A Grand Unified Theory of the AI Hype Cycle

🌈 Abstract

The article discusses the cyclical nature of the history of artificial intelligence (AI) research, where periods of hype and rapid progress are followed by periods of disappointment and stagnation, known as "AI winters". The author outlines the common patterns observed in these cycles, including the resource-intensive nature of AI research, the tendency to overpromise revolutionary possibilities, and the eventual rebranding of the technology under different names as its limitations become better understood.

🙋 Q&A

[01] The history of AI research

1. What are the common patterns observed in the cycles of AI research?

• AI research goes through cycles, where each cycle has a specific technology or approach (denoted as "N") that initially requires significantly more computing resources than the average available at the time.
• The increased funding and resources lead to immediate results, as the technology was previously resource-constrained.
• The revolutionary possibilities of "N" are often overhyped, leading to unreasonable expectations.
• As the limitations of "N" become more apparent, researchers and industry quietly stop calling their tools "AI" and start using more specific terms.
• Eventually, the funding for "N" dries up, leading to an "AI winter", after which researchers move on to the next approach or technology ("M").

2. How do the cycles of AI research compare over time?

• Each cycle has been larger and lasted longer than the previous one.
• The current hype cycle is unlike any that have come before, with more money involved and a more commercial focus, rather than being primarily driven by research funding.

3. What is the author's view on the current state of AI research?

• The author cannot predict when the current "mania" will end and the bubble will burst.
• The author states that computers cannot think, and the problems of the current "AI" will not all be solved within "5 to 20 years", as often promised.

[02] The author's perspective

1. What is the author's overall perspective on the history of AI research?

• The author acknowledges that each cycle of AI research has produced genuinely useful technology, but that the progress follows a sigmoid curve, not an exponential one.
• The author suggests that the initial promises of AI research often imply or state outright that pouring more resources into the technology will lead to endless improvements, when the reality is that these strategies inevitably have a limit.

2. What is the author's advice or message to the reader?

• The author cannot predict when the current "mania" will end, but they can tell the reader that computers cannot think, and the problems of the current "AI" will not all be solved within the promised timeframes.
• The author is available for consulting work if the reader's organization could benefit from expertise on topics like "what are we doing that history will condemn us for".