Google has asked all employees to spend two to four hours manually correcting answers from chatbot Bard before release. This should prevent errors when the chatbot is released.
Google hopes that if employees take the time to correct Bard, the chatbot’s answers will be better, reports CNBC. The email also contained a manual on how employees should correct the chatbot and how answers from the chatbot should sound. Google emphasizes that the chatbot should not appear to display any emotion.
It is customary for employees to test a new function internally, but now Bard is under more scrutiny after an error in the demonstration last week. In addition, the chatbot said that the James Webb telescope was the first to photograph an exoplanet, but that is not true.
It is not the first time that AI systems from large tech companies have to rely on human corrections. All tech companies with digital assistants, like Apple Siri, Amazon Alexa, Microsoft Cortana and Google Assistant, relied on at least until 2019 human corrections to correctly interpret voice commands .
Google wants to release Bard soon, as Microsoft is in the process of giving users access to the new Bing integrating OpenAI’s GPT technology. That demonstration also contained a series of inaccuracies, while users now also getting weird answers. Tweakers published a on Thursday backstory on search engine chatbots.
