Titled “Generative AI at the Crossroads: Light Bulb, Dynamo, or Microscope?” the paper, authored by Martin Neil Baily, David M. Byrne, Aidan T. Kane, and Paul E. Soto, explores whether genAI represents a fleeting innovation or a groundbreaking force akin to past general-purpose technologies (GPTs) such as electricity and the internet.

The Fed economists ultimately conclude their “modal forecast is for a noteworthy contribution of genAI to the level of labor productivity,” but caution they see a wide range of plausible outcomes, both in terms of its total contribution to making workers more productive and how quickly that could happen. To return to the light-bulb metaphor, they write that “some inventions, such as the light bulb, temporarily raise productivity growth as adoption spreads, but the effect fades when the market is saturated; that is, the level of output per hour is permanently higher but the growth rate is not.”

Here’s why they regard it as an open question whether genAI may end up being a fancy tech version of the light bulb.

GenAI: a tool and a catalyst

According to the authors, genAI combines traits of GPTs—those that trigger cascades of innovation across sectors and continue improving over time—with features of “inventions of methods of invention” (IMIs), which make research and development (R&D) more efficient. The authors do see potential for genAI to be a GPT like the electric dynamo, which continually sparked new business models and efficiencies, or an IMI like the compound microscope, which revolutionized scientific discovery.

The Fed economists did cautioning that it is early in the technology’s development, writing “the case that generative AI is a general-purpose technology is compelling, supported by the impressive record of knock-on innovation and ongoing core innovation.”

Since OpenAI launched ChatGPT in late 2022, the authors said genAI has demonstrated remarkable capabilities, from matching human performance on complex tasks to transforming frontline work in writing, coding, and customer service. That said, the authors said they’re finding scant evidence about how many companies are actually using the technology.

Limited but growing adoption

Despite such promise, the paper stresses that most gains are so far concentrated in large corporations and digital-native industries. Surveys indicate high genAI adoption among big firms and technology-centric sectors, while small businesses and other functions lag behind. Data from job postings shows only modest growth in demand for explicit AI skills since 2017.

“The main hurdle is diffusion,” the authors write, referring to the process by which a new technology is integrated into widespread use. They note that typical productivity booms from GPTs like computers and electricity took decades to unfold as businesses restructured, invested, and developed complementary innovations.

“The share of jobs requiring AI skills is low and has moved up only modestly, suggesting that firms are taking a cautious approach,” they write. “The ultimate test of whether genAI is a GPT will be the
profitability of genAI use at scale in a business environment and such stories are hard to come by at present.” They know that many individuals are using the technology, “perhaps unbeknownst to their employers,” and they speculate that future use of the technology may become so routine and “unremarkable” that companies and workers no longer know how much it’s being used.

Knock-on and complementary technologies

The report details how genAI is already driving a wave of product and process innovation. In healthcare, AI-powered tools draft medical notes and assist with radiology. Finance firms use genAI for compliance, underwriting, and portfolio management. The energy sector uses it to optimize grid operations, and information technology is seeing multiples uses, with programmers using GitHub Copilot completing tasks 56% faster. Call center operators using conversational AI saw a 14% productivity boost as well.

Meanwhile, ongoing advances in hardware, notably rapid improvements in the chips known as graphics processing units, or GPUs, suggest genAI’s underlying engine is still accelerating. Patent filings related to AI technologies have surged since 2018, coinciding with the rise of the Transformer architecture—a backbone of today’s large language models.

‘Green shoots’ in research and development

The paper also finds genAI increasingly acting as an IMI, enhancing observation, analysis, communication, and organization in scientific research. Scientists now use genAI to analyze data, draft research papers, and even automate parts of the discovery process, though questions remain about the quality and originality of AI-generated output.

The authors highlight growing references to AI in R&D initiatives, both in patent data and corporate earnings calls, as further evidence that genAI is gaining a foothold in the innovation ecosystem.

Cautious optimism—and open questions

While the prospects for a genAI-driven productivity surge are promising, the authors warn against expecting overnight transformation. The process will require significant complementary investments, organizational change, and reliable access to computational and electric power infrastructure. They also emphasize the risks of investing blindly in speculative trends—a lesson from past tech booms.

“GenAI’s contribution to productivity growth will depend on the speed with which that level is attained, and historically, the process for integrating revolutionary technologies into the economy is a protracted one,” the report concludes. Despite these uncertainties, the authors believe genAI’s dual role—as a transformative platform and as a method for accelerating invention—bodes well for long-term economic growth if barriers to widespread adoption can be overcome.

Still, what if it’s just another light bulb?

For this story, Fortune used generative AI to help with an initial draft. An editor verified the accuracy of the information before publishing.