Introduction to Ai: What It is and Isn'T

• A. A physical robot that performs manual labor
• B. The ability of computer systems to perform tasks that typically require human intelligence, such as reasoning, learning, and perception ✓
• C. Any software program that stores and retrieves large amounts of data
• D. A network of computers that communicate over the internet

Why: AI refers to computer systems performing tasks that normally require human intelligence, such as learning, reasoning, and understanding language or images. It is not defined by physical form, data storage, or networking alone.

• A. 1956, at the Dartmouth Summer Research Project ✓
• B. 1936, in Alan Turing's paper on computable numbers
• C. 1969, with the launch of ARPANET
• D. 1997, when Deep Blue defeated Garry Kasparov

Why: John McCarthy coined the term 'artificial intelligence' for the 1956 Dartmouth Summer Research Project, widely regarded as the founding event of AI as a field. Turing's 1936 work laid theoretical groundwork but did not coin the term.

• A. Narrow AI is designed for a specific task, while general AI would match human-level flexibility across a wide range of tasks ✓
• B. Narrow AI runs on small computers, while general AI requires supercomputers
• C. Narrow AI is always rule-based, while general AI is always machine learning
• D. Narrow AI is illegal in most countries, while general AI is permitted

Why: Narrow AI excels at specific tasks such as image recognition or spam filtering. General AI would be able to apply intelligence flexibly across many domains like a human can. Hardware size and legal status are not what separates the two categories.

• A. ELIZA ✓
• B. Logic Theorist
• C. SHRDLU
• D. Perceptron

Why: ELIZA, created by Joseph Weizenbaum at MIT in the mid-1960s, used pattern matching and scripted responses to mimic a Rogerian therapist. Logic Theorist proved mathematical theorems, SHRDLU manipulated blocks in a virtual world, and the Perceptron was an early neural network model.

• A. Governments banned AI research after public protests
• B. Early promises outpaced the technology's real capabilities, leading to disappointment and cuts in funding ✓
• C. Computers became too powerful for AI algorithms to run efficiently
• D. All AI researchers moved to other fields after the Dartmouth conference

Why: In the 1970s and late 1980s, overly optimistic predictions were not met by the limited computing power and methods of the time, causing funders to withdraw support. The result was periods of stalled progress known as AI winters.

• A. To serve as the final output the model produces
• B. To provide examples from which the model learns patterns ✓
• C. To define the exact rules the programmer wants the model to follow
• D. To store the model's code after it has been written

Why: Training data consists of examples that the model analyzes to identify patterns and relationships. The model learns from these examples rather than being explicitly programmed with rules for every case.

• A. A physical robot that performs tasks autonomously
• B. A fixed set of hand-written if-then rules
• C. A mathematical function with adjustable parameters that maps inputs to outputs ✓
• D. A database that stores every possible answer to a question

Why: A model is a mathematical function whose parameters are adjusted during training so that its outputs become increasingly accurate for given inputs. It is not a physical device, a rule list, or a lookup database.

• A. The model memorizes every training example exactly
• B. The model's parameters are adjusted to reduce errors on data ✓
• C. The programmer manually updates the model's rules after each mistake
• D. The model gains consciousness and understanding of the data

Why: Learning means that the model's internal parameters are gradually adjusted based on how wrong its predictions are, so that performance improves. Memorization alone does not generalize, and learning does not imply consciousness.

• A. Because new data is always cheaper to collect than training data
• B. Because models must be tested on data that is always larger than the training set
• C. To check whether the model generalizes to new examples rather than just memorizing training data ✓
• D. Because the training data becomes corrupted once the model has been trained

Why: Testing on held-out data reveals whether the model has learned general patterns. A model that performs well only on its training data may have overfit, meaning it memorized specifics rather than learning transferable patterns.

• A. A system groups customers into clusters without any labels provided
• B. A model is given emails labeled 'spam' or 'not spam' and learns to classify new emails ✓
• C. A program follows a fixed recipe of steps written by a developer
• D. A robot explores a maze and receives no feedback at all

Why: Supervised learning uses labeled examples, where the correct answer is provided for each training input. Clustering without labels is unsupervised, fixed recipes involve no learning, and feedback-free exploration is not supervised learning.

• A. By being manually reprogrammed by engineers after every error
• B. By finding patterns in example data and adjusting its internal settings to reduce mistakes ✓
• C. By memorizing every possible answer before it is deployed
• D. By following a fixed list of if-then rules written in advance

Why: Machine learning systems learn from examples by detecting patterns in data and adjusting their internal parameters to make fewer errors over time, rather than relying solely on hand-written rules or memorization.

• A. Only the input data with no additional information
• B. A randomly generated number and a timestamp
• C. An input paired with the correct output or answer ✓
• D. A list of rules that the system must follow

Why: In supervised learning, each training example pairs an input (such as an image) with its correct label or output (such as 'cat'), which lets the system compare its guesses against the right answers.

