Anthropic Economic Index, Virtual Agent Economies, AlterEgo and How People Use ChatGPT

A recent NBER paper, “How People Use ChatGPT,” provides one of the first academic breakdowns of user behavior, revealing trends that both confirm and challenge popular narratives about generative AI. The discussion also explores Anthropic’s latest economic index on global AI adoption, a DeepMind paper on agent economies, and the competing visions for AI-powered wearables.

How People Use ChatGPT

The most significant finding from the NBER paper is that the primary use for ChatGPT is information gathering, effectively making it a “Search GPT 2.0.” This finding suggests that, for now, the technology is being used more as an enhancement to an existing paradigm (search) rather than as a completely new category of tool.

However, the true economic impact may not be in the standalone chat interface but in how the technology is embedded via APIs into other software and platforms.

Key Usage Patterns and Surprises

• Writing vs. Content Generation: In education, 50% of the use is for “writing.” However, a closer look reveals that two-thirds of this activity is not generating new content, but rather editing, critiquing, and translating existing text. This highlights its power as a sophisticated editorial tool.
• A Tech Bubble?: The data indicates that some of the most hyped use cases are surprisingly niche among the general user base. Coding accounts for only 4% of messages, with personal reflection/therapy at less than 2% and gaming at less than 0.5%. This suggests a disconnect between the focus of the tech community and mass-market adoption.
• Worker vs. Non-Worker Profiles: An interesting shift has occurred from early professional adopters to a majority (70%) of “non-workers.”
  • Workers use it for knowledge-intensive tasks, writing, and programming.
  • Non-workers use it for everyday life tasks and creative pursuits like generating images and art.
  • Together, information seeking, practical guidance, and writing tasks account for approximately 80% of all usage.

The conversation suggests we may be in the early days of discovering the technology’s “killer app,” similar to how the spreadsheet unlocked the PC’s potential beyond simple tasks like storing recipes. The current chat interface may be limiting the scope of user imagination.

Anthropic’s Economic Index and the Global AI Gap

Anthropic’s second Economic Index expands its analysis beyond the US, revealing a strong correlation between a country’s GDP and the adoption of AI tools like Claude. This raises concerns about a widening “AI gap,” where wealthier nations reap the benefits of AI while lower-income countries are left behind.

However, the report also offers a more nuanced view:

• Usage Density: By normalizing for the working-age population, the index shows that smaller, tech-advanced nations can lead in per-person usage.
• Regional Utility: Different tools find different product-market fit in various regions. For example, in India, half of all Claude usage is for coding tasks, a stark contrast to the general trend seen with ChatGPT. Outliers like Singapore and Canada are also significantly “over-indexed,” using the tool far more than their GDP would predict.
• The Open Source Factor: The data is for a paid, proprietary model. The landscape of AI adoption might look entirely different if it accounted for free, open-source alternatives, which are highly popular in global developer communities.

The situation mirrors the adoption curve of social media a decade ago. The key challenge now is to bridge the access gap ethically, avoiding the creation of “information monopolies” under the guise of providing free, curated access.

The Dawn of Agent Economies

A speculative paper from DeepMind, “Agent Economies,” explores the future implications of AI agents interacting with each other in the market. This scenario introduces novel risks, as seen in algorithmic trading or the current “algorithmic war” in job recruiting, where AI-generated applications are pitted against AI-powered filters.

From Agent Companies to Economies

While the concept of a fully autonomous agent economy seems distant, the intermediate step of “agent companies” is already emerging in open-source projects:

• MetaGPT: A project that simulates an entire software startup with agents fulfilling roles for product management, engineering, and deployment.
• AI Scientist: A team of agents capable of formulating a hypothesis, designing and running an experiment, and writing a research paper. One of its generated papers was successfully accepted into the ICLR conference.

A key research question is how to make these agent-driven systems “steerable.” However, skepticism remains, as even heavily regulated human-driven markets, like finance, are difficult to steer, often relying on crude “circuit breakers” during crises. The future of agent interactions may depend on designing the right systems for fairness, credit assignment, and incentives.

The Future of AI Wearables: AlterEgo vs. Meta

Two competing visions for the future of wearable AI have emerged:

1. AlterEgo (Invisible AI): A prototype device worn behind the ear that reads neuromuscular signals from the face and throat, allowing a user to communicate silently and internally with an AI. It represents a new, non-vocal communication plane.
2. Meta Ray-Ban Glasses (Visible Interface): An augmented reality approach that overlays information, such as language translations or notes, onto a user’s field of view. It focuses on making existing AI interactions more convenient.

While the technology is impressive, the primary challenge remains finding the daily-use killer application. Demos showcasing identifying architectural styles or texting in a noisy room feel niche. The most compelling and valuable use case for a technology like AlterEgo is its potential to empower individuals with speech impediments or motor neuron diseases, providing them with an invaluable communication tool.

• Tags:
• Generative AI
• ChatGPT
• Large Language Models
• DeepMind
• Anthropic
• Agent Economies
• AI Wearables