Generative AI may seem magical, but behind its development are specialized professionals, such as engineers and analysts, who work methodically to create it.
Their role is to evaluate the accuracy of chatbots, contributing to the continuous improvement of their performance.
Read also: Apple & Google likely to collaborate to create artificial intelligence on iPhones
In this context, concerns are being raised around Gemini, Google's advanced AI system, after the implementation of new guidelines towards contractors involved in its development.
The contractors, working through Hitachi subsidiary GlobalLogic, evaluate the AI models’ responses based on parameters such as “truthfulness.” In the past, they could skip prompts that required specialized knowledge — in cardiology, for example — when they didn’t have the necessary scientific training. However, under Google, that ability has been limited.
Now, contractors are asked to rate the parts of the answer they understand and note that they do not have specialized knowledge. They are only allowed to skip a prompt if it is missing critical information or contains harmful content.
See more: Catly has no Generative AI, Blockchain or NFTs
The previous guideline explicitly stated: “If you do not have the specialized knowledge (e.g., programming, mathematics) to evaluate the prompt, please skip it.” This policy was designed to ensure greater accuracy by referring specialized issues to competent evaluators. Instead, the new guideline raises concerns, especially in sensitive areas like health.
Contractors have expressed concerns about the reduced reliability of responses when asked to assess issues outside their expertise. One of them commented characteristically: “I thought the bypass was there to ensure accuracy by referring the issue to someone more specialized.”
Read more: China unveils rules for genetic artificial intelligence (AI)
This policy change brings to the fore concerns about whether Gemini will be able to provide reliable information to users, particularly on technical or specialized issues, such as rare diseases.
Source: techcrunch
