ABSTRACT:
The most recent laureate of the Nobel Memorial Prize in Economic Sciences, Joel Mokyr, has been recognized for his work on the power of ideas in economic development. The economic historian has shown how the propositional knowledge about natural phenomena feeds the prescriptive knowledge of practical techniques. Furthermore, the impacts of both types of what he called useful knowledge are enacted through social processes that lead to innovation [3]. The broad access to the existing ideas big and small and to the thinking of the idea makers drives the economic and societal development of humanity. The ideas, however radical, emerge and are built in a long chain, or sometimes interdisciplinary web, of their predecessors. We could not have CRISPR gene editing without the discovery of DNA; we could not have SpaceX without the rocketry visions of Robert Goddard in the United States or Ary Sternfeld in France and in the former Soviet Union. The development of generative artificial intelligence (GenAI) based on large language models (LLM) has been formed by a long path of ideas, some of which turned out to be a detour of fruitful semantic AI, to the resurrection of the idea of neural networks, all in the intellectual and social processes. It is quite unlikely that we would see the today’s blossoming of GenAI without the connectivity of the Internet and the accessibility to knowledge and information furnished by the Web. The dense network of ideation has resulted in the erstwhile emergence of our knowledge economy, where the ideas interbreed and receive financial and societal support [2]. Our field of information systems (IS) does and will contribute meaningfully to the development of ideas brought to fruition by researching information technology (IT)-based ideation, co-creation in crowdsourcing, and the provision of access to structured knowledge corpora—among other contributions [1, 4]. With the speed of the Web-based amplification and the support of the mass access to the existing knowledge, small ideas can scale to a greater effect, as a new vibe-coding shortcut featured on GitHub propagating to a significant financial outcome.
The first paper in the issue showcases what we can do as a scholarly field to contribute to the emergence of new ideas. As said, in their generality they emerge on the foundation of their predecessors (in the greatest cases, the Newtonian standing on the shoulders of giants). The present work empirically studies the functioning of online platforms that assist ideation by providing access to hyperlinked knowledge and enabling collective action in collaborating on new ideas, directly or indirectly. Yimei Zhou, Qian Tang, Vincent Z.W. Mack, and Shao Yi Liaw investigate the operation and effects of the curated hyperlinked knowledge made accessible on the platforms (say GitHub or Wikipedia) on the success of novel ideas through the recombination and enhancement of their predecessors. Thus, the authors study the interaction of the content of the idea networks with their structure. They base themselves in the knowledge-collaboration theory and determine the roles of the idea makers’ expertise and co-creation of new ideas by the collaborating contributors. Idea networks allow the tracing of collaborative development of an idea and serve the authors in drawing the empirically-based conclusions on the process of recombinative collaborative ideation.
In a related next paper, the authors investigate a key outcome of the collaboration between humans and LLMs. Monideepa Tarafdar, Martin Adam, and Long The Nguyen study this co-creation on knowledge-based tasks with respect to the factuality: the correspondence of the outcome to reality and already known facts. As we know full well, challenges abound here, from the mildly called hallucinations to the outright rubbish. The authors investigate empirically how a human’s reflection can influence this correspondence to reality. Tarafdar and colleagues establish how the human-LLM interaction modes (conversational versus adversarial) and the human cognitive states influence the factuality. The results are important and can serve in the design of human-LLM systems that will increasingly define the functioning of organizations. Since ideas lead to further innovation, we can expect that LLMs are not the last word in AI, yet the results obtained here should be valid in the future.
Delving further into our understanding of working with GenAI, Guohou Shan, Michael Rivera, Subodha Kumar, and Pawan Anand study the effects of its use on the coding performance of software developers. The use is now common, and the research is of the moment. In a multimethod research program, the authors laudably performed both a longitudinal field study and a randomized experiment. They are able to lead their work and explain their results through the lens of cognitive load theory. And the most interesting of these results is that the human-GenAI co-creation can deliver higher quality code with diminished cognitive effort. The contribution is both epistemologically significant and bears implications for the presently challenged software-developer marketplace.
In another of their many roles, algorithmic systems—both LLM-based and simply scripts—serve commonly as conversational agents (chatbots). They make mistakes. How do their users, willing or not, witting or not, react to these errors? Alfred Benedikt Brendel, Sascha Lichtenberg, Fabian Hildebrandt, and Alan R. Dennis conduct an empirical study comparing users’ responses in the light of the attribution theory. The results establish that the perceived humanness of a chatbot makes its relatively minor errors appear insignificant to the user (well, people make mistakes, and this looks like a person) and do not affect users’ satisfaction with the system. The authors conclude that the agents more likely to make mistakes (such as LLM-based for now) should be designed more anthropomorphically, in contrast to the script-based responders.
