martin A. Fischler and L. Stephen Coles and Jay M. Tenenbaum},
  title = {{Forecasting and Assessing the Impact of Artificial Intelligence on Society}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {1973},
}
Figure