AI in education
Contents
AI in Education
Survey/study of
- 2023-08: Perception, performance, and detectability of conversational artificial intelligence across 32 university courses
- 2023-10: Employees secretly using AI at work.
- 2023-10: Survey shows students using AI more than professors.
- 2023-11: ChatGPT has entered the classroom: how LLMs could transform education
AI improves learning/education
- Mollick, Ethan R. and Mollick, Lilach and Bach, Natalie and Ciccarelli, LJ and Przystanski, Ben and Ravipinto, Daniel, AI Agents and Education: Simulated Practice at Scale (June 17, 2024). The Wharton School Research Paper. doi: 10.2139/ssrn.4871171
- Can enable personalized education.
- Generative AI for Programming Education: Benchmarking ChatGPT, GPT-4, and Human Tutors
- GPT4 can out-perform human tutors.
- Keppler, Samantha and Sinchaisri, Wichinpong and Snyder, Clare, Backwards Planning with Generative AI: Case Study Evidence from US K12 Teachers (August 13, 2024). doi: 10.2139/ssrn.4924786
- Teachers benefit from using AI as a co-pilot to aid in tasks (planning, how to teach topic, explore ideas).
- There is smaller utility in using AI purely as a text-generator (to make quizzes, workbooks, etc.).
AI harms learning
- A real-world test of artificial intelligence infiltration of a university examinations system: A “Turing Test” case study ** Current grading systems cannot detect AI.
- Bastani, Hamsa and Bastani, Osbert and Sungu, Alp and Ge, Haosen and Kabakcı, Özge and Mariman, Rei, Generative AI Can Harm Learning (July 15, 2024). The Wharton School Research Paper.doi: 10.2139/ssrn.4895486
- Access to ChatGPT harmed math education outcomes.
Software/systems
- GPTutor (code)
- EduChat: A Large-Scale Language Model-based Chatbot System for Intelligent Education
- Eureka Labs (founded by Andrej Karpathy) aims to create AI-driven courses (first course is Intro to LLMs)
Individual tools
- Chatbot (OpenAI ChatGPT, Anthropic Claude, Google Gemini)
- NotebookLM: Enables one to "chat with documents".
AI for grading
- Can Large Language Models Make the Grade? An Empirical Study Evaluating LLMs Ability To Mark Short Answer Questions in K-12 Education (preprint)
Detection
- Do teachers spot AI? Evaluating the detectability of AI-generated texts among student essays
- GenAI can simulate student writing in a way that teachers cannot detect.
- AI essays are assessed more positively than student-written.
- Teachers are overconfident in their source identification.
- Both novice and experienced teachers could not identify texts generated by ChatGPT vs. students
AI/human
AI out-performs humans
Tests
- 2023-07: SciBench: Evaluating College-Level Scientific Problem-Solving Abilities of Large Language Models
- 2024-06: A real-world test of artificial intelligence infiltration of a university examinations system: A “Turing Test” case study
- AI scores higher than median students.
Creativity
- 2023-09: Best humans still outperform artificial intelligence in a creative divergent thinking task
- Best humans out-perform AI at creativity. (By implication, median humans may not.)
- 2024-02: The current state of artificial intelligence generative language models is more creative than humans on divergent thinking tasks
- 2024-02: Felin, Teppo and Holweg, Matthias, Theory Is All You Need: AI, Human Cognition, and Causal Reasoning (February 24, 2024). doi: 10.2139/ssrn.4737265
- Argues that human "theory-based" creativity is better than AI "data-based".
- 2024-07: Pron vs Prompt: Can Large Language Models already Challenge a World-Class Fiction Author at Creative Text Writing?
- Top human (professional author) out-performs GPT4.
- 2024-09: Can LLMs Generate Novel Research Ideas? A Large-Scale Human Study with 100+ NLP Researchers
- LLMs can be creative
Professions
- 2024-03: Influence of a Large Language Model on Diagnostic Reasoning: A Randomized Clinical Vignette Study
- GPT4 improves medical practitioner work; surprisingly, GPT4 alone scored better than a human with GPT4 as aid (on selected tasks).
- Humanity's Last Exam
- Effort to build a dataset of challenging (but resolvable) questions in specific domain areas, to act as a benchmark to test whether AIs are improving in these challenging topics.
AI improves human work
- TBD
Uptake
TBD
