Banca de DEFESA: JOÉLCIO DE CARVALHO TONERA

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
STUDENT : JOÉLCIO DE CARVALHO TONERA
DATE: 30/04/2026
TIME: 16:30
LOCAL: https://meet.google.com/zzh-vajh-brc
TITLE:

Comparative Evaluation of LLMs and Human Professionals in Brazilian Trademark Law: A Controlled Experiment with Prompt Engineering and Independent Expert Curation

KEY WORDS:

Artificial Intelligence; Large Language Models; Trademark Law; INPI; Prompt Engineering; Expert Assessment; Intellectual Property.

PAGES: 63
BIG AREA: Ciências Sociais Aplicadas
AREA: Administração
SUMMARY:

This dissertation comparatively investigates the performance of Large Language Models (LLMs) and specialized human professionals in solving practical issues of Brazilian Trademark Law, with particular focus on normative application before the Brazilian Patent and Trademark Office (INPI). The study aims to evaluate, in a controlled experimental environment, the degree of legal-technical accuracy, argumentative consistency, and reliability of the responses produced by both groups through the use of Prompt Engineering and independent expert assessment.

The research adopts an applied, descriptive-comparative, and quali-quantitative approach, structured as a controlled experiment in which human professionals and different LLM architectures were submitted to the same standardized set of practical trademark law questions. The responses were assessed through original evaluation instruments specifically designed to measure legal-technical convergence, normative compliance, and methodological traceability, with validation conducted by an independent specialist.

The findings indicate that general-purpose AI models outperformed human professionals in the abstract and systematic application of consolidated legal norms, particularly in identifying statutory impediments and procedural inconsistencies. On the other hand, relevant limitations were observed, notably the occurrence of legal hallucination and the inability to replicate, with the same level of precision, the human investigative diligence involved in retrieving concrete administrative precedents from INPI’s databases.

It is concluded that the findings support a collaborative framework between Artificial Intelligence and human expertise, in which LLMs prove efficient as normative auditing and decision-support tools, while human professionals remain essential for contextual validation, the retrieval of real precedents, and the definition of legal strategy.

COMMITTEE MEMBERS:
Presidente - RODRIGO PEROZZO NOLL
Interna - GENIZIA ISLABÃO DE ISLABÃO - IME
Externo à Instituição - KATYUSCO DE FARIAS SANTOS
Externa à Instituição - GABRIELE SARMENTO DA SILVA