Research

Generative AI

Abstract Strategic decisions are uncertain and often irreversible. Hence, predicting the value of alternatives is important for strategic decision making. We investigate the use of generative artificial intelligence (AI) in evaluating strategic alternatives using business models generated by AI (study 1) or submitted to a competition (study 2). Each study uses a sample of 60 business models and examines agreement in business model rankings made by large language models (LLMs) and those by human experts. We consider multiple LLMs, assumed LLM roles, and prompts. We find that generative AI often produces evaluations that are inconsistent and biased. However, when aggregating evaluations, AI rankings tend to resemble those of human experts. This study highlights the value of generative AI in strategic decision making by providing predictions.

Abstract Creativity is core to being human. Generative artificial intelligence (AI)—including powerful large language models (LLMs)—holds promise for humans to be more creative by offering new ideas, or less creative by anchoring on generative AI ideas. We study the causal impact of generative AI ideas on the production of short stories in an online experiment where some writers obtained story ideas from an LLM. We find that access to generative AI ideas causes stories to be evaluated as more creative, better written, and more enjoyable, especially among less creative writers. However, generative AI-enabled stories are more similar to each other than stories by humans alone. These results point to an increase in individual creativity at the risk of losing collective novelty. This dynamic resembles a social dilemma: with generative AI, writers are individually better off, but collectively a narrower scope of novel content is produced. Our results have implications for researchers, policy-makers, and practitioners interested in bolstering creativity.

Select media coverage

Research shows AI can boost creativity for some, but at a cost, NPR, July 12, 2024

Faster to compose, but more boring: what AI does to stories, The Naked Scientists Podcast, July 12, 2024

AI prompts can boost writers’ creativity but result in similar stories, study finds, The Guardian, July 12, 2024

AI can make you more creative—but it has limits, MIT Technology Review, July 12, 2024

Experiment finds AI boosts creativity individually — but lowers it collectively, TechCrunch, July 12, 2024

How AI can make your workplace more equitable, Forbes, August 24, 2023

Abstract We introduce a framework for understanding the impact of generative AI on human work, which we call the human–AI task tensor. A tensor is a structured framework that organizes tasks along multiple interdependent dimensions. Our human–AI task tensor introduces a systematic approach to studying how humans and AI interact to perform tasks, and has eight dimensions: task definition, AI contribution, interaction modality, audit requirement, output definition, decision-making authority, AI structure, and human persona. After describing the eight dimensions of the tensor, we provide illustrative frameworks (derived from projections of the tensor) and a human–AI task canvas that provide analytical tractability and practical insight for organizational decision-making. We demonstrate how the human–AI task tensor can be used to organize emerging and future research on generative AI. We propose that the human–AI task tensor offers a starting point for understanding how work will be performed with the emergence of generative AI.

Abstract At the heart of scientific discovery are expert researchers who identify research ideas worthy of inquiry. While generative artificial intelligence (AI) technologies—large language models, in particular—have been found to be widely adopted by knowledge workers for various tasks, their use in assisting researchers with the generation of research ideas, one of the most fundamental tasks in science, remains unexplored. In a randomized online experiment with 310 researchers across disciplines, we study how providing generative AI ideas affects researchers’ self-evaluation of their subsequent proposals, and their overall research agenda. We do not find any average effect on their assessment of the proposals’ novelty or feasibility. However, research experience is an important moderator: experience negatively moderates the effect of generative AI on perceived novelty and impact of the proposal, and on their own research agendas. Further analyses of the quantitative data and open-ended qualitative participant responses reveal the underlying mechanisms. Less experienced researchers accept and integrate generative AI, because it triggers new lines of thinking and validates their existing ideas. By contrast, more experienced researchers show aversion, primarily due to discounting outside ideas, hesitancy toward technology, and a perceived challenge to one’s identity. Our findings contribute to the innovation literature by offering insights into generative AI’s role in the production of new research directions, and to the growing literature on generative AI’s role in complementing human tasks. 

