Teaching resources

Most students have heard about AI, but do they know how it works? 

Do they know its strengths and weaknesses? Do they know how to assess the risk, and how to design countermeasures to mitigate risk?  

The AI Conundrum is required reading in a range of courses from Business Analytics to Journalism, and from Intro to Computer Science to Public Health Communications. Our goal is to share teaching materials to increase understanding and ethical use of AI.  

All materials are available under the MIT Creative Commons Attribution-NonCommercial-ShareAlike (CC BY-NC-SA) license. This license is commonly applied to materials available on MIT OpenCourseWare (OCW). 

Track 1: 30 Minutes to 1 Hour Introduction to AI

Purpose: A brief introduction to AI to engage students in recognizing AI thinks differently than humans, and the implications of these differences, especially in precision and the risk when AI operates in an open environment vs. in a closed environment where a human is at the helm or in the loop. 

10min: Introduction to How AI Thinks Differently Than Humans: (Video)

10min: Self Assessment - How Much Do You Know About How AI Really Works?

• 8min to take self assessment
• 2min debrief on the scores (87% of business people score 36 or lower and have substantial room to grow in terms of understanding how AI works. Generally only computer scientists score 95 or higher).  

10min: Discussion of Implications of AI Risk Framework, and different use cases where AI can perform well versus where it may be more risky.

• Using the framework introduced in the video (precision, rationale, and open vs. closed systems), brainstorm the risk levels of different AI applications. 
• Consider discussion of facial recognition. Facial recognition has led to false arrests (Clearview AI). At the same time, facial recognition has been used to find missing children that have been sex trafficked (Spotlight). It has also been used for personal photo libraries to sort photos (Mylio). A key factor to consider in risk is how the technology is applied. At the same time, discussion may turn to the challenges in controlling how technology is applied, and how risk may increase as AI becomes more capable and people become more adroit at using it.

To improve self assessment score, read Part 1 of The AI Conundrum.

Additional Resource: AI Bot (vector database) of book and training materials https://www.speakerrex.com/bot.html

Track 2: 

Purpose: An introduction to AI to engage students in recognizing AI thinks differently than humans, as well as hands-on with GenAI for image and for text.  This module provides a combination of instructional content and hands-on short labs to bring the concepts to life. The module provides practical examples of using AI in advertising to make predictions and personalize content, as well as Retrieval Augmented Generation to generate AI Personas. Track 2 module briefly addresses the strengths and weaknesses of AI. It addresses the implications of differences in how AI thinks versus how humans think, especially in precision and the risk when AI operates in an open environment vs. in a closed environment where a human is at the helm or in the loop. It offers a few hands-on exercises to bring the concepts to life. 

30 minutes - Part 1
20min: Introduction to How AI Thinks Differently Than Humans + Advertising Use Case: (Video)

10min: Self Assessment (How Much Do You Know About How AI Really Works?)

• 8min to take self assessment
• 2min debrief on the scores (87% of business people score 36 or lower and have substantial room to grow in terms of understanding how AI works. Generally only computer scientists score 95 or higher).  

30 minutes - Part 2

9min: Fundamentals of How AI Works (video)

9min: Exercise: Text to Image Generation (lab)

• Text to Image
• In-painting (fill)
• Out-painting (expansion) and role of context

12min Debrief (video-8:55+)

1 hour - Part 3

18min Gradient Descent and Bias In AI (video)

22min: How Large Language Models (LLMs) Work (video)

• 8min: Prompt Library Exercise: Explore Anthropic Prompt Library and test out at least two different prompt strategies/use cases, such as "Corporate Clairvoyant." 
• Extra time? Try the console/API to send prompts and receive response. This is a key mechanism for building AI workflows that go beyond the basic chat interface. 

5min: Retrieval Augmented Generation (RAG), and Claritas Personas Case Study (video)

• 5min optional exercise: Engaging with personas with different reference data and comparing results (Jenna, Striving Selfies vs. Alex, Young Digerati). Consider asking questions to each such as whether you prefer to pay for an App and have no ads, or have a free App with ads, or the type of car they plan to buy next, or favorite chain restaurants. 

10min: Discussion of Implications of AI Risk Framework, and different use cases where AI can perform well versus where it may be more risky.

• Using the framework introduced in the video (precision, rationale, and open vs. closed systems), brainstorm the risk levels of different AI applications. 
• Consider discussion of facial recognition. Facial recognition has led to false arrests (Clearview AI). At the same time, facial recognition has been used to find missing children that have been sex trafficked (Spotlight). It has also been used for personal photo libraries to sort photos (Mylio). A key factor to consider in risk is how the technology is applied. At the same time, discussion may turn to the challenges in controlling how technology is applied, and how risk may increase as AI becomes more capable and people become more adroit at using it.

