Generative
AIFOR BUSINESS TRANSFORMATION
Cornell

Certificate Program

We will open by highlighting recent advances in artificial intelligence (AI) that have enabled machines to develop new forms of cognition — classifying, manipulating, and generating linguistic and visual artifacts — and look at underlying technologies that have the potential to go further and automate tasks commonly associated with the highest levels of human intelligence. We’ll then examine the conceptual underpinnings behind recent advances in generative AI (e.g., image generators like Mid-Journey and large language models like GPT), including learning algorithms, the idea of pretraining, the role of generic and domain-specific data, and the architectural decisions that drive the performance of these modules. This will help build a clear assessment of the current capabilities and uses of AI to make informed guesses on the future potential.

We’ll extrapolate the patterns from past general-purpose technologies to see both the promise and the peril of AI — how AI may create opportunities for new products, services, productivity improvement, and new business models, as well as how it may fail to live up to its promise, leading to unintended consequences and causing harm to society. We’ll apply these patterns to one industry to illustrate that promise and peril. We’ll close by taking stock of our skill sets and how they match — or don’t match — the ones required for this journey.

TOPICS COVERED

• Automating Cognitive Work: From Numbers to Language and Visuals
• Machine Learning: Big Data, Classification, Computer Vision, Language, and Cognition
• Large Models and Pretraining Architectures
• Generative AI: LLMs (GPT-X) and Image Generators
• Project Work: General-Purpose Technologies: From the Steam Engine to AI in Your Industry

We will start this course by providing an overview of the available AI tools and how individuals can use these tools as co-pilots to increase personal productivity — how to pick the right tool, provide context, write better prompts, and validate the output generated. We’ll then discover how AI can be used to improve the effectiveness of organizational processes: automation, cheaper instrumentation and data gathering, tighter monitoring and control, analysis and causal inference, work modularization, and flexization. We’ll apply these templates to identify opportunities for using AI in organizations.

TOPICS COVERED

• Tool Selection. Language Tools: Context, Prompts, and Fine-Tuning
• Image Tools: Prompts, In-Painting, and Other Tricks
• Automation. Information Enablement: Monitoring and Analysis
• Work Redesign: Flexibility and Modularization
• Project Work: Identifying Productivity Enhancement Opportunities

Advances in AI enable new business models. Technology changes the economics and operational trade-offs of a business. Entrepreneurial leaders respond by reinventing their business models — rethinking the scope of their business, the price structures, the timing of decisions, the employment model, and their operations strategy, among other elements. Leveraging these opportunities requires leaders to make strategic choices: what aspects of AI models to build, buy, or borrow; how to use and monetize proprietary training data; and how to time investments.

We will prepare you to answer these questions by building an economic understanding of modern large AI models: the pretraining architecture, the scaling laws, data obsolescence, the computation of cost reductions, the costs and benefits of generic pretraining data, the costs and benefits of proprietary data for fine-tuning, and vertical and horizontal competitive dynamics.

TOPICS COVERED

• The Economics of Artificial Intelligence and Other Digital Technologies
• New Business Models: Scope, Sequence, Players, Platforms, Automation, Intermediation, and
• Disintermediation
• Pretraining Architecture, Scaling Laws
• Technology Competition. Horizontal and Vertical
• Project Work: New Business Model and/or Your AI Strategy

AI models provide few performance guarantees. The latest generation of models are trained on unfiltered internet-scale data; as a result, the models have unpredictable outcomes which can create legal liability. There are also significant open concerns around copyright. Furthermore, the models may be used by unscrupulous actors in unintended ways.

TOPICS COVERED

• Ethical Concerns
• Model Safety in the Wild
• Legal Concerns: Copyright and Liability
• Project Work: Your AI Model Card

We will explore a counterintuitive, risk-limiting approach to developing and launching new AI-based ventures. We’ll use the examples of several high-profile venture failures to understand alternate development strategies and build intuition on how we should develop our new ventures. We’ll then examine a set of techniques to create low-cost, low-risk prototypes for AI models.

TOPICS COVERED

• Failing Fast, Failing Cheap
• Maximum Informational Return Experiments (instead of MVPs)
• Experimentation Templates
• Canonical Study Designs: A/B Testing, Before-and-After Testing, and Difference in Differences
• Seven Rules of Highly Effective Experimenters
• Project Work: Your AI Experiment