AI & Teaching Workshop

HMC Math Department Retreat

2024-08-21

About Me

  • Originally from Connecticut
  • Undergrad at LeTourneau Univ. in Longview, TX Engineering with an Electrical Concentration (2010-2014)
  • Masters and PhD in EE at Caltech (2014-2019)
  • Harvey Mudd (2019-Present)

Workshop Resources

URL: https://tinyurl.com/muddmathai

QR Code

What tools are part of your classroom community?

  1. Chalkboard
  2. Whiteboard
  3. Slides
  4. Tablet
  5. Think-pair-share

What are the values embedded in these tools?

  1. Chalkboard: historical context, culture, limit speed
  2. Whiteboard: cleanliness, limit speed
  3. Slides: accessibility, efficiency
  4. Tablet: accessibility, limit speed
  5. Think-pair-share: community, conversation

What are the values embedded in Generative Artificial Intelligence?

What is artificial intelligence?

Key Terms and Concepts

Generative AI is a program capable of generating content like text, images, video, and computer code. Most genAI systems respond to a prompt entered in natural language by a user.

  • Large language model (LLM): Program that behaves like very sophisticated autocomplete. Examples: GPT-4, Gemini.
  • Generative Pre-trained Transformer (GPT): The specific architecture (also known as foundation model) of LLM first developed by OpenAI.
  • Multimodal: The ability to process and generate different types of media. For example, input text and output an image or video. Examples: Midjourney, DALLE-3, Sora.
  • Supervised machine learning: An machine learning algorithm trained using labeled data.
  • Unsupervised machine learning: A machine learning algorithm trained on unlabeled data.

An Abbreviated History of genAI

A rough timeline:

  • 1956: Field of AI research begins at workshop at Dartmouth College.
  • 1960s-1980s: Machine learning using neural networks is explored.
  • Late 2000s: Development of deep learning. Deep refers to multiple hidden layers in the network.
  • 2014: Generative Adversarial Networks (GANs) are developed.
  • 2018: OpenAI invents the Generative Pre-trained Transformer (GPT) architecture.
  • 2022: ChatGPT released to the public in November based on GPT-3.5.

Discriminative AI vs. Generative AI

Discriminative models focus on classification and prediction based on given data.

  • Facial recognition: Identify persons in photo library.
  • Speech recognition: Siri or Alexa.
  • Medical diagnosis: Classifying your MRI or X-ray images.
  • Autonomous navigation: Tesla full self driving.
  • Recommendation systems: Amazon predicting what you might want to buy.

Generative models are capable of creating outputs drawn from the distribution of input data they were trained on.

  • Text generation: LLMs which generate text from a prompt.
  • Image generation: Midjourney or DALLE to generate new images.
  • Drug discovery: Alpha Fold for predicting protein structure.

The medium is the message

A technology becomes a medium as it employs a particular symbolic code, as it finds its place in a particular social setting, as it insinuates itself into economic and political contexts. A technology, in other words, is merely a machine. A medium is the social and intellectual environment a machine creates.

Understanding Media (1994)

Marshall McLuhan

Neil Postman on the message in the medium

A news show, to put it plainly, is a format for entertainment, not for education, reflection or catharsis. And we must not judge too harshly those who have framed it in this way. They are not assembling the news to be read, or broadcasting it to be heard. They are televising the news to be seen. They must follow where their medium leads. There is no conspiracy here, no lack of intelligence, only a straightforward recognition that “good television” has little to do with what is “good” about exposition or other forms of verbal communication but everything to do with what the pictorial images look like.

Amusing Ourselves to Death (2005)

Neil Postman

Whose interest?

[Y]ears from now..it will be noticed that the massive collection and speed-of-light retrieval of data have been of great value to large-scale organizations but have solved very little of importance to most people and have created at least as many problems for them as they may have solved.

Amusing Ourselves to Death (2005)

Neil Postman

What is technology?

