Syllabus walkthrough, evaluation breakdown, team formation. Dr. Addas research overview.
Defining AI, ML, and the two-track structure of this course. Where each type is used.
Impact on industries, organizations, and individuals. Real org success stories and usage data.
Hands-on: open Colab, run first Python notebook, explore Gemini AI features.
AI is reshaping every industry. The question is no longer whether AI will affect your organization — it's whether you'll be the leader who shapes that adoption or the one who reacts to it. MBUS 854 is designed for general managers, not data scientists. You will learn enough to lead, evaluate, and govern AI initiatives without writing production code.
Dr. Addas's research focuses on the intersection of AI, technology adoption, and organizational outcomes. The course draws from real cases and current research — expect frameworks you can use in the boardroom, not just the classroom.
Predictive AI learns patterns from labeled historical data to predict an outcome. Output is a number or category (e.g., will this customer default? Is this image a cat?). Accuracy is measurable against known ground truth.
Generative AI learns to produce new content — text, images, code, audio — by modeling probability distributions over data. Output is open-ended. Evaluation is harder (what is a "correct" paragraph?).
| Dimension | Predictive AI | Generative AI |
|---|---|---|
| Output type | Number, label, probability | Text, image, code, audio |
| Training data | Labeled (X → Y pairs) | Unlabeled at scale |
| Evaluation | Accuracy, AUC, F1 | Human judgment, benchmarks |
| Business use | Credit scoring, churn, fraud | Drafting, search, code assist |
| Key risk | Bias in historical data | Hallucination, misuse, IP |
A key strategic question in AI adoption: does this AI replace human judgment (automation) or enhance it (augmentation)? The right choice depends on task complexity, accountability needs, and error tolerance.
Cross-Industry Standard Process for Data Mining. The industry standard workflow for ML projects. You'll use this structure in the individual assignment.
| Phase | What Happens | Key Leader Questions |
|---|---|---|
| 1. Business Understanding | Define the problem, success criteria | What decision does this model support? |
| 2. Data Understanding | Explore, describe, quality check | Do we have enough? Is it biased? |
| 3. Data Preparation | Clean, transform, engineer features | What's being excluded and why? |
| 4. Modeling | Select and train algorithms | What trade-offs does this model make? |
| 5. Evaluation | Test performance, check for bias | What does a false positive cost us? |
| 6. Deployment | Integrate into workflow, monitor | Who is accountable when it's wrong? |
Stanford AI Index 2025 and industry surveys reveal the most common enterprise use cases:
| Use Case | Share of Users | Example |
|---|---|---|
| Content creation | 23% | Drafts, marketing copy, reports |
| Technical assistance | 21% | Code generation, debugging, SQL |
| Research & summarization | 18% | Literature review, competitive intel |
| Customer service | 15% | Chatbots, ticket triage, FAQ |
| Data analysis | 12% | Spreadsheet generation, dashboards |
Used for the Predictive AI track and individual assignment. A cloud-based Jupyter notebook environment — no local Python install needed. Gemini AI features are embedded: you can ask Gemini to explain code, generate functions, or debug errors directly in the notebook.
What you need to know: Cells run sequentially. Variables persist across cells in the same session. If a cell fails, fix it and re-run — don't just re-run all cells from the top blindly.
Used for the Generative AI track (Sessions 3–4) and the group project. PartyRock is AWS's no-code GenAI builder — you can prototype an AI app without writing code. Bedrock is the enterprise platform underneath it, giving access to models from Anthropic (Claude), Amazon (Titan), Meta (Llama), and others.
The group project will require building a working GenAI prototype on Bedrock. Start getting familiar with PartyRock early — it's the fastest path to understanding what Bedrock can do.
While the course is tool-agnostic, being fluent in at least one major LLM interface is expected. For session prep, research, and report drafting, Claude (Anthropic) and ChatGPT (OpenAI) are the most capable general-purpose options. Gemini Ultra is the Google equivalent and is directly integrated into Colab.
Watch the assigned Google Colab introduction video before class. You should arrive knowing how to: