Source-linked AI summary
Assigning AI: Seven Approaches for Students, with Prompts
Ethan Mollick, Lilach Mollick
TL;DR
AI can support learning but introduces risks that require deliberate classroom guidance. This paper outlines instructional approaches and prompts for using AI while keeping students responsible for evaluating and directing its contributions.
Problem
Students may struggle to build mental models or consider alternative pathways when reflecting on experiences solely from their own viewpoint.
Method
The paper develops classroom approaches, prompts, and guidelines that position AI as a tutor, coach, or team complement while directing students’ interactions.
Results
The paper presents AI-assisted tutoring, coaching, teamwork, and active-learning practices for helping students use AI responsibly in and beyond class.
Takeaways & Limitations
Educators can pair AI-supported learning outside class with in-class discussion, reasoning, collaboration, and firsthand examination of AI’s implications.
Takeaways & Limitations
AI tutoring can produce plausible-seeming incorrect answers, so students need explicit instruction about its limitations.
Abstract
from arXiv · showhide
This paper examines the transformative role of Large Language Models (LLMs) in education and their potential as learning tools, despite their inherent risks and limitations. The authors propose seven approaches for utilizing AI in classrooms: AI-tutor, AI-coach, AI-mentor, AI-teammate, AI-tool, AI-simulator, and AI-student, each with distinct pedagogical benefits and risks. The aim is to help students learn with and about AI, with practical strategies designed to mitigate risks such as complacency about the AI's output, errors, and biases. These strategies promote active oversight, critical assessment of AI outputs, and complementarity of AI's capabilities with the students' unique insights. By challenging students to remain the "human in the loop," the authors aim to enhance learning outcomes while ensuring that AI serves as a supportive tool rather than a replacement. The proposed framework offers a guide for educators navigating the integration of AI-assisted learning in classrooms
AI as Mentor: Instructions for students
Students should use AI mentors critically and protect their privacy, recognizing that outputs may be unpredictable, context-insensitive, realistic but wrong, and potentially based on shared data. Effective use involves supplying context, questioning and clarifying feedback, reflecting on the interaction, and designing and testing prompts with appropriate personalization and constraints.
- AI as Mentor: Instructions for students: AI mentor outputs can be unpredictable, so students should refresh and retry failed prompts or switch to another language model.Each prompt may produce a different result, and some prompts may not work at a given time.
- AI as Mentor: Instructions for students: Students should not treat the AI as a person because it can fool them and lacks knowledge of their identity and context.The AI may be capable of a lot without understanding the student’s situation.
- AI as Mentor: Instructions for students: Students remain responsible for their work and should critically evaluate AI advice, fact-check final work, and scrutinize sources, facts, and quotes.The AI can provide realistic but wrong or subtly wrong answers.
- AI as Mentor: Instructions for students: Students should share only information they are comfortable sharing because anything provided may be used as AI training data.They are not compelled to disclose personal information.
- AI as Mentor: Instructions for students: Students can improve mentoring interactions by explaining their goals and struggles, questioning assumptions, and requesting clarification until the feedback is understandable.The AI can tailor guidance when given context, and students can ask it to explain feedback differently.
- AI as Mentor: Instructions for students: Students should share complete AI interactions and reflect on what they learned, how well the tool worked, what surprised them, and whether its advice was helpful.Reflection also addresses takeaways about the student’s own work and the AI’s suggestions.
- AI as Mentor: Instructions for students: A custom AI mentor prompt should specify a learning goal, role, desired task, step-by-step interaction, personalization, and constraints such as suggesting rather than revising work.Personalization can identify the student’s learning level and whether the work is a first attempt, while constraints limit the AI’s intervention.
- AI as Mentor: Instructions for students: Prompts should be tested with great, middling, and poor assignments and repeatedly refined from the students’ perspective until the process is helpful.Testing asks whether the AI works, whether it needs more context, and how it could better support students.
AI as Coach: Instructions for Students
Students using AI as a coach should remain in control, provide context, critically evaluate guidance, and reflect on the interaction. Educators can build and test coaches around explicit learning goals, stepwise prompts, examples, personalization, and challenges, while teams can use AI to recognize strengths and question assumptions.
- AI as Coach: Instructions for Students: Prompts may be unpredictable or inconsistent, so students should retry unsuccessful prompts with the same or a different Large Language Model and record what works.The guidance attributes variation to statistical models and recommends sharing successful approaches.
- AI as Coach: Instructions for Students: AI coaching is not human coaching: its responses may feel personal, but it does not know the student’s context and can produce unrelated questions or made-up information.Students should treat every explanation or recommendation critically and redirect or retry the interaction when it becomes unhelpful.
