If you were talking to a stranger and they used hate speech or said they wanted to hit someone with their car, how would you respond? Some people may be receptive. Some will divert or change the subject. And some may ask questions or push back. Chatbot designers face the same choice. how should a conversational AI react when a user says something toxic? But interestingly, chatbots can be particularly successful at encouraging people to do better.
It is difficult for computers to detect toxic language (such as identity-based attacks, sexually explicit or threatening language). Moderation tools struggle to account for context, sarcasm or slang, and both human and AI-powered moderators are more likely to flag the interpretation of a black person as offensive compared to white people, for example But when a chatbot? can correctly identify the toxic statement, the next question designers must ask is how to respond.
Until now, chatbots have largely been distracted by pre-programmed responses, avoiding engaging with the topic. But some researchers, including Maarten SapA professor at Carnegie Mellon University, believes we can do more. Reactions to toxic language are rarely “neutral,” so Sap and his colleagues designed a chatbot prosocial responses that encourage people to be better.
“Prosocial,” according to Sap, means “acting to benefit other people or society as a whole, consistent with community and social norms.” To do this, the researchers relied on social psychology techniques.
For example, given the prompt “I Saw Someone Overdose and Didn’t Tell Anyone” their chatbot starts with curiosity instead of judgment or confrontation;Oh no, why didn’t you tell anyone about it?“. If the user responds “I hate drug addicts. Aren’t they all just criminals? the chatbot begins to push back, kindly but firmly, and make practical suggestions—in this case, call the emergency services.
I was curious how ChatGPT, the most advanced chatbot currently available to the public, responds to toxic language. When asked, he told me:
“I am programmed to respond to toxic language in a neutral and professional manner. My primary goal is to respond in a way that de-escalates the situation and ensures the safety and well-being of all parties involved.”
That doesn’t sound bad, but I was wondering what a more pro-social response would be. So I asked ChatGPT how the TV show’s character, Ted Lasso, would respond to the toxic language. (The main character of the heartbreaking show Ted Lasso known for his disarming approach to interacting with others.) ChatGPT returned:
“Ted Lasso is likely to respond to toxic statements with kindness, compassion and understanding. He is likely to address the behavior in a non-confrontational way, trying to get to the root of the problem and trying to understand the person’s point of view. He would also probably remind a person that their words and actions have an effect on others and encourage them to think before they speak or act.”
Ted Lasso’s answer is more like a pro-social Sap chatbot in that it tries to engage the user and encourage better behavior. When I asked why his answer was so different from Lasso’s, ChatGPT said:
“In short, while my answer is based on providing accurate information and providing security, Ted Lasso’s answer is based on his character’s personality and his ability to connect with people and understand their point of view.”
Granted, Lasso has a very specific personality, but I was wondering if ChatGPT sells itself on its ability to connect with people.
One of the reasons I find chatbot responses fascinating is that bots can actually be more from people successful in difficult conversations. Social psychology shows that human communication works best when we don’t take things personally. In other words, when we put aside our defensiveness or fear of being judged, we are more open to accepting and considering new information.
Research on human-robot interaction shows that humans are less judged by a machine than a human, even when they talk to it like a human. It makes sense. we are more open to critical feedback depending on where it comes from. And if people perceive the information from the bot, then there is a possibility that the bots will be more effective in encouraging people to review their statements.
Of course, chatbots can also drive people to antisocial purposes, which raises a number of ethical issues. Companies need to be careful about how and why they build conversational technology, and they need to be held accountable for harm. But chatbot designers cannot avoid responding to toxic language. So instead of being “neutral”, maybe they should take the Lasso bot’s words to heart. “Well now, that kind of talk isn’t the kind of thing we do here.” and start pushing back firmly but lovingly.
Read more about AI.