ID: 1929
Presenting Author: Cristian Pérez
Session: 581 - Experiences in enhancing communication through the application of AI in IA
Status: pending
The presentation highlights the need to supervise AI-based follow-up analyses to avoid bias or errors, discussing implications for practitioners and decision-making processes.
In medical and psychological terms, a hallucination is a sensory perception that occurs without an external stimulus yet possesses the apparent reality of a genuine perception. Hallucinations differ from illusions, which involve misinterpretations of real stimuli. In Artificial Intelligence (AI), a hallucination denotes the generation of information that appears plausible but lacks factual or data-based support. Such outcomes may include fabricated facts, citations, or inconsistent interpretations. Within AI-assisted report analysis, hallucination arises when the system produces conclusions or summaries not substantiated by the original data. This occurs due to ambiguous prompts, contextual bias, or overgeneralization. Mitigation strategies include prompt precision, verification mechanisms, and systematic human review. These safeguards are critical in environmental applications, where accuracy, transparency, and traceability carry legal implications. In Chile, follow-up reports derived from obligations under the Environmental Impact Assessment (EIA) process are submitted by permit holders to the environmental authority via an official public platform. Providing false or incomplete information in this context entails legal consequences under current environmental regulations. This presentation examines best practices to minimize AI hallucinations during the review of follow-up reports. Emphasis is placed on prompt engineering, model validation, and human oversight to enhance reliability, ensure data integrity, and maintain compliance with environmental reporting standards.
Dr. Cristian Pérez. Environmental compliance practitioner. Lecturer on environmental topics.
Veterinarian with a Master degree on environmental planning and management and Dr. in Sciences.
Coauthor 1: Gino Araya