Skip to content

Generative AI and Multi-Agent Networks for Creating Digital Twins (Kenyon College's Integrated Program for Humane Studies Program Fall 2024)

License

Notifications You must be signed in to change notification settings

jon-chun/GenAI-Multi-Agent-Networks-and-Digital-Twins

Repository files navigation

Motivations

How is AI changing humanity? This course explores the realities of labor, knowledge work, and automation. Can the current generation of rapidly improving autonomous AI agents collaboratively (or independently) craft novels, social media political campaigns, commercial brand marketing strategies, financial analyst reports, software apps, music videos, or write original scientific research papers? For those with basic Python skills and advanced curiosity, get hands-on experience building your own personal digital twin and a peek into the future.

Requirements, Philosophy, and Project-Based Learning

Although helpful, no maths beyond HS and introductory Python are essential. Unlike traditional CS/AI courses, we do not build from the bottom-up starting with comprehensive theory. Instead, we follow a top-down approach starting with real-world problems at a high-level of abstraction. We present an intellectual framework for each student to understand the underlying theory according to their individual preparation and perspectives.

Rapid AI progress paradoxically makes AI research both more complex and accessible. Traditional narrow AI research is rapidly advancing, papers have been exponentially growing in number, and expertise is becoming more specialized and/or deep. Conversely, human-centered AI applications have become more performant, easy to develop, and focused on domain expertise rather than specialized technical knowledge. Our goal is to train emerging experts from all fields to systematically, creatively and iteratively think in precise, comprehensive, and quantifiable terms that can translate ideas into well-designed and high-impact AI applications.

Learning and projects are customized to each student’s individual background, domain expertise, as well as current and future interests. Students from diverse fields fill in knowledge gaps and foster cross-pollination of ideas and insights. By the end of the course, every student should have an online digital portfolio of 4 industry standard AI projects and one original research project for future work or grad school reference.

More details on our human-centered AI curriculum, research and previous student research projects here and at digital.kenyon.edu/dh.

Fee

  • No textbooks, instead, ~$100-$150 in cloud subscriptions
    • $20/mo OpenAI ChatGPT
    • $10/mo Google Colab
    • API usage fees (IPHS can provide assistance if necessary)

Syllabus

IPHS 391 Syllabus in PDF

IPHS 391: Frontiers in Generative AI

This is an upper-division course offers an in-depth exploration of advanced AI concepts, focusing on interdisciplinary applications of large language models, AI information systems, and autonomous agents. Students will engage with a progressive curriculum, starting with a review of Python and a series of four hands-on projects, equipping them with the skills and knowledge necessary to innovate in the rapidly evolving field of artificial intelligence.

Course Details

  • Instructor:** Jon Chun
  • Institution:** Kenyon College
  • Term:** Fall 2024
  • Schedule:** Wednesday Evenings 7pm-10pm
  • Location:** Timberlake #5 (Evans Conference Rm)
  • Credit:** 0.5/4
  • Section:** 00
  • CRN:** 80044
  • Class Size:** 15
  • Prerequisites:** Introductory Python programming experience

Course Outline

Week 1-3: OpenAI API Function Calling App

Learning Goals

  • Understand the basics of AI and its applications
  • Learn to use Python and Prompt Engineering with the OpenAI API
  • Develop a basic function-calling application

Description

Introduction to AI concepts and OpenAI API. Students will learn about GPT models and create a simple chatbot with distinct persona and memory capabilities.

Detailed Content

NOTE: Content will be added each week for Fall 2024

Reflection Questions

  1. How do API function calls enhance the capabilities of language models?
  2. What are the ethical considerations when designing AI chatbots with personas?
  3. How might this technology impact various industries?
  4. What are the potential risks of relying too heavily on AI-generated content?
  5. How does the concept of "prompt engineering" relate to working with language models?
  6. In what ways could function calling be used to make AI systems more controllable?
  7. How might the widespread use of AI chatbots affect human communication skills?
  8. What are the implications of AI systems having short-term and long-term memory?
  9. How could this technology be misused, and what safeguards might be necessary?
  10. How does the development of AI chatbots relate to the Turing test?

Week 4-6: Embeddings and Explainability

Learning Goals

  • Understand embeddings and their role in language models
  • Learn to apply embeddings in AI applications
  • Explore methods for AI explainability

Description

Deep dive into embeddings, their applications, and techniques for making AI models more interpretable and explainable.

Detailed Content

Reflection Questions

  1. How do embeddings capture semantic meaning in language?
  2. What are the challenges in making complex AI models explainable?
  3. How might improved AI explainability impact fields like healthcare or criminal justice?
  4. What are the ethical implications of using AI systems that can't be fully explained?
  5. How do embeddings differ across languages, and what are the implications for multilingual AI?
  6. In what ways could explainable AI help build trust between humans and AI systems?
  7. How might embeddings be used in fields outside of natural language processing?
  8. What are the limitations of current explainability techniques in AI?
  9. How could explainable AI impact the development of AI regulations and policies?
  10. What role does bias play in embeddings, and how can it be mitigated?

