By 1987, artificial intelligence (AI) was riding high on the success of expert systems, rule-based programs that simulated human decision-making. Companies were investing millions into AI research, and AI-powered automation seemed poised to revolutionize industries.
Then, almost overnight, the AI industry collapsed. Funding dried up, research slowed, and AI was once again seen as a failed technology. This period—known as the Second AI Winter—lasted through the 1990s, leading to widespread disillusionment with AI.
This article explores the causes of the Second AI Winter, its impact on AI research, and how AI eventually recovered.
What Was the Second AI Winter?
An AI Winter refers to a period when funding, research, and public interest in AI decline due to unmet expectations.
- The First AI Winter (1974–1980s) followed the Lighthill Report (1973) and funding cuts in neural networks.
- The Second AI Winter (1987–1990s) happened after expert systems failed to meet commercial expectations.
The Second AI Winter was even more devastating than the first, nearly shutting down AI research entirely.
The Rise of Expert Systems: AI’s First Commercial Boom (1980s)
Before the Second AI Winter, expert systems were seen as the future of AI.
✅ Businesses adopted AI-powered expert systems to automate decision-making.
✅ AI was generating billions of dollars in investments.
✅ Japan’s Fifth Generation Computer Project (1982) aimed to make AI-powered computers.
For the first time, AI was not just an academic experiment—it was seen as a practical business tool.
But by 1987, the hype came crashing down.
Causes of the Second AI Winter
1. Expert Systems Were Too Expensive to Maintain
- Knowledge bases required constant updating.
- Human experts had to program every new rule manually.
- Maintenance costs were higher than expected.
Companies realized that expert systems were not as efficient as originally promised.
2. Expert Systems Couldn’t Adapt to New Problems
- Unlike modern AI, expert systems didn’t learn—they only followed pre-programmed rules.
- Real-world problems were more complex than rule-based AI could handle.
For example:
❌ A medical AI system trained in one type of diagnosis would fail if symptoms didn’t match its rules.
❌ Business AI systems were inflexible and couldn’t adjust to new market conditions.
3. The Fall of the Fifth Generation Computer Project (Japan, 1982–1992)
- Japan had invested billions in AI-powered computing, hoping to create AI-driven personal computers.
- By the late 1980s, the project failed to deliver breakthroughs, leading to global skepticism about AI.
4. Economic Recession and Shift in Priorities
- The stock market crash of 1987 forced companies to cut spending.
- AI was seen as a luxury, and businesses prioritized traditional computing over AI research.
5. The Rise of Classical Computing Methods
- In the 1990s, database-driven software and traditional computing were cheaper and more reliable than AI.
- Companies preferred rule-based automation and statistics over uncertain AI models.
As a result, AI companies collapsed, research slowed, and AI funding dried up.
Immediate Effects of the Second AI Winter
🚨 Massive AI Layoffs
- AI startups went bankrupt, and thousands of AI researchers lost jobs.
- Universities cut AI funding, shifting focus to computer science and data processing.
🚨 Decline of Expert Systems
- Companies abandoned expert systems in favor of database-driven automation.
- AI was seen as impractical, and many believed AI research would never recover.
🚨 AI Becomes a “Dirty Word”
- Investors and governments avoided AI funding, considering it a failed technology.
- The term “artificial intelligence” disappeared from major tech conferences for nearly a decade.
For the second time in 20 years, AI research had collapsed—and this time, it seemed AI was finished for good.
How AI Recovered From the Second AI Winter
Although AI research was nearly abandoned, some researchers continued working in the shadows, laying the foundation for AI’s comeback in the 2000s.
✅ Machine Learning Gained Interest (1990s)
- Researchers explored new AI models, including genetic algorithms and Bayesian networks.
- Statistical AI (data-driven learning) replaced rule-based expert systems.
✅ Neural Networks Were Revived (1990s–2000s)
- In 1986, Geoffrey Hinton had revived neural networks with backpropagation.
- By the late 1990s, computer hardware was powerful enough to train larger networks.
✅ Big Data and the Internet (2000s)
- The rise of the internet and cloud computing provided large datasets for AI training.
- Companies like Google, Amazon, and Microsoft began experimenting with AI-powered search and automation.
✅ Deep Learning Breakthroughs (2010s)
- Convolutional Neural Networks (CNNs) revolutionized computer vision (e.g., facial recognition, self-driving cars).
- Recurrent Neural Networks (RNNs) improved speech recognition and language processing.
- AI-powered assistants like Siri, Alexa, and Google Assistant emerged.
By the 2010s, AI had fully recovered, leading to the current AI revolution we see today.
Lessons from the Second AI Winter
The Second AI Winter (1987–2000s) taught valuable lessons:
1. AI Needs Real-World Use Cases
- The failure of expert systems showed that AI must solve real problems—not just theoretical ones.
- Modern AI focuses on practical applications like speech recognition, autonomous vehicles, and medical diagnosis.
2. AI Must Adapt and Learn
- Expert systems failed because they couldn’t learn.
- Today’s AI models use deep learning and self-improving algorithms to handle complex, changing environments.
3. Computational Power Matters
- AI in the 1980s was limited by slow computers.
- Modern AI thrives due to fast GPUs, cloud computing, and massive datasets.
4. Hype Can Be Dangerous
- AI’s downfall in the 1980s was caused by overpromising and underdelivering.
- Today’s AI researchers set realistic expectations, balancing optimism with caution.
How the Second AI Winter Shaped Modern AI
The Second AI Winter (1987–2000s) was a dark period for artificial intelligence, nearly destroying the field.
❌ AI research funding collapsed.
❌ Expert systems failed to meet expectations.
❌ AI companies went bankrupt, and interest in AI disappeared.
However, AI did not die. Instead, researchers used this period to refocus on machine learning, neural networks, and data-driven AI.
Today, AI is more powerful than ever, thanks to the lessons learned from past failures. From self-driving cars to ChatGPT, modern AI avoids the mistakes of the past, ensuring that we never enter another AI Winter again.