Abstract Generative artificial intelligence (AI) has the potential to change society and work, but little is known about the psychological reactions people have when using it on a day-to-day basis. We conducted a five-day, within-person experience-sampling experiment with participants working across a diverse range of jobs and industries. Each day, we experimentally varied whether people were told to use generative AI as much as possible—where they were free to decide how to use generative AI in their daily tasks—or refrain from using it. Our investigation focused on three domains. The first domain primarily concerns how using generative AI alters the nature, sequence, and scope of work tasks. The second domain explores how perceived effort changes with AI use, influencing perceptions of job demands, work meaningfulness, and anticipated peer appreciation. The third domain addresses anxiety- related effects, including concerns about job insecurity, accuracy, and reputational risks. Our results reveal that on days when individuals used generative AI, they altered the nature, scope, and sequence of their tasks, perceiving their work as more demanding yet more appreciated by peers. Notably, we did not find evidence of increased job insecurities or work- related concerns.

Social media

Abstract This paper examines technology adoption where sets of new members – generational cohorts – enter the population over time. I look at the diffusion of a social media platform, Twitter, through eight annual generational cohorts of newly premiering television shows, debuting between 2006 and 2013. I first show that a new generational cohort’s rate of adoption is affected by the behaviour of prior cohorts that are present in the existing population. I then show how the macro-diffusion patterns change over the generational cohorts. These results have implications for organisations developing and managing new technologies. Incumbents must account for a distinct diffusion process that occurs with each new generational cohort to defend against the constant threat of a rival technology diffusing more widely in a new generational cohort. Organisations developing newer technologies can enter and successfully compete in a market with a widely diffused incumbent technology if they are able to attract a newer generational cohort.

Abstract Platform governance helps align the activities of participating actors to deliver value within the platforms. These platforms can operate in environments where governance is intentionally or conventionally weak in favor of open access, frictionless transactions, or free speech. Such low- or no-governance environments leave room for illegitimate actors to penetrate platforms with illegitimate content or transactions. We propose that an external observer can employ transparency mechanisms to establish “soft” governance that allows participants in a low-governance environment to distinguish between sources of legitimate and illegitimate content. We examine how this might work in the context of disinformation Internet domains by training a machine learning classifier to discern between low-legitimacy from high-legitimacy content providers based on website registration data. The results suggest that an independent observer can employ such a classifier to provide an early, although imperfect, signal of whether a website is intended to host illegitimate content. We show that the independent observer can be effective at serving multiple platforms by providing intermediate prediction results that platforms can align with their unique governance priorities. We expand our analysis with a signaling game model to ascertain whether such a soft governance structure can be resilient to adversarial responses.

Select media coverage
New academic tool used to identify fake news domain names, TechRadar, November 12, 2020
Boffins devise early-warning system for fake news: AI fingers domains that look sus, The Register, November 12, 2020

Abstract Event studies, which have significantly advanced mergers and acquisitions (M&A) research, obtain excess returns based on a theory linking a firm's shareholder returns to those of the market. For outcomes lacking such a theory, we propose an empirical approach using a synthetic control method with machine learning to link outcomes for the acquirer or target to those for a group of comparison firms. We discuss the method's assumptions, its close parallel to event studies, and its difference in weighting comparison firms (based on data versus derived from theory). We provide an illustration of Dollar Tree's acquisition of Family Dollar, by analyzing shareholder returns (to demonstrate consistent results with an event study), realized cost and sales synergies, and customer sentiment (derived from more than 52 million Twitter messages). We highlight this method's potential—for M&A and other areas of strategy research—to open up new lines of inquiry.

Abstract We study how social media followers of an organization—a resource that resides outside its boundaries—provides value to the organizations. We theorize that both follower stocks and flows lead to higher performance outcomes but through different mechanisms: stocks raise the general profile of an organization, and flows trigger information delivered to new followers’ networks. We further hypothesize on an organizational characteristic that moderates these two effects in opposite directions: an organization’s intensity of interest is the extent to which it inspires a connection with people beyond consumption. We theorize that the positive effect of stocks is stronger for organizations with a low intensity of interest, and the effect of flows is stronger for organizations with a high intensity of interest. We test our theory using followers of television shows on Twitter and Nielsen viewership ratings as a performance measure and find support for our hypotheses. We control for alternative explanations like inherent differences across shows, bandwagoning, show quality, and a show’s “buzz.” Our results suggest that an organization’s social media followers contribute to its value in distinct ways from other external resources.