To improve scores on self-assessment, read Part 1 of The AI Conundrum. Ask the AI Book Bot (vector database) to teach you the concepts, or ask clarifying questions. https://www.speakerrex.com/bot.html

Track 3: 

Purpose: Take students through a progressive set of lectures and labs that provide an introduction to AI to engage students in recognizing AI thinks differently than humans, provide the foundations of how AI works, and how it can be applied in GenAI applications, prompting, computer vision, and mixed AI workflows and Agentic Systems. Track 3 is a survey of these topics and includes over a dozen hands-on exercises to provide hands-on exposure to the various AI systems.

Part 1: Approximately 1 Hour, 35 minutes of video + 2 labs: 

Introduction to How AI Thinks Differently Than Humans + Advertising Use Case (video) 

Lab 1: How much do you know about how AI really Works? (self assessment) 

Lab 2: Natural Language Processing (NLP) and Large Language Models (LLMs) to summarize the training video and ask questions of the content (all labs are at https://speakerrex.com/2024)

• First 30min is same as in Track 2, with 20min of video + 8min lab with 2min debrief on the scores (87% of business people score 36 or lower and have substantial room to grow in terms of understanding how AI works. Generally only computer scientists score in the 95 or higher category). 
• Second section of video includes approximately 10 minutes addressing some of the risks of AI. 
• Lab 2 shifts to a more utilitarian focus, and shows how to use AI to take the transcript from the video, run it through a large language model (LLMs) such as Claude, Gemini, Meta or Open AI and convert it to a summary where the student can ask questions and extract answers from the transcript via the chat interface. This lab highlights a strength of AI.

Part 2: Approximately 1 Hour, 50 minutes of video + 1 lab:

The Fundamentals of How AI Works (video)

• How AI Thinks: Universal Approximation and X, Y pattern fitting
• Exercise: Text-to-image generation (lab instructions)
• How AI Learns: Gradient Descent (and why it can lead to bias in AI due to correlation - note: bias can emerge due to bias in the underlying dataset, as in the case of the wolves and huskies example, or due to spurious correlations, as in the pixel for the baby in the bassinet. In general, AI tends to intensify bias due to the probabilistic nature of AI output.) 

Part 3: Approximately 1 Hour, 42 minutes of instructional video + 3 labs + 5 minutes of Q&A:

How Large Language Models (LLMs) Work (video)

• How LLMs produce output
• How to prompt 
• Exercise: Personas
• Exercise: Building in reasoning through prompts
• Exercise: Jailbreaking LLMs

Part 4: Vector Database, Computer Vision and Multimodal AI 

(Edited video and description coming soon)

• Unedited Video

Part 5: AI Workflow and AI Agents 

• Unedited Video

If you have a specific way you are teaching The AI Conundrum, contact us and share your materials with other instructors. 

What students are saying in their written reviews:

"I give this read five stars because, ultimately, it is a great blend of entertaining and witty while also being extremely informative. I appreciate that the book tackles the incredibly complex subject of AI in a digestible way that anyone can gain insight from. I would recommend this to anyone interested in the topic, regardless of their level of familiarity with artificial intelligence, machine learning, or any related subject."

"For someone pretty uninformed about AI (only really knowing about some of the hype) this book provided amazing background, context, examples, and explanations of how AI has limitations and why those limitations happen. However, at times it dives deep into an explanation revolving around math that is pretty hard to grasp unless you kind of already understand those topics. I understand that it might need to be done this way and that it's more on me, but I wish it was a little more simplified."

(AUTHOR NOTE: We are revised the math section to make the nine pages that explain how AI works more accessible to those that aren't confident in math). 

"Part 1 was particularly great because I think it was overly simple and used different examples that helped me understand a complex topic."

"I enjoyed the dive into the two heads behind the book and their individual backgrounds. I also loved the discussion of AI's relation to optical illusions, such as the checkers board light illusion. Finally, I loved the discussion about the challenge AI has recognizing diversity, as well as its inherent bias; I thought your take on the topic was very nuanced and gave an objective description of it, rather than making it in any way politicized (as many outlets who discuss that subject tend to do)."

Extra Credit assignment: Read Part 1, and apply the AI Risk Framework to a use case example selected by the student. Explain the risk and countermeasures the student would apply and why.  

In development

Assignment: Read Part 1 and Chapters 10 & 11 in Part 2. 

Discuss why AI polarizes, and how the business model for media contributes to polarization. Discuss how misinformation spreads, and potential strategies to increase media literarcy. 

Most text books explain the code, but leave out the deeper explanation of AI's strengths and weaknesses. Part 1 of The AI Conundrum does an excellent job in augmenting the technical text books on AI by offering a more complete understanding of how AI works. Part 3 does an excellent job at explaining AI Bias and offering specific approaches to mitigating bias. 

Contact us and we will happily provide materials and collaborate.

A one day course, built around The AI Conundrum, assigns the book as pre-read, and then gets executives hands-on with AI, paralleling Part 1, in the morning. Before the lunch break, executives analyze an AI business case using the AI Risk Framework, as well as the cost/benefit analysis of AI vs. the Next Best Alternative. In the afternoon, sessions focus on building an AI Governance framework, and evaluating specific use cases to identify weak points and to consider countermeasures to mitigate risk.