AI, networks and Mechanical Turks (2025) (www.ben-evans.com)

A new perspective on the intersection of AI, user behavior, and recommendation systems was presented in a recent analysis, highlighting the enduring challenge of the cold start problem in machine learning. As established platforms like Amazon and Google leverage user data to inform their recommendations, they remain limited by their inability to understand the deeper reasons behind user actions. While they can identify correlations, such as buying packing tape and bubble wrap, they fail to connect these insights to broader contexts, like moving house. The analysis emphasizes that large language models (LLMs) can transform this landscape by offering a more nuanced understanding of both what products and content users engage with and why, enabling richer correlations that were previously unattainable. This advancement could allow companies to bypass the traditional user acquisition hurdles by leveraging general-purpose LLM APIs that draw from collective knowledge rather than relying solely on in-house data. As businesses explore new strategies for gathering insights—like using camera roll analysis for user preferences—they may harness the potential of LLMs to create a holistic view of user behavior. This shift has profound implications for AI/ML innovation, as it promises not only to enhance recommendation accuracy but also to offer entirely new methods for addressing the saturation of choices in today’s digital marketplace. While the future remains uncertain and reminiscent of the early days of the internet, it signals a crucial evolution in how AI can reshape user interactions and product discovery.