Today’s AI excitement feels unprecedented—every company racing to integrate large language models, billions in investment, and breathless predictions about transformation. But we’ve been here before. The current wave of AI enthusiasm isn’t the first time corporations have bet big on artificial intelligence to revolutionize their operations. Understanding what happened during the last major AI boom in the 1980s and 1990s—and the parallel Knowledge Management movement that promised to capture and scale organizational expertise—offers crucial lessons about both the promise and the pitfalls of transformative technology.

The 1980s: Extraordinary Investment and Grand Visions

The 1980s saw extraordinary corporate investment in AI, particularly expert systems and knowledge-based reasoning. Companies believed they could capture expert knowledge in rule-based systems. GE developed DELTA for locomotive repair diagnostics, reportedly saving millions annually. Digital Equipment Corporation built XCON to configure VAX computer systems, processing thousands of orders and becoming one of the most successful early expert systems. American Express created expert systems for credit authorization.

Case-Based Reasoning (CBR) emerged as a promising alternative—solving new problems by adapting solutions from similar past cases. Inference Corporation, Cognitive Systems Inc., and others built commercial CBR platforms for help desk support, legal research, medical diagnosis, and design assistance.

The vision was intoxicating: capture retiring experts’ knowledge, standardize decision-making, reduce training costs, and scale expertise globally. AI would fundamentally re-engineer corporate operations.

The Knowledge Management Movement (Late 1980s-1990s)

Knowledge management (KM) emerged with broader ambitions than AI, aiming to capture all organizational knowledge—documents, processes, lessons learned, and tacit knowledge. Companies like Lotus (Notes/Domino), Microsoft, and specialized vendors built platforms for knowledge repositories and collaboration.

KM recognized technology alone wasn’t enough, emphasizing communities of practice and knowledge-sharing cultures. Firms like McKinsey, Ernst & Young, and Accenture built massive internal KM systems to leverage knowledge across global practices.

The reality proved messy. Knowledge repositories became overstuffed “knowledge graveyards” with primitive search. People didn’t naturally document knowledge, and systems felt like extra work rather than enablers.

What Went Wrong: The AI Winter Returns

Technical Limitations: Expert systems were brittle—working well in narrow domains but failing catastrophically outside them. Knowledge acquisition took far longer and cost more than anticipated. As business rules changed, updating thousands of interconnected rules became unmanageable. CBR systems struggled with retrieval at scale and adapting cases to different situations. Symbolic AI couldn’t handle uncertainty or learn from data well.

Economic Reality: Development costs were astronomical—often millions per system—with hard-to-prove ROI. Specialized LISP machines became obsolete as PCs grew powerful. Many systems never left pilot projects or were abandoned when key champions departed.

The Hype Cycle: Vendors overpromised dramatically. When systems couldn’t deliver transformative results, disillusionment hit hard. Funding dried up in the late 1980s/early 1990s as companies recognized the gap between promise and reality.

Knowledge Management Challenges: The “if you build it, they will come” approach failed. Tacit knowledge proved much harder to capture than explicit knowledge. Knowledge quickly became outdated without good validation mechanisms. Search was too primitive for large repositories. Cultural resistance—knowledge hoarding for job security, “not invented here” syndrome, and lack of time—undermined adoption.

Changing Technology Landscape: The internet and web browsers in the mid-1990s shifted attention and resources. Data warehousing, business intelligence, and ERP systems offered more immediate, measurable value. The PC revolution made expensive, specialized AI systems seem anachronistic.

What Actually Worked

Not everything failed. Specific, narrow expert systems like XCON saved real money. Credit card fraud detection evolved from rule-based to hybrid approaches. Manufacturing diagnostics and scheduling systems succeeded in controlled environments. Cultural lessons about knowledge sharing influenced later collaboration tools. CBR found lasting niches in help desk systems and design reuse.

Legacy and Lessons

The 1980s-90s AI and KM wave left important legacies. Companies learned that technology without process change and cultural buy-in fails—lessons that informed later enterprise software implementations. Much of today’s AI renaissance builds on symbolic AI research from that era, now combined with machine learning and neural networks that learn patterns from data rather than requiring explicit programming.

The oversell created skepticism that persisted for decades. When modern AI emerged in the 2010s, there was initial wariness about “AI hype” precisely because of this history.

The goal of capturing and leveraging organizational knowledge remains valid. Today’s approaches—using machine learning, natural language processing, better search, and sophisticated knowledge graphs—are finally delivering on those old promises with fundamentally different technical approaches.

The early excitement faded because the gap between vision and capability was too large given 1980s-90s technology. Symbolic AI hit fundamental limits, knowledge engineering didn’t scale, and the economics didn’t work. But the problems those pioneers identified were real, and we’re now revisiting them with dramatically more powerful tools.