Chapter 18 Exercises: Generative AI — Multimodal

Chapter 18 Exercises: Generative AI — Multimodal

Section A: Recall and Comprehension

Exercise 18.1 Define the following terms in your own words, using no more than two sentences each: (a) diffusion model, (b) multimodal AI model, (c) inpainting, (d) deepfake, (e) content provenance, (f) voice cloning.

Exercise 18.2 Explain the core intuition behind how diffusion models generate images. Why is it important for business leaders to understand this mechanism, even at a conceptual level?

Exercise 18.3 Describe the difference between the "aesthetic quality gap" and the "accuracy gap" in AI-generated images. Using the opening scenario of this chapter, explain why the accuracy gap is more consequential for businesses than the aesthetic gap.

Exercise 18.4 List three business applications of text-to-speech technology and three business risks of voice cloning. For each risk, propose one mitigation strategy.

Exercise 18.5 Summarize the current limitations of AI video generation. What are the primary constraints that prevent video generation from being commercially viable for branded advertising as of 2026?

Exercise 18.6 Explain the distinction between the "training data question" and the "output question" in generative AI intellectual property law. Why does each question matter for a business using AI-generated content?

Exercise 18.7 What is the C2PA standard, and why might it become important for businesses that use or produce digital content? Name three types of organizations that have committed to the standard.

Section B: Application

Exercise 18.8: Content Strategy Assessment You are the marketing director for a mid-sized e-commerce company that sells home furnishings (furniture, lighting, textiles). Your current content production budget is $400,000 annually, split between product photography ($200,000), lifestyle images ($120,000), and social media graphics ($80,000).

(a) Identify which content categories are most suitable for AI-assisted generation and which should remain primarily human-produced. Justify your reasoning.
(b) Estimate the potential cost savings if you adopt AI-assisted generation for the categories you identified. What new costs would you incur (QA, legal review, tooling)?
(c) Design a quality assurance pipeline for AI-generated product images. What specific checks would you include, and who would perform them?
(d) Draft a brief (one-page) proposal for your CEO explaining the opportunity, the risks, and the recommended approach.

Exercise 18.9: Code Generation Tool Evaluation Your company's CTO is considering deploying GitHub Copilot Enterprise for a 25-person development team. She asks you to prepare a business case.

(a) Based on the productivity data cited in this chapter, estimate the expected productivity gains. Express this in terms of hours saved per week and equivalent full-time employees (FTEs).
(b) Identify three categories of development tasks where code generation tools are likely to deliver the most value and three where they are likely to be least helpful or potentially harmful.
(c) Tom's experience at Athena revealed that Copilot-generated code nearly introduced data quality bugs. What organizational safeguards would you recommend to capture the benefits while mitigating this risk?
(d) Calculate the break-even point: at what level of productivity improvement does the cost of GitHub Copilot Enterprise licenses pay for itself, given the team's average fully-loaded salary cost?

Exercise 18.10: IP Risk Assessment Your company has been using an AI image generation tool (without indemnification) to create social media marketing content for the past six months. The legal team has just read about the Getty v. Stability AI lawsuit and is concerned.

(a) Conduct a retrospective risk assessment: what specific IP risks has the company been exposed to during the past six months?
(b) Develop a forward-looking IP risk mitigation plan using Lena Park's five-point framework from the chapter. For each recommendation, specify the concrete actions your company should take.
(c) Your CMO argues that the risk is minimal because "nobody is going to sue us over social media graphics." Prepare a memo explaining why this reasoning is flawed, using specific examples from the chapter and current case law.

Exercise 18.11: Build-vs-Buy Analysis for Generative AI A healthcare company wants to use generative AI to produce patient education materials — simple explanatory documents with text and illustrations explaining common medical procedures. The materials must be accurate, culturally sensitive, and available in six languages.

(a) Evaluate the three deployment models (API, fine-tuned, self-hosted) for this use case. For each model, identify the primary advantages and risks specific to healthcare.
(b) Which deployment model would you recommend, and why? Consider cost, accuracy requirements, data privacy regulations (HIPAA), and organizational capability.
(c) What human oversight mechanisms would you require before any AI-generated patient education material is distributed?

Exercise 18.12: Deepfake Preparedness Plan You are the chief communications officer for a publicly traded consumer goods company. Design a deepfake response plan that addresses the following scenarios:

(a) A deepfake video of your CEO making offensive remarks goes viral on social media.
(b) A competitor uses AI-generated images that closely resemble your flagship product in their advertising.
(c) A fraudster uses voice cloning to impersonate your CFO and request a wire transfer from your finance team.

For each scenario, outline: (i) immediate response steps (first 2 hours), (ii) short-term actions (first 48 hours), and (iii) preventive measures you would implement in advance.

