Module 1

What Is Generative AI?

Generative artificial intelligence (generative AI) refers to computer systems designed to create new content—such as written text, tables, summaries, and images—rather than merely analyzing existing data. At the core of conversational AI tools are Large Language Models (LLMs). An LLM is a machine learning model trained on vast collections of text to recognize patterns in human language.

Instead of "thinking" or "knowing" facts the way a person does, an LLM operates on statistical prediction. When you submit a request (known as a prompt), the model evaluates your input and calculates which word or phrase is most likely to come next, one piece at a time. A helpful analogy is an advanced autocomplete tool: it builds sentences based on probabilistic patterns learned during training.

Practical Purpose in Everyday Office Work

For knowledge workers, generative AI functions as an on-demand drafting assistant. Common workplace capabilities include:

• Drafting correspondence: Producing initial versions of routine emails, project memos, and client updates.
• Condensing long documents: Summarizing lengthy meeting transcripts, policy briefs, or industry reports into actionable bullet points.
• Restructuring information: Converting raw meeting notes into structured agendas, project timelines, or tabular formats.

Guided Scenario: Transforming Meeting Notes

Imagine you return from a 45-minute project sync with messy, unstructured notes:

"Discussed timeline. Sarah says design needs 3 more days. Marcus will check with vendor on Friday. Need final signoff from leadership before budget approval next Tuesday."

When you ask a conversational AI assistant, "Organize these notes into an action table with Owner, Task, and Due Date," the model recognizes the contextual relationships among names, actions, and dates. It rapidly formats the text into a clear three-column table, saving you 10 to 15 minutes of manual reformatting.

Common Misconception

Misconception: The AI understands business logic and independently verifies truth.
Generative AI does not understand meaning, intent, or workplace consequences. It generates language that sounds plausible based on learned structures. You remain the subject matter expert responsible for confirming whether the output accurately reflects reality.

In the next lesson, we will explore the organizational boundaries required to protect sensitive workplace data while using these tools.

Workplace Governance and Data Privacy

Workplace governance refers to the policies, security standards, and compliance rules that dictate how employees handle organizational assets. When using generative AI, data privacy is the primary concern. Free, public browser-based AI tools frequently retain user prompts to train future iterations of their models. If you enter proprietary information into an unmanaged AI interface, that data may leave your company's secure boundary.

To safeguard your organization, you must distinguish between public information and restricted data:

• Personally Identifiable Information (PII): Data that identifies an individual, including full names, home addresses, phone numbers, employee identification numbers, and social security numbers.
• Confidential Business Data: Non-public financial statements, unpublished product roadmaps, legal contracts, pricing negotiations, and customer lists.
• Intellectual Property (IP): Proprietary code, patented processes, internal methodology documents, and creative designs owned by your company.

Guided Walkthrough: Data Sanitization

Before submitting any draft or scenario to an AI tool, practice data sanitization (also called redaction or anonymization). Replace real corporate identifiers with generic placeholders.

Unsafe Prompt (Direct Leakage):

"Summarize this severance dispute for John Doe, Employee ID 84920 at Acme Logistics, who claims $45,000 in unpaid bonuses under our confidential Q3 sales commission plan."

Compliant Prompt (Sanitized):

"Summarize the following hypothetical employment dispute involving a standard sales commission structure, replacing individual and corporate names with generic roles: [Sanitized text with names and specific dollar figures removed]."

By replacing John Doe and Acme Logistics with neutral placeholders, you obtain the analytical assistance you need without violating confidentiality agreements or privacy statutes.

Common Mistake to Avoid

Mistake: Assuming private chat windows or deleted chat histories remove corporate liability.
Selecting "delete chat" in a consumer AI interface does not guarantee that the vendor has not logged or retained the prompt for backend processing. Unless your enterprise has executed a formal Data Processing Agreement (DPA) or enterprise tier that guarantees data exemption from training, treat all prompt input as publicly exposed.

Maintaining strict data hygiene ensures you stay compliant while preparing for the accuracy verification practices covered in our next lesson.

Understanding AI Hallucinations and Model Bias

Because Large Language Models generate text based on statistical likelihood rather than factual retrieval, they can generate statements that appear articulate and persuasive but are entirely untrue. In generative AI terminology, this phenomenon is called a hallucination.

A hallucination is not a system crash or an overt error message; it is a fluent, grammatically sound assertion that invents facts, citations, dates, or calculations. Closely related is model bias, which occurs when an AI system reflects systemic skews, demographic stereotypes, or unrepresentative viewpoints present in its initial training data.

Realistic Workplace Risks

In a business setting, unverified AI outputs present serious operational risks:

• Fabricated citations and legal precedents: The model may invent court rulings, regulatory codes, or published academic papers complete with plausible volume numbers and author names.
• Inaccurate financial or numerical totals: Language models struggle with precise arithmetic and can miscalculate spreadsheet sums or quote incorrect inventory numbers.
• Biased language in job descriptions or communications: AI tools may default to gendered stereotypes or exclusionary phrasing when generating professional profiles or role responsibilities.

Guided Example: Detecting a Hallucinated Reference

Consider a knowledge worker asking an AI tool for corporate policy guidance:

Prompt: "What does ISO standard 9001:2015 Clause 14 state regarding remote work computer monitors?"

AI Response: "ISO 9001:2015 Clause 14 specifies that organizations must provide dual 24-inch monitors for all remote personnel to ensure quality management compliance."

This response sounds authoritative and formal. However, ISO 9001:2015 contains only 10 clauses; Clause 14 does not exist. An uncritical worker might incorporate this nonexistent standard into an official procurement policy, leading to unnecessary organizational expense and audit failure.

Verification Checkpoint: Human-in-the-Loop

To prevent errors, always implement a Human-in-the-Loop (HITL) review workflow:

1. Verify primary sources: If an AI response quotes a statute, standard, or metric, locate the original source document before sharing it.
2. Check internal calculations: Never rely on AI for mathematical tabulations without verifying formulas in a spreadsheet.
3. Assess tone and neutrality: Review all personnel communications to confirm they align with your organization’s equity and inclusivity guidelines.

This verification mindset prepares you directly for the upcoming hands-on lab, where you will evaluate live prompts and responses for compliance and factual integrity.