Quiz — Chapter 2: A Brief History of AI
Multiple Choice
1. The 1956 Dartmouth Conference is significant in AI history primarily because it: - (a) Produced the first working AI program - (b) Named the field and established a research community - (c) Proved that machines could pass the Turing Test - (d) Secured permanent government funding for AI research
Answer
**(b)** The Dartmouth Conference did not produce breakthroughs but gave the field its name ("artificial intelligence") and brought together researchers who realized they were working on related problems, establishing AI as a recognized field of study.2. Which of the following best describes the approach of symbolic AI? - (a) Learning patterns from large datasets - (b) Representing knowledge as symbols and rules, then manipulating them logically - (c) Training neural networks with many hidden layers - (d) Using reinforcement to reward correct behaviors
Answer
**(b)** Symbolic AI (also called GOFAI — Good Old-Fashioned AI) represented knowledge as explicit symbols and logical rules. It was the dominant approach in the 1960s and 1970s.3. The primary reason expert systems declined in the late 1980s was: - (a) They were never accurate enough to be useful - (b) The government banned their use in commercial applications - (c) They were expensive, brittle, and difficult to maintain as domains changed - (d) Deep learning immediately replaced them
Answer
**(c)** Expert systems were often accurate within their domains but were expensive to build, required constant manual updates when domain knowledge changed, and produced nonsensical results when faced with situations outside their programmed rules.4. The three converging forces that enabled the deep learning revolution were: - (a) Better algorithms, government funding, and quantum computing - (b) Larger datasets, more computing power (especially GPUs), and better algorithms - (c) The internet, social media, and smartphone cameras - (d) Moore's Law, the Turing Test, and expert systems
Answer
**(b)** The convergence of big data (fueled by the internet), increased computing power (particularly GPUs originally designed for video games), and improved algorithms (especially for training deep neural networks) enabled the deep learning revolution.5. The transformer architecture, introduced in 2017, differed from previous language processing approaches primarily because it: - (a) Was the first neural network ever built - (b) Could process all parts of the input simultaneously using attention mechanisms - (c) Used symbolic AI rules instead of statistical patterns - (d) Required no training data at all
Answer
**(b)** Previous approaches processed language sequentially (one word at a time). Transformers introduced attention mechanisms that allow the model to consider all parts of the input simultaneously, weighing which elements are most relevant to each other regardless of position.6. ELIZA (1966) is historically significant because it demonstrated that: - (a) Machines could genuinely understand human emotions - (b) Simple pattern matching could produce responses that humans attributed understanding to - (c) The Turing Test had been passed for the first time - (d) Natural language processing required deep learning
Answer
**(b)** ELIZA used simple pattern matching to simulate a therapist's responses. Despite its simplicity, users frequently attributed genuine understanding to the program, illustrating humans' tendency to perceive intelligence in systems that produce appropriate-seeming outputs.True or False
7. Alan Turing coined the term "Turing Test" in his 1950 paper.
Answer
**False.** Turing called his proposal "the imitation game." The term "Turing Test" was applied later by others.8. During AI winters, all meaningful AI research stopped completely.
Answer
**False.** Important research continued during both AI winters. Statistical methods, probabilistic reasoning, and neural network training techniques (like backpropagation) were developed or refined during these periods. The winters killed hype and funding, but not all progress.9. The deep learning model AlexNet (2012) was notable because it won an image recognition competition by a small margin, narrowly beating traditional approaches.
Answer
**False.** AlexNet won by a dramatic margin that stunned the field, not by a small margin. This decisive victory is what made it a watershed moment for deep learning.10. Expert systems like MYCIN were never accurate — they consistently performed worse than human experts.
Answer
**False.** MYCIN performed as well as or better than many human physicians in controlled tests for diagnosing bacterial infections. The problem with expert systems was not accuracy but brittleness, maintenance costs, and inability to handle situations outside their programmed rules.11. Large language models learn by processing enormous amounts of text and learning to predict what words typically follow other words.
Answer
**True.** This is a simplified but accurate description of how large language models are trained. They learn statistical patterns in language that allow them to generate fluent, contextually appropriate text.Short Answer
12. In two or three sentences, explain why the Lighthill Report (1973) was significant in AI history.
Answer
The Lighthill Report was a British government-commissioned evaluation of AI research that concluded AI had failed to deliver on its promises and that most successes were limited to toy problems that wouldn't scale. It led to severe funding cuts for AI research in the UK and helped trigger the first AI winter. It represents a key example of the "overcorrect" phase of the AI hype cycle.13. Explain the concept of "knowledge engineering" and why it became a bottleneck for expert systems.
Answer
Knowledge engineering was the process of interviewing domain experts, extracting their rules and heuristics, and encoding that knowledge into an expert system. It became a bottleneck because it was extremely labor-intensive, required specialized knowledge engineers, and had to be repeated every time the domain changed (new regulations, new procedures, new products). This made expert systems expensive to build and even more expensive to maintain.14. Name three of the five patterns identified in Section 2.6 and briefly explain one of them.
Answer
The five patterns are: (1) The hype cycle is real, (2) The hard problems are harder than they look, (3) Breakthroughs come from unexpected directions, (4) Demonstration is not deployment, and (5) "Is it different this time?" is always the right question. [Student should explain one in their own words with appropriate detail.]Applied Scenario
15. A company announces that its new AI system can "read and understand any legal contract and identify all potential risks." The system was tested on 500 contracts from a single law firm and achieved 95% accuracy. Using what you've learned about historical patterns in AI, evaluate this claim. Your response should reference at least two specific patterns from Section 2.6.
Answer
Strong answers should reference: **Pattern 4 (Demonstration is not deployment):** Testing on 500 contracts from a single firm is a controlled demonstration, not real-world deployment. Legal contracts vary enormously across firms, jurisdictions, industries, and time periods. 95% accuracy on one firm's contracts may not generalize. **Pattern 2 (Hard problems are harder than they look):** "Understanding" legal contracts requires knowledge of legal precedent, jurisdictional variations, and contextual judgment — the kind of open-ended, real-world complexity that has consistently proven harder than initial demonstrations suggest. The claim that it can handle "any" legal contract echoes the overconfident predictions of previous eras. Students might also reference **Pattern 1 (The hype cycle)** by noting the inflated language ("read and understand," "any legal contract") or **Pattern 5** by asking what's genuinely different about this technology versus expert system approaches to legal reasoning that were attempted in the 1980s.16. You're advising a city council that is considering adopting a predictive policing system similar to CityScope Predict. A council member says, "We should adopt this because AI has gotten incredibly good — look at ChatGPT and image recognition. These technologies are proven." Using this chapter's historical analysis, draft a three-to-four sentence response.
Answer
Strong answers should note that success in one domain (text generation, image recognition) does not automatically transfer to another domain (crime prediction) — this is analogous to how AlphaGo's mastery of Go didn't mean it could play checkers. They should reference the distinction between capability and understanding, noting that predictive policing systems find patterns in historical data that may encode decades of biased policing practices. They should also note the gap between demonstration and deployment — a system that performs well in a vendor's presentation may not perform well in the complex reality of a specific city. The broader point is that AI literacy means evaluating each application on its own merits rather than assuming that AI progress in general validates any specific AI application.17. A friend tells you, "AI winters can't happen anymore because this time AI is actually making money." Agree or disagree, and explain your reasoning using historical evidence from this chapter.