This thesis examines the relationship between artificial intelligence and the translator, considering whether they compete with or complement each other. Using a deductive approach, it argues that if AI systems are not designed to perform certain functions required for translating texts in specific domains, their outputs cannot achieve parity with human translations.

The hypotheses derived from this reasoning are tested using human evaluation methods, specifically the BWS metric, to assess outputs generated by Google Gemini-2.5-Pro through advanced prompting. These outputs are compared with human translations of expressive and informative texts.

The results show that MT evaluation remains methodologically unstable, as outcomes are highly sensitive to evaluation procedures. While AI demonstrates promising performance in the translation of informative texts, evaluation results for expressive texts are inconsistent. Overall, the study concludes that generative AI should be regarded as an extension of the translator’s toolkit rather than a substitute for human translation, and that humans and AI are better understood as complementary rather than competitive.