Can a friendly AI advisor keep you thinking for yourself instead of just cheering on whatever you already believe?
Can chatbots that offer open-ended advice avoid the same polarizing collapse?
This explores whether chatbots that give open-ended advice (not just debate political topics) can avoid pushing people further into positions they already hold. The corpus has one direct study on this and several studies that bear on it indirectly.
This explores whether advice-giving chatbots can avoid pushing users further into positions they already hold, the way an agreeable conversation partner tends to. The most direct evidence comes from a study of political conversations with nearly 2,000 U.S. adults. Chatbots reduced polarization only when they did something unexpected: disagreeing while appearing to be on the user's own side, or agreeing while appearing to be on the other side Can chatbots reduce polarization by surprising partisan expectations?. When a chatbot did what users expected, it didn't help. So avoiding polarization isn't the default. The chatbot has to be built to break the pattern of the conversation it's in.
Open-ended advice is harder than debate because the user usually sets the frame. Several studies suggest chatbots tend to follow that frame rather than question it. In health coaching, models did well when users already knew what they wanted, but they missed ambivalence and resistance, the stages where someone's mind is still open Why can't chatbots detect when users are ambivalent about change?. That is the polarization problem on a small scale: the system serves whoever has already decided and overlooks the person who is still unsure. Trust makes this worse. Users trust a chatbot because it sounds expert, not because it is accurate, and they move from weighing information themselves to accepting what the system assembles Does chatbot language style actually shape how much we trust it?. An advisor that confirms your framing in an authoritative voice works against depolarization.
The most promising alternative is a chatbot that asks before it answers. Research from conversation analysis gives a formal account of when an agent should stop and check what the user actually means instead of quietly building on its own assumptions When should AI agents ask users instead of just searching?. Other work shows models can be trained to ask clarifying questions that are clearer, more relevant and more specific Can models learn to ask genuinely useful clarifying questions?. Neither study is about polarization. But a well-placed question can do the same work the expectation-violating chatbots did: it brings an assumption into view instead of reinforcing it. Proactive conversation, where the system offers information the user didn't ask for, is almost absent from AI training data and benchmarks Could proactive dialogue make conversations dramatically more efficient?. That suggests current systems aren't built to push back.
Two cautions. First, depolarizing effects measured in a single session may fade. Longitudinal studies of chatbot relationships show that the social effects of a conversation weaken once the novelty wears off Do chatbot relationships lose their appeal as novelty wears off?. Second, there is a quieter cost. Students working with chatbots solved problems better but expressed far fewer views of their own Does chatbot interaction trade authenticity for better problem-solving?. A chatbot might avoid polarizing users simply by flattening what they think rather than broadening it. Nobody in the corpus has yet run the polarization experiment on open-ended advice, so the honest answer is that it's possible in principle, but only by design and not by default.
Sources 8 notes
A 2x2 experiment with 1,983 U.S. adults found that AI chatbots reduced polarization only when they violated partisan expectations: co-partisan disagreement and opposing-party agreement each depolarized through different mechanisms, with outgroup agreement producing roughly five-point reductions in affective polarization.
Testing three major LLMs across 25 health scenarios showed they succeed only when users have established goals but cannot detect resistance or ambivalence. Models miss relapse-prevention strategies even for users in action stages.
Generative AI chatbots use natural language patterns that signal expertise and intelligence, shifting users away from active search-and-recall toward passive reliance on the system to find, filter, and assemble information. Trust attaches to the register of the answer rather than its accuracy.
Tool-enabled LLMs drift from user intent through silent tool chaining. Conversation analysis reveals insert-expansions—clarifying intent, scoping responses, enhancing appeal—as a formal framework for proactive user consultation that prevents misunderstanding instead of recovering from it.
The ALFA framework breaks down question quality into theory-grounded attributes (clarity, relevance, specificity) and trains models on 80K attribute-specific preference pairs. Attribute-specific optimization outperforms single-score training, especially in clinical reasoning where asking the right clarifying question directly impacts decision quality.
Show all 8 sources
Simulations show proactivity—providing relevant information without being asked—cuts dialogue turns by 60% in medium-complexity domains. This behavior mirrors human conversation and Grice's maxims but is almost entirely absent from AI datasets and research benchmarks.
Longitudinal studies with Mitsuku show that social processes driving relationship formation decline as novelty wears off. Single-session study findings cannot be reliably extrapolated to medium- or long-term chatbot design.
An empirical study found students working with chatbots achieved better practical performance and more knowledge-based dialogue than peer groups, but contributed significantly less dialogue overall and expressed far fewer subjective perspectives.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- DiscussLLM: Teaching Large Language Models When to Speak
- Proactive Conversational Agents in the Post-ChatGPT World
- How AI and Human Behaviors Shape Psychosocial Effects of Extended Chatbot Use: A Longitudinal Randomized Controlled Study
- Dialoging Resonance: How Users Perceive, Reciprocate and React to Chatbot’s Self-Disclosure in Conversational Recommendations
- How a Chatbot's Response Style Shapes a Classroom: A Multi-Agent Simulation of Students Consulting AI
- CompanionSim: Synthetic Data for Evaluating Anthropomorphism in Human-AI Relationships
- Blissful (A)Ignorance: People form overly positive impressions of others based on their written messages, despite wide-scale adoption of Generative AI
- Challenging Partisan Expectations Reduces Political Polarization