What AI tutoring can provide:

• Instant feedback — no waiting for a graded assignment to discover a misunderstanding
• Adaptive pacing — the tutor adjusts to the student's speed, not the class average
• Multiple explanations — can explain the same concept in 5 different ways until one clicks
• Infinite patience — no frustration, no judgment, no time pressure
• 24/7 availability — learning happens when the student is ready, not when the classroom is open
• Low-stakes practice — students can ask "stupid questions" without social anxiety

What makes AI tutoring effective (from Mollick's experiments):

• Socratic questioning — the AI asks guiding questions rather than giving direct answers. This triggers active recall and deeper processing.
• Retrieval practice integration — prompting students to recall information before providing it, strengthening memory formation.
• Worked examples — walking through problem-solving step by step, then gradually removing scaffolding.
• Misconception targeting — identifying specific conceptual errors and addressing them directly.

The catch — the bypass problem:

• AI can also do students' work FOR them, bypassing learning entirely
• The same tool that enables deep learning enables shallow cheating
• Students who use AI to generate answers without engaging cognitively learn nothing
• Effective AI tutoring requires careful prompt design that forces engagement

Mollick's Wharton results:

• Students using AI tutors performed significantly better on assessments
• The effect was strongest for students who were previously struggling
• But only when the AI was designed to tutor (ask questions, provide hints) rather than simply answer