While AI ethics interventions often focus on how researchers should navigate consequential choices, they may overlook a prior question: when do researchers recognize they are making a decision at all? This qualitative study examines how academic NLP teams confront “decision moments” – junctures where latent alternative paths could be considered. We propose a railyard problem analogy: where trolley problems presume a discrete choice between visible options, railyard problems concern whether alternative paths register as possibilities at all. Drawing on decision-tracing interviews across four NLP projects, we demonstrate how technical defaults, institutional structures, and tacit norms (infraethics) combine to organize research as a human-infrastructural process. Many consequential outcomes arise through "passive decisions", where alternatives exist but never become sufficiently visible, viable, or voiced (VVV) to warrant deliberation; "active decisions" only emerge when VVV conditions are met. Our analysis suggests ethics interventions should cultivate the collaborative conditions under which alternatives become recognizable.
Around 90% of CEOs see AI as the “most critical technology for ensuring future profitability and competitiveness.” At the same time, up to 95% of AI projects fail. Currently, little is known about the leaders who approve and guide AI initiatives. We call them AI Deciders. This study investigates how AI Deciders reason about AI benefits and risks, and how their knowledge about AI influences their decisions on what and where to innovate. We interviewed AI Deciders across diverse organizations. We found no ideation. AI Deciders just consider one concept at a time. Design and HCI played no role in deciding what to build. Many AI Deciders overestimate AI’s benefits while underestimating risks. Based on these findings, we identified opportunities for design and HCI to support impactful and responsible AI innovation. This should reduce AI project failure.
Expectation cues such as source labels, expertise signals, or identity-based indicators can bias how humans interpret and evaluate information. In high-stakes domains like healthcare, education, and law, such biases threaten the objectivity of decision-making. As LLMs increasingly provide decision support in these contexts, this study aims to examine whether LLMs exhibit expectation-driven bias akin to that of humans. Across two experiments (N = 1260), we manipulated expectations via priming statements and measured shifts in judgment scores. In both humans and LLMs, higher expectations led to more favorable evaluations for suggestions of equivalent quality, and greater mismatches between expectations and actual performance produced stronger judgment distortions. Notably, humans tended to adjust their evaluations unconsciously, whereas LLMs revised their outputs in a consistent and traceable manner. These findings reveal both shared sensitivities and distinct adjustment patterns, offering design insights for building expectation-aware AI systems that promote fair and transparent human–AI interaction.
Social media has blurred the line between professional journalism and personality-driven commentary, yet we know little about how users evaluate credibility and engage with news from influencers and legacy media when they appear in the same feed. This short paper investigates how political ideology and news source type shape perceived credibility and follow intentions on Instagram. We conducted a mixed-methods experiment where U.S.-based participants (N=120) viewed a set of real news posts and rated the credibility of four accounts (two legacy media–based, two influencer-based), balanced by ideology (two left-leaning, two right-leaning), and indicated whether they would follow each account. Our findings suggest that perceived credibility on Instagram is multi-dimensional, rooted in ideological alignment, yet moderated by institutional signals and perceived authenticity. These insights highlight how platform design and source dynamics can reinforce selective exposure, with implications for both mitigating polarisation and strengthening trust in online news ecosystems.
