Artificial Intelligence in the Personal Computer Environment, Today and Tomorrow

Tom J. SchwartzIJCAI 1985

Abstract

In the mid-1980s, Artificial Intelligence (AI) began a significant migration from elite research laboratories and expensive mainframes to the burgeoning world of Personal Computers (PCs). This paper documents this transition, specifically focusing on "expert systems." It argues that the rapidly increasing computational power of PCs is democratizing AI, moving it from a "laboratory curiosity" to a strategic commercial tool applicable to diverse fields such as locomotive repair and financial services.

Introduction: The Democratization of AI

In the early 1980s, AI was synonymous with "Big Iron": massive mainframes or dedicated, expensive Lisp Machines from manufacturers like Lisp Machines Inc. or Symbolics. This created a high barrier to entry for businesses wanting to leverage Expert Systems—software designed to emulate human decision-making.

The "watershed" point arrived when machines like the IBM PC-XT and the early Macintosh reached sufficient computing power to support complex rule-based logic. This shift reduced development costs and allowed for widespread deployment, realizing Alan Kay’s "Dynabook" concept of portable, personal intelligence.

Market Survey and Methodology

The analysis is based on an extensive survey of the AI software ecosystem available for PCs in 1985. This includes implementations of Lisp, PROLOG, Smalltalk, and various "expert system shells." By 1981, the power of individual workstations had already begun to rival mainframes, setting the stage for a robust desktop AI market.

Table 1: Survey of AI software tools, pricing, and rule capacities available for PCs in 1985.

As shown in the survey data, the market was surprisingly mature by 1985. Pricing for these tools ranged from as little as $20 for hobbyist-level software to $15,000 for high-end professional systems. Rule capacities varied significantly, with some systems supporting up to 5,000 rules, while smaller "technician systems" focused on under 200 rules.

Technical Insights and System Capabilities

A key insight from this era is the rise of "small" expert systems that did not require a dedicated "knowledge engineer." Instead, users could build systems themselves using "induction extraction tools" that learned by example. Furthermore, the boundary between PCs and high-end workstations began to blur as desktop power increased.

Table 2: Technical features of 1985 AI tools, including external hooks and implementation comments.

Technical features such as "hooks to the outside world" allowed these systems to interact with other software and data sources, a precursor to modern integrated AI applications. Many tools utilized inductive learning, where rules were induced from examples (like spreadsheets), laying the groundwork for later logic-based machine learning.

Deployment Success and Results

The transition to PC-based AI yielded several notable successes:

  • GE’s Cats-1: A 1,200-rule system for diesel locomotive repair.
  • PUFF: A lung disease diagnostic tool, originally written in EMYCIN, which was successfully rewritten in BASIC to run on an Apple II.

These examples demonstrate that AI could serve as a "learning curve accelerator," disseminating routine expertise to less experienced staff and enhancing organizational consistency in decision-making.

Limitations and Future Outlook

Despite these advances, 1980s AI faced the "Common Sense" barrier—the inability of expert systems to handle general knowledge outside their narrow logical domains. Most systems were also limited to a few hundred rules, making them unsuitable for massive, complex environments.

However, the vision of the era—deploying AI on "hand-held calculator-like devices" using application-specific integrated circuits (ASICs)—has largely been realized today through mobile Neural Processing Units (NPUs). The next frontier remains the automation of knowledge acquisition, reducing the bottleneck of manual rule-coding and allowing every expert to become their own knowledge engineer.

Summary updated

BibTeX
@inproceedings{schwartz1985artificial,
  author = {Tom J. Schwartz},
  title = {{Artificial Intelligence in the Personal Computer Environment, Today and Tomorrow}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {1985},
}
Figure