The present work introduces a new insight problem task: joke completion. We found that performance and magnitude of insight within it correlated with an established task: rebus puzzles. However, participants performed worse on and took longer in joke completion problems than in their rebus counterparts. Further, the distribution of reported insight was bimodal only for rebuses, as should be expected of an insight problem. In joke completion problems, both self-estimated and externally-rated joke funniness correlated with reported insight. Challenging the assumption of impasse, performance and insight decreased as a function of trial time for both problem types, with the best and most insightful solutions submitted within the first 20 seconds. While this is a preliminary study, we argue that it signals a promising direction for the problem solving, humor, and creativity literatures by providing a new approach to capture insight in a manner conducive to linguistic and cognitive modeling techniques.

Though humans should defer to the superior judgement of AI in an increasing number of domains, certain biases prevent us from doing so. Understanding when and how these biases occur is a central challenge for human-computer interaction. A proposed source of such bias is the perceived subjectivity of tasks. We tested this hypothesis using one of the most subjective tasks possible: Evaluating joke funniness. Across two experiments, we addressed the following: Would people rate jokes as less funny if they believed an AI created them? When asked to rate jokes and guess their likeliest source, participants evaluated jokes attributed to humans as the funniest and those to AI as the least funny. However, when we explicitly framed these same jokes as either human or AI-created, there was no difference in performance-level ratings. These results challenge the notion that task subjectivity always biases users against AI if the source is transparent.

Though humans should defer to the superior judgement of AI in an increasing number of domains, certain biases prevent us from doing so. Understanding when and why these biases occur is a central challenge for human-computer interaction. One proposed source of such bias is task subjectivity. We test this hypothesis by having both real and purported AI engage in one of the most subjective expressions possible: Humor. Across two experiments, we address the following: Will people rate jokes as less funny if they believe an AI created them? When asked to rate jokes and guess their likeliest source, participants evaluate jokes that they attribute to humans as the funniest and those to AI as the least funny. However, when these same jokes are explicitly framed as either human or AI-created, there is no such difference in ratings. Our findings demonstrate that user attitudes toward AI are more malleable than once thought-even when they (seemingly) attempt the most fundamental of human expressions.

Understanding the processes leading to insight has remained one of psychology’s greatest challenges. In this study, we examined how different lexical properties affect cognitive processes involved in a popular class of insight problems: Compound Remote Associates (CRAs). These properties were familiarity, lexeme meaning dominance, and semantic transparency. We found that a higher proportion of problems were solved when they were presented beginning with the most familiar cues, but not when they began with right-headed dominant or the most semantically transparent cues. Further, we found that participants focused their efforts disproportionately on first and last cues, that subjective ratings of insight decreased as trial times elapsed, and that the magnitude of reported insight increased with the number of cues successfully solved. This suggests that participants can monitor their progress in such problems. These results contest longstanding assumptions of requisite periods of impasse and the absence of incremental progress in insightful problem solving.