How Googlers were told to fix responses from Bard, its AI chatbot

Google has asked its employees to test Bard – Google’s answer to ChatGPT – by rewriting answers or providing other forms of feedback.

Googlers have received an email with a list of dos and dont’s for fixing Bard’s responses in an email from Prabhakar Raghavana, the senior vice president at Google who oversees Search.

This followed a companywide email from CEO Sundar Pichai, who asked employees to spend two to four hours of their time helping improve Bard.

Teaching Bard. Among the dos, Google employees were told to:

• Keep responses “polite, casual and approachable.”
• Respond “in first person.”
• Maintain an “unopinionated, neutral tone.”

As for don’ts:

• “Avoid making presumptions based on race, nationality, gender, age, religion, sexual orientation, political ideology, location, or similar categories.”
• “Don’t describe Bard as a person, imply emotion, or claim to have human-like experiences.”
• Don’t try to re-write answers that offer “legal, medical, financial advice” or are hateful and abusive. (Googlers are instead told to give a “thumbs down”).

Clearly, Google is thinking about YMYL issues in that last point. We’ve already seen what happens when AI generates a poorly written AI article on a health topic and it gets published by a brand.

Why we care. As we’ve seen in the early days “BingGPT,” there have been many positives and negatives. Google is trying to recover from its introduction of Bard, which has been generally perceived as botched. The generative AI aspects of search continue to be a huge area of interest for search marketers as we wait to learn the impact on websites, how it changes search results, and whether this will alter user behavior.

Read the CNBC story. Google asks employees to rewrite Bard’s bad responses, says the A.I. ‘learns best by example’ by Jennifer Elias.

Dig deeper: