Mind Mapper: Adaptive AI Systems That Understand How You Think

Today, we're introducing Mind Mapper, a wearable AI system that learns your verbal behavioral pattern to help you overcome your vulnerabilities.

Published at ACM IUI 2026, Mind Mapper explores how AI can learn the patterns behind your behavior in a privacy-preserving manner while observing the way you naturally speak. Wearing a smartwatch for 6-12 days, our system discovered behavioral patterns like:

• "When presented with scarcity cues or limited-time offers, I tend to prioritize immediate purchase over long-term fit"
• "When learning new concepts, I often prefer hands-on examples over abstract explanations"
• "When faced with conflict, I tend to withdraw rather than speak up"

Participants rated these patterns as accurate (86%), uniquely personal, and useful. Nearly half said seeing their patterns made them want to change their behavior.

Cognitive Mapping for AI Support & Scaffolding

Current AI assistants are impressive—but fundamentally reactive. They respond when you ask, remember what you've told them, and optimize for your stated preferences. What they don't do is build an understanding of how you think.

Instead, by analyzing everyday conversations through a privacy-preserving pipeline, Mind Mapper builds interpretable if-then rules: "If X situation happens, then Y behavior occurs." The patterns are built by accumulating evidence over time to predict how you'll likely respond in recurring contexts. This enables a range of applications from anticipatory support, content adaptation, reflection, and training new patterns to simulation of likely behavior.

How It Works

Mind Mapper runs on a Samsung Galaxy Watch, transcribing speech in real-time with aggressive privacy protections:

• Raw audio is never stored—only transcribed text
• All personally identifiable information is automatically anonymized
• Everything stays encrypted on your device, accessible only via your password
• You review and approve all data before any analysis happens
• All processing uses zero-data-retention API settings

The behavioral mining itself uses a multi-stage LLM architecture that generates, tests, and refines patterns with supporting evidence—producing human-readable rules rather than black-box predictions.

What We Learned

We deployed Mind Mapper with 15 participants (12 completed) for 6-12 days each, capturing over 700 hours of real-world conversations. The results validated both the technical approach and the human value:

• Accuracy: Patterns were rated significantly more accurate than baseline LLM approaches (86% vs 76%)
• Uniqueness: When shown anonymized patterns from other participants, people identified their own 81% of the time
• Usefulness: 55% rated patterns as useful or very useful for reflection and behavior change
• Emotional impact: While mostly neutral-to-positive, negative emotional reactions strongly correlated with desire to change (participants confronted with uncomfortable truths often wanted to improve)

Perhaps most tellingly, participants described the experience as "surprisingly accurate," "eye-opening," and occasionally "uncomfortably honest"—the kind of feedback you get when something truly understands you.

Looking Forward

Mind Mapper represents a shift from AI that remembers facts about you to AI that understands how you think. This isn't just about better personalization—it's about cognitive support that meets you where you are.

The research opens pathways for:

• Just-in-time interventions that prevent rather than correct mistakes
• Educational systems that adapt to individual reasoning styles
• Therapeutic tools that recognize and help reframe maladaptive patterns
• Decision aids that account for your specific biases

Of course, with power comes responsibility. Detailed behavioral models could enable manipulation if misused. Our work prioritizes local data ownership, transparent inference, and user control—but the field will need continued vigilance as these capabilities mature.

The Full Story

Mind Mapper will be presented at ACM IUI 2026 on March 23, with our full research available here. We're also releasing the system implementation as open source at github.com/mlfluidinterfaces/mindmapper.

We believe the most powerful AI won't just be smart—it will be understanding. Not in some sentient sense, but in the practical way a good teacher, coach, or friend understands: knowing not just what you said, but how you tend to think, and offering support shaped by that knowledge.