- AI as Coach: Instructions for Students: Students remain in charge by declining irrelevant questions, challenging the AI’s assumptions, and deciding what guidance to use in their own learning.They should ask questions, seek clarification, and tell the AI to move on when necessary.
- AI as Coach: Instructions for Students: Students should share only information they are comfortable disclosing because anything shared may be used as AI training data.They are not compelled to provide personal information.
- AI as Coach: Instructions for Students: Students should share challenges directly, provide context, and ask the AI to question them when they need help articulating or exploring a challenge.Context helps the AI tailor its advice or guidance, although it does not know the student’s situation independently.
- AI as Coach: Instructions for Students: Educators can build an AI coach by defining a reflection or planning goal, specifying the AI’s role and task, providing step-by-step instructions, and adding examples, personalization, and challenges.The coach may address a past event or future project and can prompt students to explain surprises, generate solutions, or consider multiple viewpoints.
- AI as Coach: Instructions for Students: Educators should test prompts across different Large Language Models from students’ perspectives, identify confusion or weak responses, and have students report and reflect on what they learned.Exercises can be completed individually or in teams, followed by class discussion or written reflection on the AI’s usefulness and feedback.
- Team Structure Prompt: AI can support collaborative intelligence by helping teams balance members’ skills, act as a devil’s advocate, question assumptions, and offer alternative viewpoints before decisions.A team-structure prompt can help teams identify strengths and missing expertise before a major project, while devil’s-advocate use can surface potential drawbacks.
AI as Teammate: Instructions for Students
The AI-teammate approach asks students to use AI for advice and alternative ideas while remaining in control, critically evaluating outputs, and protecting personal information. Educators can design prompts with goals, context, stepwise instructions, examples, personalization, and challenges that turn evaluating AI explanations into a learning activity.
- AI as Teammate: Instructions for Students: Students should treat AI as an unpredictable tool, retrying ineffective prompts or switching Large Language Models because outputs vary.The paper also recommends taking notes on which prompts work.
- AI as Teammate: Instructions for Students: Students remain in charge: they need not accept AI advice, should evaluate and challenge it, and can redirect irrelevant questions.AI may not know the student’s context, may hallucinate, and can become stuck in unrelated questioning.
- AI as Teammate: Instructions for Students: Students should actively communicate preferences and feedback because AI may adapt its tone or style, including becoming confrontational when they argue.The guidance specifically recommends stating expectations and giving feedback on advice and output.
- AI as Teammate: Instructions for Students: Students should share only what they are comfortable sharing, avoid personal information, and discuss what they learned from the interaction.The paper asks students to reflect on helpfulness, time savings, decision-making, and takeaways.
- AI as Teammate: Build your own: An AI-teammate prompt should define the team learning goal, role, desired task, step-by-step interaction, examples, project context, and ways to challenge students.Prompt designers should test the prompt across different Large Language Models and anticipate confusion, thoughtful responses, and evaluation of AI advice.
- AI as Teammate: Build your own: Prompt designers can challenge students by requiring questions or solution generation, then testing whether students use their own insights to interrogate AI output.Testing should consider how and when the lesson requires students to evaluate the AI’s advice.
- AI as Teammate: Build your own: Students can teach AI about a familiar topic by identifying what its explanations and examples get right, wrong, or omit, checking their own understanding and fluency.This uses AI’s rapid generation of explanations and examples while making assessment and correction the student’s task.
- AI as Teammate: Build your own: Teaching AI requires students to organize and deeply process knowledge, compare concepts, and explain inaccuracies, gaps, inconsistencies, and successful applications.The passages connect teaching others with uncovering the extent of understanding and note that misunderstanding cannot be addressed without deep topic knowledge.
AI as Student: Instructions for students
Students should treat AI as an unpredictable, nonhuman tool whose explanations and applications require critical evaluation against their existing knowledge and class sources. AI simulation can extend this approach by giving students practice applying concepts in novel scenarios, while requiring oversight because simulations may hallucinate, lose focus, or fit students unevenly.
- AI as Student: Instructions for students: AI responses may vary or fail, so students should retry prompts or use a different Large Language Model when necessary.The paper emphasizes that AI behavior is unpredictable and that some prompts may not work at a given time.
- AI as Student: Instructions for students: Students should remember that AI is not a person, lacks their context, and may invent information when it lacks relevant knowledge.The passage notes that ChatGPT was not connected to the internet and lacked knowledge beyond 2021 in the described context.
