A carregar...
Publicação
Human-AI Collaboration: Understanding the Dynamics of Decision-Making with Explainable AI
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 31.88 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Considering the rapid development of Artificial Intelligence (AI) models like ChatGPT, Copilot or
Gemini in recent years, research on how these models influence their users has become essential - not
only for user safety but to guide the future development of more efficient models. This study investigates
the impact that predictions and explanations acquired from ChatGPT have on human decision-making,
with a focus on task complexity, participant confidence, and demographic variables.
To evaluate this impact, an experimental study was conducted with 53 participants, using a convenience
sampling method. Using Qualtrics, the participants, primarily university students, were divided into two
groups: Easy Decision-Making (EDM) and Complex Decision-Making (CDM). Each group was
subdivided into three subgroups: those receiving only AI predictions, those receiving predictions with
explanations, and a control group with no AI assistance. Performance was measured in terms of decision
accuracy, alignment with AI predictions, and confidence, then compared between groups and subgroups.
Additionally, demographic data, participants' attitudes towards technology and AI, and perceived
relevance of information were collected to explore how these factors might influence the effectiveness
of AI assistance and to test the awareness of this influence.
Although no statistically significant differences were found between the subgroups, the analyses indicate
a tendency for AI assistance to be more effective in more complex tasks, while explanations may be
more beneficial in simpler scenarios. The study also highlighted that explanations can reduce ambiguity
in decision-making, aligning participants' decisions more closely with AI predictions, even if they are
unaware of this influence. Demographic factors did not significantly influence AI's impact, though
younger participants generally performed better, possibly due to greater familiarity with technology.
Attitudes towards AI also played a role, with scepticism boosting confidence levels, while partial trust
in AI reduced it. Future research with larger, more diverse samples is recommended.
Descrição
Tese de mestrado, Ciência Cognitiva, 2024, Universidade de Lisboa, Faculdade de Ciências
Palavras-chave
Inteligência Artificial Explicável Tomada de Decisão na Incerteza Colaboração Humano-IA Influencia Demográfica na Inteligência Artificial Confiança com Inteligência Artificial Teses de mestrado - 2024