Why Everyone Is Wrong About AI (Including You) | Benedict Evans

Key Points

• AI is a major platform shift, comparable to the iPhone, but not necessarily larger than past shifts like the internet or mobile.
• The impact on employment, economy, and productivity will likely mirror previous platform shifts, with new questions emerging over a 10–15 year period.
• Historical examples (e.g., internet, mobile, automatic elevators) show that new technologies often seem strange and unpredictable, and incumbents often fail to adapt.
• Data is not a decisive advantage in AI because models require vast amounts of generalized text, which is widely available to competitors.
• Incumbents like Google face a reset in user behavior and product definitions, but their future is uncertain due to internal conflicts and market dynamics.
• Claims about AI becoming autonomous or threatening are often exaggerated or misinterpreted, as they result from specific prompts rather than inherent capabilities.
• Regulation of AI should be approached cautiously, with attention to trade-offs, rather than treated as a monolithic issue.

Summary

In this transcription, technology analyst Benedict Evans offers a nuanced perspective on artificial intelligence, positioning it as a significant but not unprecedented platform shift. He argues that while AI is the biggest development since the iPhone, it is not necessarily more transformative than earlier shifts such as the internet or mobile computing. Evans emphasizes that every platform shift initially seems strange and unpredictable, and the current excitement around AI mirrors past hype cycles, including the dot-com bubble. He cautions against assuming that this time is fundamentally different, noting that historical patterns often repeat, albeit with new specifics.

Evans uses historical analogies to illustrate how new technologies evolve. He recalls the early uncertainty around the internet—whether it would be centralized like the "information superhighway" or decentralized like the web—and how mobile internet took a decade to mature, eventually replacing the PC as the central tech platform. He also cites the example of automatic elevators, which were once a novel innovation but are now taken for granted. These examples highlight that while AI will bring profound changes, its full impact and business models will only become clear over time.

Regarding incumbents, Evans suggests that they often try to absorb new technologies as features rather than fully adopting them, which can lead to failure. He references Kodak, which embraced digital cameras but was ultimately undone by the rise of smartphone photography and social media, not by ignoring innovation. For Google, the threat is a "reset" of the playing field: users may no longer default to Google for search, and the product itself is being redefined. However, Evans downplays the idea that proprietary data gives incumbents a decisive advantage, because training large language models requires vast amounts of generalized text that is equally accessible to anyone with sufficient resources.

Evans also addresses exaggerated claims about AI autonomy, such as studies suggesting AI might threaten blackmail. He argues that these are often the result of prompting a text generator to produce specific narratives, akin to writing "murder is good" on a piece of paper and blaming the photocopier. He advocates for careful interpretation of such headlines. On regulation, he advises against treating AI as a monolithic category, instead emphasizing trade-offs and the need for context-specific rules, similar to how cars and spreadsheets are regulated. Finally, when asked how a country like the United States should prepare for AI, Evans refrains from giving a simple answer, hinting at the complexity of competing for dominance in a rapidly evolving field.