Two following papers address online security, a perennial and highly weighty concern in our IS field. With the increasing power and accessibility of LLMs, security concerns can only grow (as will the capabilities of protection measures). The first of the papers, by He Li, assesses the impact of security measures taken by the participants in the platform ecosystems. Clearly, an entry of a software product with impaired security can imperil the entire ecosystem. Beyond that, these complementors of value offered by the focal platform need to compete with other such participants in the overall ecosystem (say, the data-analytics apps offered on Hadoop). Such differentiation is particularly important for the startups aiming to join the platform. Just how important is it? The author answers this question by assessing the effects of the signaled security features of the product entering the platform ecosystem on the venture funding made available to the startup. In a nuanced assessment, the work leads to a recognition of value of security features to the differentiated entry into a platform ecosystem.
The Journal of Management Information Systems (JMIS) has always recognized that design research is highly important to the weight and credibility of our IS field. The next paper addresses the AI-created exposures in the context of open-source software. Opening such software has the—by now traditional—advantages of open source, among them, rapid co-creation, many eyes on the potential faults, and the economies of production. The accessibility of machine-learning software creates, however, serious dangers. The issue is addressed by the design offered and tested by the authors of the next paper, Ben Uribe Lazarine, Sagar Samtani, Hongyi Zhu, and Ramesh Venkataraman. In the linked AI repositories, propagation of vulnerabilities is a key threat. The authors’ system counters it with the use of machine learning (use AI to protect AI). The tests show the superiority of the authors’ approach, as well as the potential of the method for even broader use in IS design.
Security underlies privacy. Yet the attainment of privacy is to a large extent user-driven. There are basically two types of users: those who are engaged in protecting their online privacy and those are not. Here, Nicholas James Brown, Vitali Mindel, Paul Benjamin Lowry, and Quinton Nottingham offer a broad perspective on the issue, basing themselves in an encompassing onto-epistemological framework. The authors shift their analysis from that of users’ concerns with the risks of privacy exposures, a more traditional approach, to that of privacy interests. On the philosophical basis the authors’ employ, this broader approach to privacy seeks to understand the meaningfulness of privacy to the users in the light of their experience and belief in the potential effectiveness of their privacy-related actions. The findings, based on the empirics with a large number of participants, show that the privacy-interests lens is more effective in explaining the proactive and sustained actions of users, as contrasted with the privacy-concerns analysis that is more apt in explaining the reactive actions. The novel approach will be helpful in system design and privacy-oriented regulation.
There are several methods for gauging strategic opportunities in the competitive marketplace. The next paper presents and shows the advantages of a novel such method. Myunghwan Lee, Gene Moo Lee, Hasan Cavusoglu, and Marc-David L. Seidel recognize the greater challenges of such assessment in the current environment of fluid corporate boundaries, massive attempted entries to the marketplaces, and multihoming of companies in business domains. The computational approach to opportunity assessment leverages the availability of extensive data, yet the use of the data has to be grounded in a sound theory. The authors ground their use of the data for opportunity seeking in the social network theory, seeking structural holes within the competitive network of firms and products. Deploying the data available on public companies, the authors show the advantages of their novel method of looking for strategic opportunities with the IS systems.
The paper by Hamed Qahri-Saremi, Nima Korzadeh, and Ofir Turel joins the body of research on online reviews, with a novel contribution: the researchers look at the review process. The quality of reviews is of course paramount to their helpfulness to the future users. In pursuit of higher quality, does it matter whether the reviewers rate first and review second (the general current approach) or vice versa? Based on a large consumer panel and cognitive analysis, the researchers show that it does. The review-first approach leads to more thoughtful reviews and well-grounded subsequent ratings of the products. Moreover, there is a significant distinction in quality between the reviews furnished from a mobile device versus a PC. Simple actions by the platforms and firms obtaining the reviews can thus make them more helpful to their customers.
Medical crowdfunding has become a large sector in the charitable area, with individuals facing healthcare-related challenges seeking funds from potential donors. Online platforms facilitate and regularize the behavior of the participants. Signaling the virtue and the need is obviously highly important for the fund seekers, and this side of crowdfunding has been quite extensively investigated by the scholars. Here, Victor Xiaoqi Wang and Hua (Jonathan) Ye analyze the behavior of the other side, that of the potential donors. The credibility of the campaign is the key to donations. Donors may be anonymous or onymous. Based on the signaling theory and the data from a crowdfunding platform, the authors establish that the anonymity of previous donors lowers the diagnosticity of the campaign and attenuates the donations of the potential subsequent philanthropists. Even though donors may gain greater virtue by remaining anonymous, privacy has its costs—and these should be made clear by the platforms.