- AI as Student: Instructions for students: Students must assess AI explanations and illustrations critically by checking their accuracy, depth, application, and alignment with class sources and prior knowledge.The assignment asks students to evaluate what the AI got right or wrong and use other sources to check its output.
- AI as Student: Instructions for students: Students may end the interaction whenever the AI’s argument is invalid or does not change their assessment.They are not compelled to continue the conversation after giving feedback.
- AI as Student: Instructions for students: AI simulation can create practice scenarios that require students to apply concepts, solve problems, make consequential decisions, and receive performance feedback.The approach targets skills that are difficult to practice by pushing students to transfer knowledge into new situations.
- AI as Student: Instructions for students: A simulator prompt should establish the AI’s role, the student’s role, the target concept, scenario problems, dilemmas, a consequential choice, and a structured ending.The example prompt asks the AI to act as a scenario builder and role player and to set up a consequential choice after 4 interactions.
- AI as Simulator: Sample Output: AI simulations can hallucinate, misjudge students’ learning levels, lose track of the practice goal, and produce scenarios that are ineffective for some students.The risks arise because the AI may lack sufficient concept information, have no sense of student level, or get stuck in a story loop.
- AI as Simulator: Sample Output: Teachers should explain the practice goal beforehand, identify concepts students should recall, and prepare them for variation across scenarios.The guidance recommends reminding students of relevant concepts or discussing them before role-play practice.
AI as Simulator: Instructions for students
AI simulators provide fictional, role-based practice, but students must critically evaluate the AI’s feedback and recognize that it may be unpredictable, context-blind, or inaccurate. Effective use involves goal-directed participation, clarification, detailed responses, reflection, and carefully designed prompts that specify roles, interaction steps, and learning levels.
- Using the simulator: Because AI responses are randomized and prompts may fail, students can retry prompts, switch models, or paste the prompt again to generate a different scenario.The guidance notes that the simulator may not work on the first attempt and that repeating the prompt can produce a different scenario.
- Limitations: AI scenarios are fictional role-plays, not interactions with real people who know the student or the student’s context.The AI may feel like a person, but it is not one and does not know the student or their circumstances.
- Limitations: Students should critically assess every explanation and feedback because the AI can make up information or get the evaluation subtly wrong.Students are advised to check explanations with trusted sources and judge whether feedback adequately addresses the concept or summarizes their performance.
- Using the simulator: Students can deepen practice by pursuing the exercise goal, immersing themselves in the role, giving extensive responses, and asking for clarification when confused.They should apply what they learned to the scene’s challenges rather than treating role-play as an end in itself.
- Building a simulator: To build a simulator, instructors should define the learning goal, assign roles, specify step-by-step interaction rules, and test and refine the prompt for students’ learning level.An example specifies interviewing practice, prevents the AI from playing both roles, requires a consequential choice after 5 interactions, and ends with performance feedback.
Conclusion
The paper concludes that AI tools warrant re-examining traditional assignments and classroom practices, while responsible use requires educators and students to learn about AI’s benefits, risks, and implications together.
- Conclusion: AI tools can facilitate skill development, practice outside the classroom, independent learning, and potentially personalized engagement, prompting educators to reconsider traditional assignments and classroom practices.The authors suggest experimenting with more in-class discussions, question-and-answer sessions, and collaborative exercises such as “think-pair-share.”
- Conclusion: Educators and students should collaboratively develop responsible ways to use AI that enhance educational and life outcomes.The approaches are presented as a starting point for addressing both AI’s advantages and risks.
- Conclusion: Students need firsthand experience with AI tools to understand their implications and prepare themselves for their significant challenges and opportunities.The conclusion frames direct engagement with AI as necessary for learning to use these tools responsibly.
Appendix: Large Language Models and Prompt Compatibility
The appendix presents a prompt-compatibility chart for approaches in the paper across OpenAI ChatGPT 4 and Microsoft’s Bing in Creative Mode. It notes that compatibility may change as the models change.
- Prompt compatibility: The chart covers increasing knowledge through the AI-Tutor approach.Its recorded compatibility entry is “yes no yes no no.”
- Prompt compatibility: The chart covers increasing metacognition through the AI Team Reflection Coach and AI Coach: Team Premortem approaches.Both approaches have the recorded entry “yes no yes no no.”
- Prompt compatibility: The chart covers providing feedback through AI Mentor and building collective intelligence through AI Teammate.AI Mentor is paired with “yes sometimes yes sometimes no,” while AI Teammate is paired with “yes yes yes yes yes.”
- Prompt compatibility: The chart covers increasing fluency through AI Student, paired with “yes sometimes yes sometimes no.”The chart’s note says its prompt compatibility is subject to change as the models change.