Artificial intelligence is often discussed in terms of faster economic growth, higher productivity, and greater corporate profits. But those are not the only outcomes that matter. Another very important question is whether AI will actually improve human welfare by broadening opportunity, strengthening institutions, and raising living standards across society.
That question is at the heart of “From Wealth to Welfare: Systemic Complementarities for the Age of AI,” a recent paper by MIT’s Isabella Loaiza and Roberto Rigobon.
“Over the past century, general-purpose technologies (GPTs), such as electricity, the combustion engine, and the Internet, have played a central role in driving global welfare,” they wrote. “They helped extend life expectancy and civil and labor rights and reduce infant mortality and poverty. However, these transformations did not happen based on the GPTs’ technical capabilities alone. We unlocked their potential by introducing necessary system-level complements into our social and economic systems.”
Artificial intelligence is the next great general-purpose technology. But, like its predecessors, whether it ultimately enhances welfare for all — or primarily creates wealth for a few — will depend less on the technology itself than on the complementary investments society chooses to make.
AI is indeed a classic general-purpose technology. As Erik Brynjolfsson, Daniel Rock, and Chad Syverson explained in their influential 2018 paper “The Productivity J-Curve,” GPTs “are the defining technologies of their times and can radically change the economic environment. They have great potential from the outset, but realizing that potential requires larger intangible and often unmeasured investments and a fundamental rethinking of the organization of production itself.”
The authors called this phenomenon the Productivity J-Curve. Like the letter “J,” productivity often declines during the initial investment phase before rising sharply once organizations have restructured themselves to take advantage of the new technology.
History provides many examples. Electricity, automobiles, computers, and the Internet all followed this same pattern. Even after they reached commercial viability, it often took decades before businesses and society learned how to reorganize around them and fully realize their benefits.
For example, productivity growth did not accelerate until roughly forty years after electricity was introduced in the early 1880s. Companies first had to redesign factories around electric motors and develop innovations such as the assembly line. Likewise, although the Internet began as a research network in the 1970s, it was not until the late 1990s and early 2000s that it became widely deployed across the economy and fundamentally transformed business models.
“Today, artificial intelligence (AI) is the rising GPT on the horizon, promising to reshape how we live and work,” wrote Loaiza and Rigobon. “However, like its predecessors, AI’s welfare-generating power is not promised without intention. If we want AI to enhance welfare rather than merely generate wealth, we must invest in the appropriate system-level complements to catalyze a shift toward our collective goals.”
The authors define system-level complements as “the social institutions, cultural norms, capital investments, and human and scientific resources required to realize the welfare benefits of a GPT.”
To illustrate the concept, they compare two very different technologies: the automobile and the Internet.
When automobiles were first introduced, they were slower, less reliable, and harder to operate than horses. Their broader societal value emerged only after governments and businesses invested in roads, traffic laws, driver education, safety standards, and liability insurance. Those complementary investments transformed automobiles from a technological novelty into a cornerstone of modern life.
The Internet followed a somewhat different path. Beginning as ARPANET in 1969 and later evolving into NSFNET, it served primarily researchers and universities for nearly two decades. Then, during the 1990s, the Internet and the World Wide Web began their transformation into the global commercial platform we know today.
That transformation also exposed important inequalities. Access to the digital economy required a personal computer, an Internet connection, and often a bank account and credit card. Those without access found themselves increasingly excluded from educational resources, employment opportunities, commerce, and information, giving rise to what became known as the digital divide.
Over time, however, technological progress dramatically lowered those barriers. Affordable smartphones, broadband wireless networks, cloud computing, and inexpensive mobile applications extended the benefits of the Internet to billions of people around the world.
According to the latest statistics from the International Telecommunication Union (ITU), approximately six billion people—almost three-quarters of the world’s population—were online in 2025, while 2.2 billion remained offline. Internet penetration now ranges from 88% to 94% in high-income countries but remains only 16% to 23% in low-income countries.
“The Internet, while transformative, demonstrates the consequences of underinvestment in these complements,” wrote Loaiza and Rigobon. “It has suffered from insufficient development of privacy and data governance norms, public data infrastructure, and equitable access policies. Free and universal Internet access, perhaps the closest digital equivalent of public roads, remains uncommon.”
“Artificial intelligence now faces a similar crossroads, presenting us with a choice: invest in the system-level complements necessary for inclusive progress or repeat these historical oversights.”
This observation leads to the paper’s central contribution: identifying five categories of complementary investments that will largely determine whether AI becomes a technology that broadly enhances human welfare rather than simply increases wealth.
1 — Complementary Capital. Like earlier GPTs, AI requires investments in both physical and intangible infrastructure.
Cars needed roads and gas stations. Likewise, “The Productivity J-Curve” emphasizes that realizing AI’s full potential requires large investments in organizational capabilities such as redesigned business processes, workforce development, competitive strategies, and new products and services.
“For AI,” the authors write, “the question remains open: What mix of public and private investment should fund it and its infrastructure?”
2 — Complementary Human Development. Technologies only scale when people know how to use them safely and effectively.
Cars required driver education. The Internet required digital literacy. AI requires a new wave of human development centered on capabilities that complement rather than compete with machines.
In their earlier paper, “The EPOCH of AI: Human-Machine Complementarities at Work,” Loaiza and Rigobon identified five uniquely human capability groups summarized by the acronym EPOCH:
- Empathy and Emotional Intelligence
- Presence, Networking, and Connectedness
- Opinion, Judgment, and Ethics
- Creativity and Imagination
- H0pe, Vision, and Leadership
They also argued for rethinking the traditional education-work-retirement lifecycle in favor of continuous reskilling and career renewal, supported by robust public and corporate transition programs.
3 — Complementary Social and Economic Institutions. Every transformative technology requires governance.
Automobiles required driver’s licenses, traffic regulations, and insurance systems. AI similarly requires institutions that define accountability while encouraging innovation.
Among the priorities identified by the authors are updated labor standards that more broadly distribute AI’s productivity gains and tax policies that appropriately balance investments in human workers and intelligent machines.
4 — Complementary Metrics. “Metrics frame our questions, guide policy, influence investment, and ultimately shape the future we build.”
Traditional productivity metrics alone are insufficient for understanding AI’s broader impact.
Instead, they call for new measures that evaluate human-AI collaboration, distinguish between augmentation and automation, track the evolution of work over time, measure AI’s regional economic effects, assess its impact on urban labor markets, evaluate new systemic vulnerabilities, and identify unintended consequences when AI optimizes narrow objectives while overlooking broader societal values.
5 — Complementary Systems Thinking. Finally, “AI should not be viewed simply as another technology but as part of a much larger sociotechnical system.”
Complexity science reminds us that local technological changes often produce unexpected consequences throughout interconnected economic and social systems. AI will not operate in isolation, and neither should our approaches to governing it.
History suggests that transformative technologies do not automatically improve society. Electricity, automobiles, and the Internet generated enormous economic and social benefits only after governments, businesses, educational institutions, and civil society built the complementary infrastructure, skills, governance, and institutions needed to realize their potential.
Artificial intelligence is unlikely to be different. The question is no longer whether AI will transform our economy — it almost certainly will. The more important question is whether we will make the complementary investments needed to ensure that its benefits are broadly shared.
As Loaiza and Rigobon conclude, AI’s long-term legacy will depend less on the technology itself than on the choices we make about the systems that surround it. Whether AI ultimately becomes another welfare-enhancing general-purpose technology is not predetermined. It is a societal choice — one that will shape not only the wealth AI creates, but also the kind of future it helps us build.