"ChatGPT is really search GPT." This succinct observation from Lauren McHugh Olende, Program Director for AI Open Innovation, encapsulates a pivotal insight from the recent "How People Use ChatGPT" paper. The report, a gold standard working paper from the National Bureau of Economic Research, reveals that the primary application for OpenAI's flagship tool is, in essence, information retrieval. This finding, alongside discussions on Anthropic's Economic Index, the rise of virtual agent economies, and cutting-edge AI wearables like AlterEgo, formed the core of a recent "Mixture of Experts" podcast. Host Tim Hwang, joined by McHugh Olende, IBM Fellow and Master Inventor Aaron Baughman, and Master Inventor Martin Keen, delved into the multifaceted implications of AI's rapid integration into society.
The NBER paper's granular analysis of ChatGPT usage patterns presents a landscape both expected and surprising. Initially, professionals were the primary adopters, leveraging the tool for specialized tasks. However, a significant shift has occurred, with Aaron Baughman noting that "70% of usage are now what we call non-workers." For professionals, AI aids in finding information, offering practical guidance, and assisting with writing. Yet, Martin Keen highlighted a crucial distinction within the "writing" category, clarifying that two-thirds of this use is dedicated to editing, critiquing, or translating existing content, rather than generating entirely new material. This nuance challenges the perception of generative AI as a purely creative engine, underscoring its current role as an augmentation tool for human-driven tasks.
This realization begs the question of whether AI is merely an advanced form of existing technology or a truly disruptive force. Tim Hwang pondered if the prevalence of "search" as the main use case was a "disappointing outcome," drawing parallels to early personal computers being used for simple recipe storage before the advent of spreadsheets unlocked their full potential. Lauren McHugh Olende attributed this current focus to an "imagination gap" regarding AI's broader capabilities. She emphasized that the true economic impact might not stem from standalone consumer tools, but from AI embedded via APIs into the everyday software we already use.
The conversation broadened with Anthropic's second Economic Index, which expanded its analysis beyond the United States. Tim Hwang questioned if a correlation between wealthier nations and higher Claude adoption signaled an "AI gap." Aaron Baughman affirmed that technologically advanced countries do indeed lead in "usage density" per capita. However, he expressed optimism, noting that individuals in some lower-income regions are "learning to work with AI," potentially bridging this divide through remote education and the identification of niche applications. The emphasis shifted from raw access to the practical "utility" and "actionability" derived from these tools.
The discussion then veered into the speculative realm of "virtual agent economies," inspired by a DeepMind paper. The concept revolves around autonomous AI agents interacting and performing tasks on behalf of humans, potentially leading to a rise of "solopreneurs" who leverage these digital entities. Aaron Baughman distinguished between "assistants" (performing direct tasks) and "agents" (working autonomously in the background), suggesting the latter represents the future. Yet, this vision also introduces new complexities, with Martin Keen humorously questioning if such an interconnected system could devolve into a giant "echo chamber" of self-referential AI. He also raised concerns about the "amount of processing and complexity" required for agents to communicate effectively, and the potential for meaning to "get lost" as information flows through multiple automated layers. The financial markets, already heavily reliant on algorithmic trading, serve as a cautionary tale of markets becoming difficult to "steer" when automation reigns.
The final segment explored "AlterEgo," a prototype for invisible AI interaction, and Meta's new Ray-Ban smart glasses. The AlterEgo demo showcased a subtle, ear-worn device that interprets "inner speech" (subvocalization) to interact with AI without any visible interface. Meta's Ray-Ban glasses, on the other hand, integrate AI capabilities into a familiar consumer product, allowing for features like live translation captions during conversations. Aaron Baughman highlighted the potential of such "non-invasive" interfaces, drawing on his background in biometrics and emphasizing that "the less physically intrusive, the better." He envisions a future where AI is seamlessly integrated into everyday objects, becoming an invisible companion rather than a clunky device.
The podcast concluded without definitive answers, underscoring the dynamic and uncertain nature of AI's evolution. The insights revealed a technology simultaneously enhancing existing human endeavors, forging new pathways for entrepreneurship, and presenting complex ethical and societal challenges related to access, utility, and autonomous operation. The future of AI, it seems, will be shaped not just by technological advancement, but by our collective imagination and deliberate choices in navigating its profound impact.
