Google asks staff to help fix AI’s ‘bad responses’ manually

Search engine giant, Google, is turning to its human employees to fix ‘bad responses’ given by its artificial intelligence (AI) chatbot Bard. Company CEO, Sundar Pichai, had asked employees to spend two-four hours every day to help improve the product.

Microsoft, which works closely with Open AI, has already adopted GPT-4 to power the chatbot in its Bing search engine, which could end up threatening Google’s dominance in the search engine market. Google unveiled its AI chatbot Bard to rival the growing popularity of OpenAI’s product.

However, Bard gave some inaccurate responses during a promotional event, which tanked Google’s stock prices by nine percent. Google is now on the back foot as it looks to demonstrate that its upcoming AI products are equally good.

How Google plans to fix Ba(r)d AI’s bad responses
Now, in an email sent by Prabhakar Raghavan, the vice president for search at Google, has asked employees to help the company’s chatbot get its answers right.

In the email, Raghavan wrote that Bard’s technology was exciting but still in its early days. The company was keen to get it right, and by participating in the internal testing, employees would help accelerate the model’s training and test out its load capacity.

Raghavan encouraged Google staffers to rewrite answers on topics they understood well but also provided them with a list of do’s and don’ts to consider before teaching the chatbot.

Among the do’s are instructions to keep the responses polite, casual, and approachable while ensuring that they are in the first person and using an unopinionated and neutral tone.

Employees were also told not to stereotype and avoid making presumptions based on race, nationality, gender, age, religion, sexual orientation, political ideology, location, or similar categories, CNBC said in its report.

Google is also trying to ensure that its chatbot does not come across as a person or display emotions, perhaps from their previous experience of a claim that the chatbot was sentient. Google has also instructed employees to downvote answers that offer legal, medical, or financial advice but insisted on not improving them.

For their efforts, Googlers will receive a “Moma badge” which they can display on their internal employee profile. The top 10 contributors will also be invited to a listening session, where they will meet with Bard’s team and have the chance to share their feedback live with them.

Employees have criticized CEO Pichai for rushing Bard’s rollout and even calling it short-sighted. Interesting Engineering