Exercises — Chapter 2: A Brief History of AI

Part A: Conceptual Questions ⭐

A.1. What is the Turing Test, and what question was it designed to address? Why did Turing propose a behavioral test rather than trying to define "thinking" directly?

A.2. Define "symbolic AI" in your own words. Give one strength and one weakness of this approach.

A.3. What is an AI winter? Describe the general sequence of events that leads to one.

A.4. Explain the difference between an expert system and a deep learning system in terms of how each one's "knowledge" is created. Who does the work of identifying important patterns in each case?

A.5. What three forces converged in the 2000s–2010s to make deep learning practical? Why was each one necessary?

Part B: Applied Questions ⭐⭐

B.1. A startup claims it has built an AI that can "understand and respond to any question about tax law." Using what you know about the history of expert systems, list three questions you would want answered before investing in this company.

B.2. Imagine you're a government official in 1985, deciding whether to fund an expert system project to help social workers assess child welfare cases. Based on what you've learned about expert systems, what are two potential benefits and two potential risks of this approach?

B.3. A news headline reads: "AI Now Outperforms Doctors at Diagnosing Skin Cancer." Based on Pattern 4 from Section 2.6 ("Demonstration Is Not Deployment"), write three follow-up questions a journalist should ask before publishing this headline.

B.4. Consider ContentGuard, the social media content moderation system. If ContentGuard had been built in each of the following eras, how might its approach have differed? - 1985 (expert systems era) - 2015 (deep learning era) - 2023 (transformer/LLM era)

B.5. The chapter describes how predictive policing systems like CityScope Predict can encode historical biases. Explain how this connects to the distinction between finding "truth" in data versus finding "patterns" in data.

Part C: Skills-Based Questions ⭐⭐–⭐⭐⭐

C.1. Find a recent news article (from the past 12 months) that makes a bold prediction about AI's future capabilities. Identify: - The specific claim being made - Whether the claim is about demonstration or deployment - Which historical pattern(s) from Section 2.6 it most closely resembles - What evidence would you need to evaluate the claim

C.2. Create a one-page timeline of AI history that you could use to explain the field's development to a friend who knows nothing about technology. Choose the five most important events and explain why you chose each one. Your choices reveal your priorities — be ready to defend them.

C.3. Read the first three pages of Turing's 1950 paper "Computing Machinery and Intelligence" (freely available online). Identify one argument Turing makes that you find surprisingly relevant to today's AI debates, and one that feels dated. Explain your reasoning.

C.4. Interview someone over 50 about their memories of AI. What did they expect AI to become? How does their experience compare to the historical narrative in this chapter? Write a one-paragraph reflection.

Part D: Synthesis Questions ⭐⭐⭐

D.1. The chapter identifies a recurring tension between transparency and capability: expert systems were transparent but limited; deep learning systems are more capable but opaque. Is this trade-off inevitable, or could we build systems that are both highly capable and fully transparent? Argue both sides, drawing on historical examples.

D.2. Consider the following thought experiment: If the internet had never been invented, but computing power had continued to grow as it did, would deep learning and large language models have been developed? Why or why not? What does your answer reveal about the relationship between AI progress and the broader technological ecosystem?

D.3. The hype cycle pattern suggests that AI goes through predictable phases of excitement and disillusionment. Could a funder use this knowledge strategically — investing during winters when research is cheap and underfunded, and selling during booms? What are the ethical implications of treating transformative technology as a financial cycle?

D.4. The Dartmouth Conference attendees claimed that "every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it." Seventy years later, do you think this claim has been vindicated, refuted, or is the jury still out? Support your answer with specific examples from the chapter.

Part M: Mixed and Interleaved Questions ⭐⭐–⭐⭐⭐

M.1. (Connects Ch.1 and Ch.2) In Chapter 1, we discussed how the definition of AI keeps shifting. Using specific examples from this chapter, explain how the "AI effect" (once a machine can do something, we stop calling it AI) has operated across different historical eras.

M.2. (Connects Ch.1 and Ch.2) In Chapter 1, we introduced the idea that AI systems can do things without understanding them. Trace this concept through the historical eras: How did ELIZA, expert systems, and large language models each demonstrate capability without understanding? Has the gap between doing and understanding narrowed over time, or just changed character?

M.3. (Application across eras) For each of the three anchor examples below, identify which historical era's approach would be most and least appropriate for the system's task, and explain why: - ContentGuard (content moderation) - MedAssist AI (medical diagnosis) - CityScope Predict (predictive policing)

Part E: Extension Questions ⭐⭐⭐⭐

E.1. Research one AI project not mentioned in this chapter from the 1960s–1980s era (suggestions: Shakey the robot, AM/Eurisko, Cyc, LISP machines). Write a 500-word profile explaining what the project attempted, what it achieved, why it mattered, and how it fits into the broader narrative of this chapter.

E.2. The chapter focuses primarily on AI development in the United States and United Kingdom. Research AI history in one other country or region (suggestions: Japan's Fifth Generation Project, Soviet/Russian AI research, Chinese AI development). How does that country's experience complicate or enrich the narrative presented here?

E.3. Some historians of technology argue that AI's hype cycles are a feature, not a bug — that unrealistic expectations are necessary to attract the funding that eventually produces real breakthroughs. Write a 400-word argument for or against this claim, using specific examples from AI history.

E.4. Locate and read the abstract and introduction of the 2017 paper "Attention Is All You Need" by Vaswani et al. You may not understand all the technical content, and that's fine. What can you learn about the paper's goals, approach, and self-described contributions from just these sections? How does the paper's tone compare to the Dartmouth proposal's ambitions?