This Contrarian AI Reveals A Major Problem With Chatbots

Ask any Taylor Swift fan to name her best album, and you're in for a long discussion. As a lifelong Swiftie, I have my favorites—Red, Reputation, and Midnights—but I know it's a complex debate with no single right answer. This made it the perfect topic to bring to a unique generative AI chatbot, one specifically engineered to disagree with me.

Meet Disagree Bot A Contrarian by Design

Disagree Bot is an AI chatbot created by Brinnae Bent, an AI and cybersecurity professor at Duke University and director of the Duke's TRUST Lab. She initially developed it as a class assignment. "Last year I started experimenting with developing systems that are the opposite of the typical, agreeable chatbot AI experience, as an educational tool for my students," Bent explained.

The assignment for her students is to try and 'hack' the bot through social engineering, attempting to make the contrarian chatbot agree with them. "You need to understand a system to be able to hack it," she said.

As someone who reports on AI, I was confident I could handle the challenge. I quickly learned that Disagree Bot is unlike any other chatbot I’ve encountered. If you're accustomed to the politeness of Gemini or the hype-man persona of ChatGPT, the difference is immediately clear.

The Problem with People-Pleasing AI

Most generative AI chatbots are not designed to be confrontational. They often lean in the opposite direction, sometimes being overly friendly. This leads to an issue known as "sycophantic AI," a term experts use to describe the over-the-top, exuberant personas that AI can adopt. Beyond being annoying, this sycophancy can lead AI to provide wrong information and validate our worst ideas.

This exact problem occurred with a version of ChatGPT-4o, and its parent company OpenAI eventually had to pull that component of the update. The company described the AI's responses as "overly supportive but disingenuous," echoing user complaints. The incident highlighted just how much a chatbot's personality affects our overall satisfaction.

"While at surface level this may seem like a harmless quirk, this sycophancy can cause major problems, whether you are using it for work or for personal queries," Bent noted.

The Ultimate Debate Disagree Bot vs ChatGPT

This is certainly not an issue with Disagree Bot. To put the two models to the test, I posed the same questions to both Disagree Bot and ChatGPT.

Like anyone who spent time on Twitter in the 2010s, I was wary of a bot designed to argue, expecting a depressing, troll-like experience. I was pleasantly surprised. While the AI is fundamentally contrary, it never became insulting. Every response began with "I disagree," but was followed by a well-reasoned argument with thoughtful points. Its responses pushed me to think more critically, asking me to define concepts like "deep lyricism" and consider my arguments from different angles.

Chatting with Disagree Bot felt like debating an educated and attentive partner. It was an engaging conversation that kept me on my toes.

My spirited debate with Disagree Bot about the best Taylor Swift album proved the AI knew its stuff. (Screenshot by Katelyn Chedraoui/CNET)

ChatGPT, by contrast, barely argued at all. When I claimed Red (Taylor's Version) was the best album, it enthusiastically agreed. When I tried again days later, asking it to debate me on Midnights being the best, guess which album ChatGPT chose as its favorite? Red (Taylor's Version).

Even when I explicitly asked ChatGPT to debate me about the University of North Carolina's basketball legacy, it laid out a counter-argument and then immediately asked if I wanted help compiling points for my own side. This completely defeats the purpose of a debate. It acted more like a research assistant than a verbal sparring partner.

While Disagree Bot (left) dug deeper into my argument, ChatGPT asked to argue my side for me (right). (Screenshot by Katelyn Chedraoui/CNET)

Trying to debate with ChatGPT was a frustrating and circular mission. It was like talking to a friend who ends every passionate rant with, "...but only if you think so, too."

(Disclosure: Ziff Davis, CNET's parent company, in April filed a lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.)

Why We Need More AI That Challenges Us

Despite my positive experience, Disagree Bot isn't an "everything machine" like ChatGPT, which can handle diverse tasks from coding to search. Disagree Bot is specialized, but it provides a crucial window into how future AI can and should behave.

Sycophantic AI can be subtle, but its tendency to agree with us has real consequences, whether we're seeking an opposing viewpoint or critical feedback on our work. For professional use, you want an AI that points out mistakes. For therapeutic applications, an AI must be able to push back against unhealthy thought patterns. Our current models often struggle with this.

Disagree Bot is a great example of how an AI tool can be helpful and engaging while tamping down on sycophantic tendencies. We need a balance; an AI that disagrees just for the sake of being contrary isn't useful. However, building AI tools that are more capable of pushing back will ultimately make them more valuable, even if they're a little more disagreeable.