People training new AI models admit they just get chatbots to do it, and experts warn this kind of ‘AI inbreeding’ could make future systems less useful than the companies expect

The approach to creating artificial intelligence has come under question following revelations that some of the people hired to do the job are secretly employing chatbots that already exist to carry out their duties. It has been noted that this could lead to future AI systems being lower in quality due to the inclusion of machine-made material into the human-created datasets.

According to reporting by New Scientist, some of the contractors who have been tasked with doing AI training said that they use chatbots like ChatGPT when carrying out conversation-related work and evaluation exercises rather than writing answers themselves.