Section C: Analysis and Evaluation

Exercise 18.13: The Creative Industry Debate The chapter presents two narratives about generative AI's impact on creative professionals: the "displacement narrative" (AI replaces creative jobs) and the "augmentation narrative" (AI changes but doesn't eliminate creative work).

(a) Identify two additional pieces of evidence (not from the chapter) that support the displacement narrative.
(b) Identify two additional pieces of evidence (not from the chapter) that support the augmentation narrative.
(c) NK argues that if every brand uses the same AI tools, brands will "start looking the same." Evaluate this argument. Under what conditions is she right? Under what conditions might she be wrong?
(d) How does the impact of generative AI on creative work compare to the impact of desktop publishing software in the 1980s-90s or stock photography in the 2000s? What historical patterns can inform our predictions about creative work?

Exercise 18.14: Professor Okonkwo's Distinction Professor Okonkwo states: "Generative AI doesn't create. It interpolates." Evaluate this claim.

(a) Explain what Okonkwo means by "interpolates" in this context. How does this relate to how diffusion models and language models work?
(b) Identify a scenario in which this framing usefully prevents a business leader from overestimating generative AI's capabilities.
(c) Identify a scenario in which this framing might cause a business leader to underestimate generative AI's capabilities.
(d) Is the distinction between "creation" and "interpolation" philosophically meaningful when applied to human creativity? Why or why not? How does your answer affect the business implications?

Exercise 18.15: The Verification Imperative The chapter argues that "the hardest part of generative AI isn't generation — it's verification." Evaluate this claim from multiple perspectives.

(a) For which business applications is the verification challenge most acute? Rank the following from easiest to hardest to verify: (i) social media marketing graphics, (ii) product catalog photography, (iii) patient education materials, (iv) code generation, (v) financial report narratives, (vi) synthetic training data.
(b) What organizational capabilities are required for effective verification? How do these differ from the capabilities required for generation?
(c) As generative AI quality improves, will the verification challenge become easier or harder? Defend your answer.

Exercise 18.16: Multimodal Model Strategy Compare and contrast the business implications of specialized models (a separate model for image generation, a separate model for text, a separate model for code) versus unified multimodal models (a single model that handles text, images, audio, and code).

(a) For what types of business use cases would specialized models be preferable? Why?
(b) For what types of business use cases would a unified multimodal model be preferable? Why?
(c) How does the build-vs-buy decision differ for specialized versus multimodal models?
(d) What are the vendor lock-in implications of each approach?

Section D: Research

Exercise 18.17: Industry Case Study Research a company (not mentioned in this chapter) that has publicly discussed its use of multimodal generative AI in its business operations. Write a 500-word analysis covering:

(a) What specific multimodal AI capabilities the company deploys
(b) The business results reported (cost savings, productivity gains, revenue impact)
(c) The challenges or limitations encountered
(d) How the company's approach aligns with or diverges from the frameworks presented in this chapter

Exercise 18.18: Legal Landscape Update Research the current status of one of the following lawsuits: (a) The New York Times v. OpenAI, (b) Getty Images v. Stability AI, (c) Authors Guild v. OpenAI, (d) the music industry lawsuits against Suno/Udio.

Write a 400-word briefing for a non-legal business audience covering: the current status of the case, any significant rulings or settlements, and the practical implications for businesses using generative AI tools.

Exercise 18.19: Content Provenance Research Investigate the C2PA standard in depth. Visit the C2PA website and review the standard's specification.

(a) How does C2PA technically work? What metadata is embedded, and how is it verified?
(b) Which companies and organizations have committed to the standard? What percentage of the content creation and distribution ecosystem do they represent?
(c) What are the limitations of the C2PA approach? How could it be circumvented?
(d) Should C2PA adoption be voluntary or mandated by regulation? Argue both sides.

Section E: Athena Application

Exercise 18.20: Athena's Content Creation Roadmap Based on the Athena updates in this chapter (AI-generated marketing images, visual QA pipeline, IP risk mitigation), design a 12-month roadmap for expanding Athena's use of multimodal generative AI.

(a) Identify three additional content categories (beyond product photography) where Athena should pilot AI generation. For each, estimate the business case and identify the primary risks.
(b) Propose an organizational structure for Athena's AI content team. What roles are needed? How does this team interact with the existing marketing, legal, and technology teams?
(c) Define the metrics Athena should track to measure the success of its AI content strategy. Include both efficiency metrics (cost, speed) and quality metrics (error rate, brand consistency, customer perception).
(d) NK says "If every brand uses the same AI tools, don't we all start looking the same?" How should Athena ensure that its AI-generated content maintains brand distinctiveness?

Answers to selected exercises are available in Appendix B.