Confidence Interval - An AI tool for learning critical thinking (vaughntan.org)

CONFIDENCE INTERVAL is a new web app designed to teach critical thinking by inverting the usual LLM-chatbox dynamic: instead of generating polished essays for students, it uses an "iterative scaffold" of carefully sequenced prompts and a LLM-powered “Socratic mirror” to reflect and reframe a user’s own claims. The tool guides students step-by-step to surface assumptions, sharpen value judgments, consider audiences, and strengthen evidence—keeping humans in charge of all subjective decisions while using AI for information retrieval, reframing, and systematic prompting rather than content generation. Early tests with first‑year undergraduates showed dramatic results: in a single two‑hour session, students turned vague proposals into focused, well-justified arguments suitable for academic or policy use. This approach matters because it addresses a growing educational problem: students are becoming passive consumers of machine-generated output and losing the practice of meaningmaking—the uniquely human skill of deciding what matters and why. CONFIDENCE INTERVAL encodes pedagogical best practices so critical-reasoning tutoring can scale without intensive one-on-one mentoring, preserves human agency, and demonstrates a broader design principle for AI tools that explicitly separate what machines do well (organize, prompt, retrieve) from what only humans should do (value judgments, judgement of evidence, strategic tradeoffs). The project is supported by the Future of Life Foundation’s AI for Human Reasoning program and is available for beta testing.