Abstract CEOs are increasingly adopting social media to directly communicate with outside constituencies. However, some activities CEOs undertake on social media may create a tension with their fiduciary obligations as head of the firm. To better understand this tension, we consider the completion an initial public offering – instances where the CEO’s fiduciary obligations are more salient. We compare social media behaviors of CEOs of companies that completed an IPO compared with those that filed but pulled their IPO. Our findings indicate that faced with an increase in fiduciary obligations, CEO social media content shifts towards discussing performance. We also find differing social media behaviors of CEOs following an increase in fiduciary obligations depending on firm size and whether the firm is in a high-technology industry. Our results have important oversight and social implications, as CEO activism becomes more prominent.

Commercial Space

Abstract We examine how catastrophic innovation failure affects organizational and industry legitimacy in nascent sectors by analyzing the interactions between Virgin Galactic and stakeholders in the space community in the aftermath of the firm’s 2014 test flight crash. Following catastrophic innovation failure, we find that industry participants use their interpretations of the failure to either uphold or challenge the legitimacy of the firm while maintaining the legitimacy of the industry. These dynamics yield two interesting effects. First, we show that, in upholding the legitimacy of the industry, different industry participants rhetorically redraw the boundaries of the industry to selectively include players they consider legitimate and exclude those they view as illegitimate: detracting stakeholders constrain the boundaries of the industry by excluding the firm or excluding the firm and its segment, whereas the firm and supporting stakeholders amplify the boundaries of the industry by including firms in adjacent high-legitimacy sectors. Second, we show that, in assessing organizational legitimacy, the firm and its stakeholders differ in the way they approach distinctiveness between the identities of the industry and the firm. Detracting stakeholders differentiate the firm from the rest of the industry and isolate it, whereas the firm and supporting stakeholders reidentify the firm with the industry, embedding the firm within it. Overall, our findings illuminate the effects that catastrophic innovation failure has over high-order dynamics that affect the evolution of nascent industries.

Abstract Catastrophic innovation failure challenges firms to address its causes internally and reframe their innovative efforts to external stakeholders. We examine how firms respond to these concurrent challenges by contrasting Virgin Galactic (VG)’s internal and external responses following its 2014 deadly test flight crash. Internally, following the crash, the increasingly risk-averse firm made design choices that derisked its spacecraft, but also reduced its technical capabilities. Externally, VG reframed its efforts by recalibrating two elements: the promise about the experience its innovation would enable, and the risk inherent in the development. While VG’s pre-crash framing provided a granular representation of its internal design by zooming into technical details, after the crash, VG rebuilt meaning and credibility by shifting to a holistic representation that zoomed out to focus on the significance of its pursuit. We propose a new way that firms can reframe their efforts following catastrophic innovation failure: by altering the depth of their framing of promise and risk in relation to their internal design choices. We also draw on these findings to theorize a “promise–risk balance” that innovating firms face as they navigate internal development and external communication, between promising a positive outcome and acknowledging the risks inherent in the process.

Other Settings

Abstract Entrepreneurs have increasingly entered digital crowdfunding platforms as a viable option for acquiring capital. This paper examines how high-performance outliers—projects that raised substantial amounts of capital—affect other entrepreneurs’ decisions to enter crowdfunding. The present study focuses on the two largest rewardsbased crowdfunding platforms, Kickstarter and Indiegogo. Results indicate that, following outliers, entry was relatively higher on the platform with laxer, entrant-friendly governance (i.e., Indiegogo). This effect was more pronounced among low-quality entrants and moderated by projects in categories that have higher capital requirements. The findings suggest that differences in platform governance influence how subsequent entrepreneurial entrants behave.

Abstract Mandatory information disclosure regulations seek to create institutional pressure to spur performance improvement. By examining how organizational characteristics moderate establishments' responses to a prominent environmental information disclosure program, we provide among the first empirical evidence characterizing heterogeneous responses by those mandated to disclose information. We find particularly rapid improvement among establishments located close to their headquarters and among establishments with proximate siblings, especially when the proximate siblings are in the same industry. Large establishments improve more slowly than small establishments in sparse regions, but both groups perform similarly in dense regions, suggesting that density mitigates the power of large establishments to resist institutional pressures. Finally, establishments owned by private firms outperform those owned by public firms. We highlight implications for institutional theory, managers, and policymakers.