Publicação
How prompt engineering influences GenAI: The role of examples
| Resumo: | Generative AI (GenAI) is increasingly employed in ideation and innovation contests, where it supports creativity by expanding the solution space and stimulating originality. In this context, prompts serve as the interface with AI, and the inclusion of examples emerges as a key design factor. Prior studies on human solvers highlight both benefits—reducing ambiguity and guiding idea generation—and risks, such as fixation and reduced variety. This thesis investigates whether these dynamics also hold when the solver is GenAI. The research draws on data from the Crowdsourcing Lab at the University of Palermo (2024–2025), where students addressed creative challenges with and without GenAI. Prompts were coded according to three dimensions identified in the literature: (i) presence of examples, (ii) number of examples, and (iii) type of examples (valid “positive” examples vs. invalid “negative” ones). Statistical analysis through linear regressions (Stata14) assessed the impact of these factors on creativity scores. Findings show that GenAI does not reproduce the same patterns observed with humans. The presence of at least one positive example significantly enhances creativity, confirming its orienting role. By contrast, providing too many examples reduces creativity, indicating statistical convergence rather than exploration. Negative examples show no significant effect, suggesting that GenAI does not activate de-fixation processes as human solvers do. This thesis contributes to the literature on prompt engineering, creativity, and crowdsourcing, showing that design principles effective with human participants require careful adaptation when applied to GenAI. |
|---|---|
| Autores principais: | La Paglia, Alice |
| Assunto: | Prompt engineering Generative AI Examples Crowdsourcing Feasibility Engagement Engenharia de prompt IA generativa Exemplos Criatividade -- Creativity Viabilidade Engajamento |
| Ano: | 2025 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso restrito |
| Instituição associada: | ISCTE |
| Idioma: | inglês |
| Origem: | Repositório ISCTE |
| Resumo: | Generative AI (GenAI) is increasingly employed in ideation and innovation contests, where it supports creativity by expanding the solution space and stimulating originality. In this context, prompts serve as the interface with AI, and the inclusion of examples emerges as a key design factor. Prior studies on human solvers highlight both benefits—reducing ambiguity and guiding idea generation—and risks, such as fixation and reduced variety. This thesis investigates whether these dynamics also hold when the solver is GenAI. The research draws on data from the Crowdsourcing Lab at the University of Palermo (2024–2025), where students addressed creative challenges with and without GenAI. Prompts were coded according to three dimensions identified in the literature: (i) presence of examples, (ii) number of examples, and (iii) type of examples (valid “positive” examples vs. invalid “negative” ones). Statistical analysis through linear regressions (Stata14) assessed the impact of these factors on creativity scores. Findings show that GenAI does not reproduce the same patterns observed with humans. The presence of at least one positive example significantly enhances creativity, confirming its orienting role. By contrast, providing too many examples reduces creativity, indicating statistical convergence rather than exploration. Negative examples show no significant effect, suggesting that GenAI does not activate de-fixation processes as human solvers do. This thesis contributes to the literature on prompt engineering, creativity, and crowdsourcing, showing that design principles effective with human participants require careful adaptation when applied to GenAI. |
|---|