AI 2.0: Introducing Network Effects to the AI Era

🌈 Abstract

The article discusses the current state of the AI industry, focusing on the shift from foundational models to AI applications. It analyzes the potential sources of defensibility for AI companies, including economies of scale, switching costs, and network effects. The article categorizes AI applications into three broad groups - AI 1.0, AI 1.5, and AI 2.0 - based on the nature of user interactions and the potential for network effects.

🙋 Q&A

[01] AI Applications and Defensibility

1. What are the key differences between traditional software companies and AI companies in terms of marginal costs? Traditional software companies had zero marginal costs to produce additional units, while AI companies have non-zero marginal costs due to the steep costs involved in training and running large language models.

2. Why are economies of scale not a viable mechanism for defensibility at the AI application layer? Computational costs are an order of magnitude lower at the application layer compared to the model layer, so the ability to pay for APIs or compute is not a sustainable advantage over future competitors.

3. What are the two main sources of defensibility for most AI applications? The two main sources of defensibility for most AI applications are network effects and switching costs.

[02] AI 1.0: Single-Player AI Applications

1. Why is it not possible for network effects to exist in single-player AI applications? Network effects require a multiplayer interaction, but single-player AI applications involve a direct interaction between the user and the AI product, which is a single-player interaction.

2. What is the author's assessment of the sustainability of single-player AI applications? The author suggests that single-player AI applications are prone to rapid commoditization, with some potentially going viral due to the novelty of AI, but this is rarely sustainable.

[03] AI 1.5: AI-Generated Content Platforms

1. How do network effects theoretically exist in AI-generated content platforms? User adoption leads to more AI-generated content, which other users can then interact with. However, the author argues that the low friction for creating supply and the fact that the content is known to be AI-generated limit the strength of these network effects.

2. What are the potential sources of switching costs in AI-generated content platforms? The author suggests that users could form emotional bonds with the AI-generated content, such as AI chatbots, which could introduce psychological switching costs. However, the author notes that there is no clear evidence of this so far.

[04] AI 2.0: AI-Enabled Marketplaces

1. What are the key characteristics of AI 2.0 applications that enable meaningful network effects? AI 2.0 applications use AI to enable higher-friction multiplayer interactions that were previously impossible. The example given is Haz, which uses AI to create a social feed based on users' past purchases, enabling interactions between users.

2. How does the transition from Web 1.0 to Web 2.0 relate to the potential for AI 2.0 applications to create network effects? The author draws a parallel between the transition from Web 1.0 (information access) to Web 2.0 (content creation and consumption) and the potential for AI 2.0 applications to enable a similar expansion in AI applications with network effects.