Data Thinking: um framework para a colaboração humano-IA na Engenharia de Dados Sintéticos

Abstract

Synthetic data engineering through large language models (LLMs) has been used as an alternative for training machine learning (ML) models in contexts where real data is limited, sensitive, or difficult to obtain. However, the absence of human supervision in the generation process can introduce inconsistencies, hallucinations, and biases into the data, compromising the quality of the information produced. In this sense, this work presents Data Thinking, a framework focused on human-AI collaboration inspired by Design Thinking principles, with a focus on supporting the planning, generation, and validation of synthetic data. The proposal was submitted to a feasibility study that investigated its applicability in synthetic data generation, aiming to evaluate the technological acceptance of the artifact from the perspective of software designers. The results indicate that the use of Data Thinking supports exploratory and reflective data engineering processes, contributing to the construction and validation of more reliable synthetic databases for training ML models.