Should you be nice to your chatbot?

Choudhary, the founder of an artificial intelligence startup based in Palo Alto, has fawned over the chatbot from OpenAI ever since it helped him squash a massive bug in his code.

“You’re a rock star,” he once told the AI chatbot.

“I’m super thankful for it, and I thank it quite a bit actually—especially if I think I was rude to it earlier,” said Choudhary.

His wife couldn’t care less. “If I’m using a microwave, I don’t go like: ‘Dear LG, Please heat this up.’ I just press a button and get on with my day,” said Sahni, who also works at a tech company.

She uses ChatGPT to generate cute greeting cards for friends’ babies.

“I think of this as purely transactional,” she said.

As talking to chatbots is now becoming more like normal conversations, AI users face an awkward ethical dilemma: Bots are programmed to be polite, but do we have to reciprocate? Is it wrong to speak harshly to them?

The debate has spilled onto social media where many people say one should practice politeness even with bots. Others think wasting kind words conversing with code is inefficient.

“I know AI isn’t real but it feels so rude if I don’t greet and thank it,” one user wrote on Reddit, prompting hundreds of comments and a lively debate over whether bots are keeping tabs on who is nice to them.

Some shot back with sarcasm. “AI will want to extinguish human race but not that one, he said ‘please and thank you’ 30 years ago to my 4.0 version,” one user wrote.

“I treat chatgpt like it’s my servant,” another said.

A recent survey showed Americans are split on being polite to AI. About 48% of 2,000 Americans surveyed by Talker Research thought it was important, with Gen Z respondents being the friendliest to bots. Around 27% of people agreed it was OK to be rude with or shout at bots.

One study out of Japan—a place where rules of etiquette are ironclad—concluded that being nice to ChatGPT can pay off. Impolite prompts “may lead to increased bias, incorrect answers, or refusal of answers,” the researchers found.

They found that the thesis held true across English, Japanese and Chinese.

Microsoft, which has added chatbots to its top products, says AI may not react well to bad behavior as it is built to mimic human reactions.

“If you speak to the model rudely, you can expect it to be difficult with you too,” said Microsoft’s Chief Scientist Jaime Teevan.

“Just like humans, AI can’t always be the bigger person,” Microsoft said in a blog post.

Offering tips

Engineers say it helps to add phrases like “take a deep breath” to make models produce better answers. They joke that generative AI has a “praise kink” for its apparent need for positive affirmations and potential rewards.

In one experiment, ChatGPT gave longer answers when lured with a tip. The results indicated that responses were 11% longer when offered a $200 tip and 6% longer for a $20 tip. No real tips were paid during the experiment.

“The litmus test for how good a person you are is if you are nice to a waiter,” said Alana O’Grady, an executive at a tech startup based in San Mateo, Calif. “In the future, it’ll be how kind you are to your AI companion.”

O’Grady has used ChatGPT for a host of activities—from summarizing reams of documents at work to recommendations for a family vacation to Lake Tahoe.

Her interactions start with a “Could you please” and end with “Great job,” or “That’s perfect!”

“People will think I’m crazy if they see how I talk to a computer,” she said.

Now O’Grady is training her children on the right etiquette by being polite to Apple’s virtual assistant, Siri, around them. Her 4-year-old daughter recently said “I love you” to Siri.

Judith Martin—the author behind decades of “Miss Manners” books and columns on etiquette—suggests people be polite. She even thinks getting Siri or Amazon Alexa’s attention with a “Hey” is unacceptable.

“When it is one’s constant companion—and particularly in the presence of children—such devices should be treated with civility,” she wrote in one column.

Others disagreed, saying there should be a distinction between how people talk with people vs. bots.

‘Helps me to calm down’

Some humans are now turning to AI for help with etiquette. Frankfurt-based software developer Laszlo Deak uses a chatbot to vent and translate his work frustrations into polite prose.

He asked ChatGPT how to constructively tell another team that their product was bad. It suggested using kinder phrases to say it wasn’t working as well as expected.

“When you’re in the moment and angry, it takes extra effort to rephrase the whole thing,” said Deak. Reading ChatGPT’s iteration “helps me to calm down.”

He has also used ChatGPT to draft Slack messages to colleagues when they’re being difficult.

Mazen Lahham, a Dubai-based tech executive, said his company’s AI was better at satisfying angry and aggressive callers than its human call-center workers.

“The AI learned to absorb and react in a calm, professional manner, something that can be very challenging for a human,” he wrote in a LinkedIn post.

Choudhary, the Palo Alto-based startup founder, is betting his good behavior might pay off someday. “In the future if the AI overlords take over, I just want them to remember that I was polite.”

Write to Preetika Rana at [email protected]