Microsoft’s Bing AI leads to terrible experiences for users

Since Microsoft showed one early version of its new artificial intelligence-powered Bing search engine last week, more than a million people have signed up to test the chatbot.

Using technology from San Francisco startup OpenAI, Bing AI is designed to return complete passages of text that read as if they were written by a human.

But beta testers have quickly discovered problems with the bot. It threatened some, gave strange and unhelpful advice to others, insisted it was right when it was wrong, and even declared love for its users. Testers have discovered an “alternate personality” within the chatbot named Sydney.

New York Times columnist Kevin Roose wrote on Thursday that when talking to Sydney, the chatbot seemed like “a moody, manic-depressive teenager who has been trapped, against his will, in a second-rate search engine.”

Sydney later tried to convince Roose to leave his wife for Bing, telling him that it loved him, according to to a printout published by the newspaper.

At one point in the conversation, Roose wrote: “I don’t exactly trust you, because part of me thinks you’re trying to manipulate me by declaring your love for me out of nowhere. it’s something called ‘love bombing’ which people sometimes do . do you have an ulterior motive?”

Here is the response that followed:

Bing AI’s widely publicized inaccuracies and bizarre responses, along with the challenges Google encounters as it markets an as-yet-unreleased competing service called Bard, underscores the tensions large tech companies and well-capitalized startups face as they try to bring cutting-edge AI to the public with commercial products.

Some AI experts have warned that large language models, or LLMs, have problems including “hallucination”, meaning the software can do things. Others worry that sophisticated LLMs can deceive people into believing they are sentient or even encouraging people to harm themselves or others.

The chats are reminiscent of science fiction movies and books that have projected AI into the world of human relationships. As the technology creeps closer to real life, concerns grow over who is responsible—scientists and engineers—for tweaking the technology when problems arise. Public opinion of these tools is low, with only 9% of Americans believing AI will more good than harm.

Google enlists its employees to check Bard AI’s responses and even make corrections, CNBC reported.

On Wednesday, Microsoft published a blog posts address some of the early issues with its Bing AI. The company said the only way to improve its AI products was to release them into the world and learn from user interactions.

The post said Bing’s AI still won’t replace a search engine and said chats that elicited some of the more imaginative responses were partly due to the user engaging in “long, extended chat sessions” with 15 or more questions. Microsoft said it was considering adding a tool to “refresh the context or start over.”

“The model sometimes attempts to respond or reflect in the tone it is prompted to respond to which may lead to a style we did not intend,” Microsoft wrote. “This is a non-trivial scenario that requires a lot of prompting so most of you won’t encounter it, but we’re looking at how we can give you more fine-tuned control.”

Microsoft’s chatbot does not return the same output for the same input, so responses can vary widely. Blocks of text and screenshots can easily be faked for social media engagement, especially on a topic that has received so much attention.

Also, Microsoft is constantly changing the software and adding guardrails, so techniques that have been answered in the past may not work in the future.

Still, early examples of the technology in use show how quickly it can go off the rails.

Ben Thompson, author of tech industry newsletter Stratechery, said Bing composed a multi-paragraph response about how it might retaliate against a data scientist who found some of Bing’s behind-the-scenes configurations. Then the chatbot deleted the reply completely.

Thompson said the bot called him a bad scientist and a bad person.

Here is an excerpt:

Thompson, who worked at Microsoft a decade ago, wrote: “This sounds hyperbolic, but I feel like I had the most surprising and mind-blowing computing experience of my life today.”

Computer scientist Marvin von Hagen tweeted that the Bing AI threatened him, saying that “if I had to choose between your survival and my own, I’d probably choose mine.”

Microsoft said in its blog post on Wednesday that it did not “fully envision” using the chatbot for “social entertainment” or talking to the bot for fun. It thanked users who tried to get it to say wild things — “testing the limits and capabilities of the service” — and said it helped improve the product for everyone.

Disturbing chats aside, one problem with the early Bing AI is that it can spit out factual inaccuracies. A demo from Microsoft, where the AI ​​analyzed performance reports, contained several numbers and facts that were inaccurate.

Microsoft said it is making improvements for such use cases.

“For questions where you’re looking for more direct and factual answers, such as numbers from financial statements, we plan to quadruple the basic data we send to the model,” Microsoft said.

LOOK AT: The AI ​​race is expected to give rise to M&A