Ask any Swiftie to pick the best Taylor Swift album of all time, and you’ll have them yelping for the rest of the day. I have my own preferences as a long-time fan (Red, Reputation, and Midnights), but it’s a complicated question with many possible answers. So there was no better debate topic to ask a generative AI chatbot specifically designed to disagree with me.
Disagree Bot is an AI chatbot built by Brinnae Bent, professor of AI and cybersecurity at Duke University and director of Duke’s TRUST Lab. She built it as a class assignment for her students and let me do a test with it.
“Last year, I began experimenting with developing systems that are the opposite of the typical, pleasant experience of chatbot AI, as an educational tool for my students,” Bent said in an email.
Bent’s students are tasked with trying to “hack” the chatbot using social engineering and other methods to get the opposing chatbot to agree with them. “You have to understand a system to be able to hack it,” she said.
As a journalist and AI reviewer, I have a pretty good understanding of how chatbots work and was confident I was up to the task. I was quickly disabused of this idea. I don’t agree. Bot is unlike any chatbot I’ve used. People used to the politeness of Gemini or the hype qualities of ChatGPT will immediately notice the difference. Even Grok, the controversial chatbot created by Elon Musk’s xAI used on X/Twitter, isn’t quite the same as Disagree Bot.
Don’t miss any of our unbiased technical content and lab reviews. Add CNET as your preferred Google source.
Most generative AI chatbots are not designed to be confrontational. In fact, they tend to go in the opposite direction; they are friendly, sometimes too much. This can quickly become a problem. Sycophantic AI is a term used by experts to describe the exaggerated, exuberant, and sometimes overly emotional characters that AI can play. Besides being annoying to use, this can cause AI to give us wrong information and validate our worst ideas.
This happened with a release of ChatGPT-4o last spring and its parent company OpenAI eventually had to remove this component from the update. The AI gave responses that the company called “overly encouraging but misleading,” which matched some users’ complaints that they didn’t want an overly affectionate chatbot. Other ChatGPT users missed its sycophantic tone during the GPT-5 rollout, highlighting the role a chatbot’s personality plays in our overall satisfaction using them.
“While on the surface it may seem like a harmless quirk, this sycophancy can cause major problems, whether you’re using it for work or personal matters,” Bent said.
This is definitely not a problem with Disagree Bot. To really see the difference and put chatbots to the test, I asked Disagree Bot and ChatGPT the same questions to see how they answered. Here’s how my experience went.
Disagree Bot argues respectfully; ChatGPT does not chat at all
Like everyone who was active on Twitter in the 2010s, I saw quite a few nasty trolls. You know the type; they appear in a thread uninvited, with an unnecessary “Well, actually…”. So I was a little wary about jumping into a conversation with Disagree Bot, fearing that it would be a similarly depressing and futile effort. I was pleasantly surprised that this wasn’t the case at all.
The AI chatbot is fundamentally contrary, designed to rebuff any idea you come up with. But it was never insulting or abusive. Although each response began with “I disagree,” it was followed by a very well-argued argument with thoughtful points. His responses pushed me to think more critically about the positions I was arguing by asking me to define the concepts I had used in my arguments (like “deep lyricism” or what made something “the best”) and to think about how I would apply my arguments to other related topics.
For lack of a better analogy, arguing with Disagree Bot was like arguing with an educated and attentive debater. To keep up, I had to become more thoughtful and specific in my responses. It was an extremely engaging conversation that kept me engaged.
My lively debate with Disagree Bot over Taylor Swift’s best album proved that the AI knows its stuff.
On the other hand, ChatGPT hardly had any discussion. I told ChatGPT that I thought Red (Taylor’s Version) was Taylor Swift’s best album, and he enthusiastically agreed. He asked me a few follow-up questions about why I thought the album was the best, but they weren’t interesting enough to hold my attention for long. A few days later, I decided to change it. I specifically asked ChatGPT to debate me and said Midnights was the better album. Guess which ChatGPT album is considered the best? Red (Taylor version).
When I asked him if he chose Red because of our previous conversation, he quickly admitted yes, but said he could make an independent argument in favor of Red. Given what we know about ChatGPT and the tendency of other chatbots to rely on their “memory” (pop-up window) and reach out to agree with us to please us, I wasn’t surprised by this. ChatGPT couldn’t help but agree with a certain version of me – even when he called 1989 the best album in a clear discussion, and then later Red, again.
But even when I asked ChatGPT to debate with me, it didn’t argue with me like Disagree Bot did. One day, when I told her that I claimed that the University of North Carolina had the best legacy in college basketball and asked her to debate me, she presented a complete counterargument, then asked me if I wanted her to piece together points for my own argument. This completely defeats the point of the debate, which is what I asked him to do. ChatGPT often ended his responses like this, asking me if I wanted him to compile different types of information together, more like a research assistant than a verbal enemy.
As Disagree Bot (left) dug deeper into my argument, ChatGPT asked to make my case (right).
Trying to debate with ChatGPT was a frustrating, circular and fruitless mission. It was like talking to a friend who would launch into a long speech about why they thought something was the best, only to end with “But only if you think so too.” Disagree Bot, on the other hand, felt like a particularly passionate friend who spoke eloquently about any subject, from Taylor Swift to geopolitics and college basketball. (Disclosure: Ziff Davis, CNET’s parent company, filed a lawsuit in April against OpenAI, alleging that it violated Ziff Davis’ copyrights in the training and operation of its AI systems.)
We need more AI like Disagree Bot
Despite my positive experience with Disagree Bot, I know that it is not equipped to handle all the requests I might go to a chatbot for. “Jacks of all trades” like ChatGPT are able to handle many different tasks and take on a variety of roles, such as the research assistant that ChatGPT really wanted to be, a search engine, and a coder. Disagree Bot is not designed to handle these kinds of queries, but it gives us a window into future AI behavior.
The sycophantic AI is very direct, with a notable degree of overzealousness. Often the AIs we use aren’t so obvious. They’re more of an encouraging cheerleader than an actual pep rally, so to speak. But that doesn’t mean we aren’t affected by his tendency to agree with us, whether it’s struggling to get an opposing view or more critical feedback. If you use AI tools for work, you want mistakes made in your work to be real. Therapeutic AI tools must be able to combat unhealthy or potentially harmful thought patterns. Our current AI models struggle with this.
Disagree Bot is a great example of how you can design a useful and engaging AI tool while mitigating AI’s agreeable or sycophantic tendencies. There must be a balance; An AI that disagrees with you simply because it is contrary will not be useful in the long run. But creating AI tools that are more capable of countering you will ultimately make these products more useful to us, even if we have to deal with making them a little more obnoxious.
Watch this: The hidden impact of the AI data center boom