Recently, people have become particularly interested in technology’s social impact. I myself am overawed by the way in which technology has acted to reorder and restructure social relations, not only affecting the relations between social groups, but also the relations between nations and individuals, and between all of us and our environment. To a new generation, many of these changed relationships appear so normal, so inevitable, that they are taken as given and are not questioned.

The Real World of Technology (1989)

Ursula Franklin’s 1989 CBC Lectures

Workshop Goals

Goals for today

By the conclusion of this workshop you will have…

  • Reflected on your values as an educator and how they are expressed in your classes.
  • Developed a curiosity to experiment with generative artificial intelligence.
  • Outlined text for your syllabus to articulate your pedagogical values and examples of how those values work themselves out in your courses.
  • Developed an AI policy for your courses.

HMC Math Dept. Goals

Practice

  1. Affection
  2. Application
  3. Inquiry
  4. Communication
  5. Technology
  6. Society
  7. Teamwork
  8. Diversity

Content

  1. Breadth
  2. Perspective
  3. Duality
  4. Proofs
  5. Tools
  6. Depth
  7. Projects
  8. Careers

Agenda for Today

Activities

  1. GenAI Demo
  2. GenAI Value Mapping
  3. Syllabus Revision

GenAI Demo

Lookee what he can do!

Yukon Cornelius and the Abominable Snow Monster

ChatGPT

ChatGPT the Homework Helper

Figure 2: Math 131 Homework

ChatGPT Takes Math 131

Figure 3: ChatGPT attempts Math 131

ChatGPT Takes Math 131

Figure 4: ChatGPT attempts Math 131

ChatGPT the Concept Explainer

Figure 5: Taylor series explanation

ChatGPT the Proof Writer

Figure 6: Math 40 proof

ChatGPT Math 40

Figure 7: ChatGPT attempts Math 40 proof

ChatGPT Math 40

Figure 8: ChatGPT attempts Math 40 proof

ChatGPT Math 40

Figure 9: ChatGPT attempts Math 40 proof

High-level Observations

  • LLM-generated text can reproduce the type of structure and format we would like to see.
  • Clean syntax is no longer a worthy proxy for clear thought.
  • The types of errors in LLM-generated text are tricky to detect.
  • The LLMs “solves” the blank page problem.
  • Generating a solution is now effectively free.
  • The floor for communicating mathematical thinking has been raised.

Activity #1: GenAI Value Mapping

Exploring alignment

Course intersections

Experimenting with AI in your courses

Some Principles

  1. Build agency
  2. Encourage reflection
  3. Explain what learning looks like
  4. Make values and learning goals explicit
  5. Find ways to show your students you trust them
  6. Build socio-technical thinking

Prototypes

For Students

  1. Homework Helper
  2. Interactive Encyclopedia
  3. Ideation Partner

For Instructors

  1. Pedagogy Coach
  2. Active Learning Activity Generation
  3. Feedback Summarization

Activity #2: Syllabus Revision

Elements of a Syllabus Policy

  1. A brief explanation of what generative AI is (here’s a nice interactive demo of transformers from the Financial Times)
  2. Your policy on generative AI use in the classroom
    • Avoid
    • Embrace
    • Critical Exploration
  3. An explanation of the reasoning behind your policy
  4. An articulation of your expectations around use if you are allowing genAI to be used (e.g., norms around disclosure, reporting requirements, etc.)

Resources

https://bit.ly/AI-Syllabi

Closing Thoughts

Some things you should discuss with students

  • When to use an LLM vs. talk to your classmate or instructor
  • The environmental considerations: the computational horsepower to run these things takes a lot of electricity and water.
  • Data privacy concerns: The ethical issues surrounding training data
  • Embedded biases

Contact Me

If you have questions or comments, please connect!

Substack: joshbrake.substack.com

Email: jbrake@hmc.edu

Twitter: @JoshuaBrake

Bluesky: @joshbrake.com