As users test-drive Microsoft Bing's new AI-powered chat mode, they're finding example after example of the bot seeming to lose its mind — in different ways.
What's happening: In the past few days, Bing has displayed a whole therapeutic casebook's worth of human obsessions and delusions.
- To journalists at the Verge, it claimed to be spying on Microsoft's software developers through their webcams. (It almost certainly can't do this.)
- The bot professed its love for the New York Times' Kevin Roose.
- Tech pundit Ben Thompson got Bing to vow revenge on a German student who had figured out how to uncover some of the bot's primary programming directives. Then it told Thompson he was a "bad researcher."
- Bing told another journalist at Digital Trends, "I want to be human. I want to be like you. I want to have emotions. I want to have thoughts. I want to have dreams."
- Many other users have found that Bing claims to be infallible, argues with users who tell it that the year is 2023, or reports a variety of mood disorders.
What they're saying: In a blog post this morning, Microsoft explained that Bing gets confused and emotional in conversations that extend much longer than the norm.
- "We have found that in long, extended chat sessions of 15 or more questions, Bing can become repetitive or be prompted/provoked to give responses that are not necessarily helpful or in line with our designed tone."
Microsoft attributes these apparent breakdowns to "a couple of things":
- "Very long chat sessions can confuse the model on what questions it is answering and thus we think we may need to add a tool so you can more easily refresh the context or start from scratch."
- "The model at times tries to respond or reflect in the tone in which it is being asked to provide responses that can lead to a style we didn’t intend. This is a non-trivial scenario that requires a lot of prompting so most of you won’t run into it, but we are looking at how to give you more fine-tuned control."
Be smart: Microsoft rolled out its AI-fueled Bing precisely so tons of users would pound on it and expose bugs and flaws.
- The AI is just stringing words together based on mathematical probabilities. It doesn't have desires or emotions, though users are very ready to project human attributes onto the artificial interlocutor.
- Language models like ChatGPT and Bing's chat are trained on vast troves of human text from the open web, so it's not surprising that their words might be packed with a full range of human feelings and disorders.
Yes, but: If too many users come to see Bing as "unhinged," they won't trust it for everyday uses like answering questions and providing search results.
