Research

When a technology captures widespread public excitement, technical terms become hyped labels. Existing studies find that innovators use these hyped labels to attract attention, signal legitimacy, and mobilize resources, but I argue that the strategy carries a hidden cost when innovations reach expert gatekeepers, the intermediaries who evaluate and certify them. Expertise enables gatekeepers to distinguish substantive uses of hyped labels from uses that outrun the underlying technical content. I theorize this response as expertise-based scrutiny and hypothesize heightened scrutiny of innovations carrying hyped labels, concentrated among expert evaluators. I test the argument in patent examination at the USPTO, exploiting AlphaGo’s March 2016 victory as an exogenous shock to AI hype, and examine applications filed before the shock but evaluated on either side of it, so that their underlying substance is predetermined. I find that, when examined by AI-expert examiners after the shock, applications carrying hyped labels became 12 percentage points more likely to be rejected for inadequate disclosure under §112, the finding that an application claims more than its technical description teaches; the increase is roughly a third of the base rate. I find no corresponding effects for subject-matter eligibility (§101) or novelty and non-obviousness (§102/§103), and no response among non-expert examiners. These findings show that hype is absorbed unevenly within innovation systems: scrutiny intensifies where hype meets the expertise required to evaluate it. For innovators, the hyped label that attracts broad audiences also invites scrutiny from the expert ones.

Using novel data on federal regulations and the patents of 1,242 firms from 1994 to 2013, we find that regulatory restrictiveness can have both a negative and positive relationship with innovation output depending on the level of regulatory uncertainty and the innovation type in question.

Applying machine learning to five decades of U.S. patent applications, we show that the rise of organizational software turns organizational innovations, long assumed unpatentable, into patentable technological ones, identifying more than 200,000 such applications.

The first systematic, large-scale comparison of DARPA with the NSF and NIH. DARPA-funded research is more disruptive and more likely to be linked to technological discoveries: DARPA disproportionately selects investigators with prior histories of disruptive research and work in Pasteur’s Quadrant.

Using novel household data, we show that firms led by CEOs with broader nonwork interests were more likely to form cross-industry alliances in response to the COVID-19 pandemic.

Linking patent and trademark records across four sharply dated buzz episodes—virtual reality, artificial intelligence, augmented reality, and blockchain—I find that US public firms with relevant prior capabilities expand patenting, inventor hiring, and skill demand after buzz onset, yet bring roughly one-quarter fewer products toward market.

Using large-scale bibliometric data and epidemic shocks as quasi-natural experiments, we find that the emergence of hot research topics widens the gender gap in publication output, driven by unequal capacity to pivot into newly salient areas.

A natural-language-processing crosswalk between 439,000 USPTO green patents and 4.25 million pages of the Code of Federal Regulations (1998–2022), using generative AI to classify command-and-control versus market-based regulation and citation network analysis to trace regulatory dynamics.

Using the launch of Amazon Web Services as a shock to VC screening, we examine how a shift in the criteria investors use to select startups propagates into the direction of scientific research, including its disruptiveness and the coupling between science and technology.