The prevailing discourse around artificial intelligence often frames it as an autonomous force, prompting anxieties about "what AI will do to us." However, Erik Brynjolfsson, a leading economist and director of the Stanford Digital Economy Lab, challenges this passive perspective, asserting that the critical question is, "What will we do with AI?" In his view, the true impediments to harnessing AI’s potential for widespread prosperity are not technological limitations but rather the inertia of human organizations, institutions, and our collective choices. Drawing on extensive research into AI’s profound impact on employment, productivity, and economic growth, Brynjolfsson champions the idea that the future is not preordained; it is a canvas we are actively painting through the decisions we make today. This perspective underscores human agency in an era of unprecedented technological capability, suggesting that the most transformative breakthroughs may reside less in algorithms and more in our adaptive capacity and intentional application.
Brynjolfsson, whose influential work like The Second Machine Age and Race Against the Machine has significantly shaped the debate on technology and the economy, emphasizes that while AI’s capabilities are advancing at an astonishing pace, our understanding of its economic implications and the necessary institutional adaptations lag far behind. His tenure at the Stanford Digital Economy Lab is dedicated to bridging this critical gap. He observes that early predictions about the speed of certain AI applications, such as fully autonomous vehicles, proved overly optimistic, taking over a decade longer than initially anticipated to reach a practical, albeit still cautious, level of deployment. Conversely, the rapid evolution of large language models (LLMs) like ChatGPT, Gemini, and Claude—which he concedes would have been considered Artificial General Intelligence (AGI) a decade ago—has far outstripped expectations. Yet, this rapid technological progress has not translated into a commensurate surge in overall productivity growth, which remains stubbornly flat. This disconnect highlights a fundamental challenge: the translation of raw technological power into tangible economic benefits requires substantial complementary investments and organizational reinvention.
This phenomenon is best understood through the "J-curve" of productivity, a concept Brynjolfsson frequently employs. Powerful general-purpose technologies (GPTs), such such as the steam engine, electricity, or the internal combustion engine, initiate a period where initial investments in new infrastructure, skills, and organizational redesign are significant and costly, while measurable output or productivity gains remain elusive or even decline. Historically, the electrification of factories in the early 20th century serves as a prime example. It took 20 to 30 years for businesses to fully re-architect their production processes, moving from a centralized power source to distributed electric motors, before productivity skyrocketed. During this prolonged period of costly experimentation and adaptation, conventional productivity metrics barely registered an improvement. Brynjolfsson posits that AI, as arguably the most impactful GPT of our era, is currently situated within this downward or flat segment of the J-curve. The essential "co-inventions" of new business processes, reskilled workforces, and entirely new business models are difficult, expensive, and time-consuming. Investments in these intangible assets, he notes, can be ten times greater than the direct capital expenditure on AI technology itself, yet they are often overlooked or poorly measured. Accelerating the upward slope of this J-curve demands a proactive and systematic approach to organizational transformation and talent development.
The "Canaries in the Coal Mine" research, conducted by Brynjolfsson’s team using granular ADP payroll data and a detailed taxonomy of AI exposure developed by researchers at OpenAI and the University of Pennsylvania, offers a more nuanced view of AI’s immediate impact on the labor market. While aggregate employment figures showed little change, a deeper dive revealed striking disparities. Workers aged 22 to 25 in occupations highly exposed to large language models experienced a significant employment decline, initially around 12-13% and later stabilizing at 16-17%, relative to their peers in less exposed roles. This effect was largely absent among older workers. Crucially, the study differentiated between AI used for automation (eliminating tasks) and AI used for augmentation (creating new skills and opportunities). Occupations where AI primarily automated tasks saw employment fall, whereas those embracing AI for augmentation experienced employment growth alongside increased productivity. This finding powerfully reinforces Brynjolfsson’s core message: the outcome is determined by how we choose to deploy the technology.
This distinction between automation and augmentation ties into the economic principle of demand elasticity and what Brynjolfsson terms the "Turing Trap." The "Turing Trap" refers to the narrow objective of making machines perfectly imitate human capabilities, as embodied by the Turing Test. While cost-cutting through automation is a valid business goal, an overemphasis on it can limit AI’s transformative potential. Instead, he advocates for a shift in objectives towards value creation, the development of entirely new products and services, enhanced customer satisfaction, and improved quality of work life. Using the analogy of demand curves, he explains that for goods with inelastic demand (like apples), making them cheaper might only slightly increase consumption, leading to reduced overall spending. However, for goods with elastic demand (like jet travel), a significant price reduction can unlock vast new markets and exponentially increase demand. AI, by making certain capabilities cheaper and more accessible, has the potential to trigger elastic demand for new services and creative endeavors, fostering growth rather than mere displacement. The challenge, therefore, is to identify and cultivate these areas of elastic demand where human creativity, amplified by AI, can generate unprecedented value.
The landscape of AI research and development has also undergone a significant shift, with much of the frontier innovation now originating from industry rather than academia. This is largely due to the astronomical costs associated with training the largest AI models, which can run into billions or even tens of billions of dollars—sums far beyond the reach of most universities or government-funded research institutions. While this concentration of power and resources in a few tech giants raises concerns, Brynjolfsson argues that academia retains a crucial role. Universities and public funding bodies are vital for supporting fundamental, "basic" research that may not have immediate commercial viability but often lays the groundwork for future breakthroughs. He cites Geoffrey Hinton, a pioneer in deep learning, who, when asked about his hardware, pointed to his laptop, emphasizing that conceptual breakthroughs don’t always require massive computing power. The symbiotic relationship between expensive, applied industrial research and curiosity-driven, fundamental academic inquiry is essential for long-term progress and societal benefit.
Ultimately, Brynjolfsson identifies himself as a "mindful optimist." This isn’t blind faith that everything will simply work out, nor is it passive pessimism. Instead, it’s an active, intentional optimism rooted in the belief that humans possess the agency to shape a desirable future. AI, he posits, acts as an "amplified intention"—a powerful tool that magnifies whatever goals and values we choose to embed within its application. With this amplified power comes a greater responsibility to consciously consider the kind of economy we wish to build, one that balances immense wealth creation with shared prosperity. The potential for AI to exacerbate wealth and power concentration is a significant concern, requiring careful attention to economic systems and governance structures. Just as different institutional choices lead to vastly different outcomes between nations, so too will our collective decisions determine whether AI leads to a future of broad flourishing or deepening inequality. This calls for an interdisciplinary dialogue, engaging not just technologists but also philosophers, humanists, economists, and policymakers, to deliberately design a future where AI serves humanity’s highest aspirations. The choices made today, Brynjolfsson asserts, will define the radically transformed world of tomorrow.
