Bubbles and the Invention-Imagination Gap | How to Read AI Usage Studies | A Conversation with De Kai

We’ve entered the Bubble Prediction phase of AI. Some see valuations floating safely forward; others warn of collapse. The real challenge isn’t forecasting the burst—it’s imagining the inventions that could justify today’s bets.

Bubbles are not new. From biotech in the 80s to the internet in the 90s to cleantech in the 00s, valuations raced ahead of reality: some companies failed, others endured. I had a front-row seat as an equity research analyst during the internet and cleantech bubbles. 

Here are 3 lessons that I think might be useful during today’s AI bubble. 

• Ignore Momentary Momentum: Valuation swings make news. NVDA nearly doubled between April and August. Does that mean the market truly believes its long-term prospects justify a $4 trillion valuation, or 50 times earnings? Maybe. But it might just be short-term traders riding momentum. Volatility makes headlines but says little about fundamentals. Sometimes momentum has little to do with the company at all. In 2007, the valuation of First Solar, cleantech’s poster child and best performing stock in the market that year, plunged by 30% at year end as hedge funds locked in bonuses. Expect the same kind of year-end games in AI stocks—noise, not signal.
• Think Portfolio: In every bubble, many companies disappear. A few—sometimes the ones that looked shaky at the time—end up defining entire industries: Genentech vs. Genex, Amazon vs. Pets.com, First Solar vs. Solyndra. Today's AI landscape shows similar patterns—while OpenAI and Anthropic grab headlines, the ultimate winners might be companies we're barely watching, perhaps those solving unglamorous infrastructure problems or finding unexpected applications in narrow domains. Or just moving slower, being more cautious, solving hard problems first (looking at you, Apple). Investors hedge this risk with baskets. They may crash—as in 2000—but over time, it delivers outsize returns. Don’t let a single company’s prospects blind you to the potential—or risk—of the overall industry. 
• Watch the Invention-Imagination Gap: Perhaps the most important lesson to apply to today’s AI industry is to imagine invention. Critics dismiss AI because today’s products aren’t impressive—and they’re right, for now. That was also the case with internet companies in the 90s when web browsing wasn’t secure (until VeriSign), when an online bookstore seemed ridiculously small and expensive (until Amazon vastly expanded and created its fulfillment infrastructure), and when relying on someone being in front of a computer seemed too limiting (until the smartphone). The gap? Invention imagination. The hardest part of bubbles isn’t imagining the crash, but the invention. Just as Amazon solved fulfillment logistics in ways that weren't obvious when it was 'just a bookstore,' AI companies today might solve cost and energy challenges through approaches we can't yet see—perhaps through dramatic efficiency gains, edge computing, or hybrid cloud-device architectures. Be skeptical of critics who insist today’s AI economics are fixed, without imagining the inventions that could change them.

Every bubble exaggerates the present and underestimates the future. Biotech, the internet, and cleantech all stumbled through frenzy and collapse, yet each remade the world in ways critics never imagined. AI will likely do the same. What matters now is separating spectacle from substance—and imagining the inventions that will bridge the gap.

How to Read AI Usage Studies: A Guide to Three Different Lenses

As a research institute that studies the human experience of AI, it's no wonder that we're interested in AI usage studies. In this essay, Helen provides a comprehensive analysis of three recent studies Anthropic's Economic Index Report, OpenAI's How People Use ChatGPT, and our own Chronicle project: How We Think and Live with AI: Early Patterns of Human Adaptation. Each study has its advantages and disadvantages, value and gaps.

The Three Research Approaches:

• Anthropic's Economic Lens: Focuses on tasks and productivity using millions of Claude conversations, finding heavy enterprise automation patterns
• OpenAI's Behavioral Lens: Examines user intent across 700+ million ChatGPT users, discovering 70% of usage is non-work related
• Artificiality Institute's Psychological Lens: Studies how AI relationships affect identity and thinking patterns through workshop observations

Key Insights:

• The same AI interaction can simultaneously serve as a productivity tool, creative partner, and psychological relationship
• Context shapes everything—different user populations (technical vs. consumer) naturally produce different usage patterns
• Large-scale behavioral studies miss crucial psychological factors that determine whether AI feels empowering or creates dependency
• Individual adaptation varies dramatically even with identical AI usage patterns

Bottom Line: No single study captures the full complexity of human-AI interaction. Understanding each methodology's strengths and blind spots helps you apply research insights more effectively to your specific context and make better decisions about AI integration.

Read more...

De Kai: Raising AI

In this conversation, we talk with De Kai, a professor, pioneering AI researcher, and author of Raising AI. Drawing on insights from developmental psychology and complex systems, De Kai's "Raising AI" framework emphasizes conscious human responsibility in shaping how these artificial minds develop. Rather than viewing this as an overwhelming burden, he frames it as an opportunity for humans to become more intentional about the values and behaviors they model—both for AI systems and for each other.

Listen/Watch here...