Week 7-9: RAG Application using Langchain

Learning Goals

  • Understand Retrieval-Augmented Generation (RAG)
  • Learn to use Langchain for RAG applications
  • Implement a RAG system to enhance language model capabilities

Description

Introduction to RAG and its implementation using Langchain. Students will create applications that combine the power of language models with external knowledge sources.

Detailed Content

Reflection Questions

  1. How does RAG improve the accuracy and reliability of language models?
  2. What are the potential applications of RAG in research and industry?
  3. How might RAG systems impact the spread of misinformation?
  4. What are the privacy concerns associated with RAG systems that access external data?
  5. How does RAG compare to fine-tuning as a method for improving language model performance?
  6. In what ways could RAG be used to make AI systems more up-to-date and relevant?
  7. What are the computational challenges of implementing RAG at scale?
  8. How might RAG systems change the way we interact with and consume information?
  9. What ethical considerations should be taken into account when designing RAG systems?
  10. How could RAG technology be used to create more personalized AI experiences?

Week 10-12: Autonomous Multi-Agent Simulations and Benchmarking

Learning Goals

  • Understand the principles of multi-agent systems
  • Learn to design and implement autonomous agent simulations
  • Develop skills in benchmarking and analyzing multi-agent systems

Description

Exploration of autonomous multi-agent systems, their simulation, and benchmarking. Students will create complex simulations to model real-world scenarios.

Detailed Content

Reflection Questions

  1. How do multi-agent systems model complex social interactions?
  2. What are the challenges in designing truly autonomous agents?
  3. How might multi-agent simulations be used to inform policy decisions?
  4. What are the ethical implications of using AI agents to simulate human behavior?
  5. How could multi-agent systems be used to solve real-world coordination problems?
  6. What are the limitations of current multi-agent simulation technologies?
  7. How might the principles of multi-agent systems apply to fields like economics or ecology?
  8. What role could multi-agent simulations play in future AI safety research?
  9. How do the concepts of cooperation and competition manifest in multi-agent systems?
  10. What are the potential risks of relying on multi-agent simulations for decision-making?

Week 13-15: Final Project Development and Presentation

Learning Goals

  • Apply learned concepts to a comprehensive AI project
  • Develop project management and presentation skills
  • Synthesize technical knowledge with real-world applications

Description

Students will finalize and present their semester-long projects, demonstrating their understanding and application of the course material.

Detailed Content

Reflection Questions

  1. How can AI technologies be combined to solve complex, real-world problems?
  2. What are the ethical considerations in deploying your AI solution?
  3. How might your project evolve with future advancements in AI?
  4. What were the most significant challenges you faced in your project, and how did you overcome them?
  5. How does your project contribute to the broader field of AI or its applications?
  6. What potential societal impacts could your project have if implemented at scale?
  7. How did the interdisciplinary nature of the course influence your project design?
  8. What aspects of AI development did you find most surprising or counterintuitive?
  9. How has this project changed your perspective on the future of AI?
  10. What further research or development would you propose as a next step for your project?

Grading and Projects

  • Class Participation: 20%
  • Weekly Quizzes: 30%
  • 4 Mini-Projects: 30%
  • 1 Final Main Project: 20%

Mini-Projects

  1. OpenAI API Function Calling Chat App
  2. Embeddings and Explainability with Huggingface
  3. RAG Application using Langchain
  4. Autonomous Multi-Agent Simulations

Final Project Submission Requirements

An original interdisciplinary research project applying at least one key technology covered in the 4 mini-projects.

Final projects are due today Wednesday, December 18, 2024 by 6:30pm per the official Kenyon 2024 Exam Week Schedule. You may request an extension until Friday at noon but no later.

  1. Please email me the following 2 parts for your Project for grading:
  • Your carefully proof-read poster in *.ppt (MS PowerPoint format NOT PDF)
  • Your code (to verify originality/plagiarism check) (e.g. a colab notebook or github repo)
  • Any data required to successfully run and test your programs (e.g. *.csv structured tables for SML or *.pdf unstructured documents for RAG)
  1. If your code is based upon one of the many agentic tutorials/examples out there:
  • Cite this in your references
  • Explain this in your poster and explain how you modified, extended, or otherwise customized it (e.g. unique dataset, prompts, agents, tools, etc)
  1. If you used AI to help write your poster, please note that in the credits for transparency (and do not have AI write the poster for you)

  2. Please do a final check on Moodle.kenyon.edu to ensure all your grades, quizzes, presentations, etc have been accurately recorded.

For detailed project descriptions and grading rubrics, please refer to the individual module files.

IPHS200: Programming Humanity Exercises

Diachronic Sentiment Analysis

Topic Modeling

NOTE: Be sure to: (a) create a Kaggle account with your non-Kenyon gmail.com account if you don't have one already, (b) [Copy and Edit] to make you own personal copy.

About

Generative AI and Multi-Agent Networks for Creating Digital Twins (Kenyon College's Integrated Program for Humane Studies Program Fall 2024)

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published