• A. To make the training process run faster
• B. To check how well the model performs on examples it has never seen before ✓
• C. To delete incorrect examples from the training data
• D. To give the model additional rules to follow

Why: A test set of unseen examples shows whether the model has genuinely learned general patterns or has simply memorized the training data, which is called overfitting.

• A. A calculator computing 2 + 2 using a built-in addition procedure
• B. A spam filter that improves after being shown many emails labeled spam or not spam ✓
• C. A thermostat turning on heat when the temperature drops below 20 degrees
• D. A program that always sorts a list alphabetically using a fixed algorithm

Why: The spam filter learns its behavior from labeled examples, adjusting as it sees more data, whereas the calculator, thermostat, and sorting program all follow fixed, explicitly programmed rules.

• A. Engineers personally review each data point and write a rule for it
• B. The model's internal parameters are adjusted to reduce errors on the examples it is shown ✓
• C. The data is copied into a database that the model can search
• D. The model is physically modified with new hardware components

Why: Training means repeatedly adjusting the model's internal numerical parameters so its predictions better match the correct answers in the training data.

• A. A paper wall calendar
• B. A spam filter that sorts unwanted emails into a separate folder ✓
• C. A manual kitchen timer
• D. A printed paper map

Why: Spam filters learn patterns from large amounts of email data to decide which messages are likely unwanted, which is a common form of AI. The other options are non-digital tools that do not learn or make predictions.

• A. Hand-written rules listing every movie
• B. A recommendation system that finds patterns in viewing behavior ✓
• C. A random number generator
• D. A database that only stores titles alphabetically

Why: Recommendation systems analyze patterns in what users watch and rate, then predict what else they might enjoy. Random selection or simple alphabetical storage would not personalize suggestions based on past behavior.

• A. A standard light switch
• B. A voice assistant on a phone that responds to spoken commands ✓
• C. A basic wristwatch with hands
• D. A mechanical door lock

Why: Voice assistants convert spoken words into text and interpret their meaning using AI models trained on large amounts of speech. The other options operate mechanically without interpreting human language.

• A. A fixed-focus lens with no software
• B. Automatic scene detection that adjusts settings for a portrait or landscape ✓
• C. A film roll that must be changed manually
• D. A lens cap that protects the glass

Why: Modern cameras use AI to recognize scenes, faces, and subjects, then adjust exposure, focus, and color automatically. A fixed lens, film roll, or lens cap involves no learning or recognition.

• A. AI never uses any data at all
• B. AI learns patterns from data to make predictions or decisions, rather than relying only on explicitly coded rules ✓
• C. AI can only perform arithmetic
• D. AI requires a human to make every single decision in real time

Why: Many AI systems improve by learning patterns from data, which lets them handle situations their creators did not explicitly program. Traditional software typically depends on rules a developer wrote in advance.

• A. AI systems learn patterns from data
• B. AI systems understand meaning the way humans do ✓
• C. AI systems can be trained for specific tasks
• D. AI systems perform better with more quality data

Why: Many people assume AI understands language and the world like humans do. Most AI systems detect statistical patterns in data without human-like comprehension, which is why they can produce fluent but incorrect answers.

• A. True, because it outperforms humans at a complex task
• B. True, because it learned the game without help
• C. False, because it is specialized and cannot transfer its skill to unrelated tasks ✓
• D. False, because chess is too simple to require intelligence

Why: Game-playing AI is narrow. A chess engine cannot cook dinner, write a poem, or play checkers without retraining. Excelling at one task does not imply general intelligence.

• A. Advanced AI systems have consciousness once they pass a complex test
• B. Current AI systems have no established evidence of consciousness or feelings, even when they talk about them convincingly ✓
• C. AI systems feel emotions but cannot express them
• D. Only large language models have consciousness

Why: An AI can produce text that sounds emotional or self-aware because it was trained on human writing. Fluent self-description is not evidence of inner experience, and there is currently no accepted test that confirms consciousness in machines.

• A. AI systems are deliberately programmed to favor certain groups
• B. AI systems learn from data that can contain human biases, which the systems can reproduce or amplify ✓
• C. AI systems cannot process numerical data accurately
• D. Objectivity is impossible for any computer program

Why: Because AI learns patterns from historical data, any biases in that data, such as unfair hiring decisions or skewed representation, can be absorbed and repeated. Careful design, testing, and auditing are needed to reduce this.

• A. AI systems follow a fixed list of rules written by programmers for every situation
• B. AI systems make decisions by generating random outputs
• C. Many AI systems, especially deep learning models, derive behavior from patterns learned in training, which can be hard to fully explain ✓
• D. AI systems always explain their reasoning step by step to users

Why: Unlike traditional software with explicit rules, many modern AI models learn internal representations from data. This makes their behavior powerful but sometimes difficult to interpret, and it means they do not reliably explain their own reasoning.

• A. Reliably knowing what happened in the world five minutes ago without access to new information
• B. Summarizing a long article into a few key points ✓
• C. Physically repairing a broken appliance in a home
• D. Experiencing emotions the way a human does

Why: Summarizing text is a pattern-based language task that modern AI systems handle well. The other options require real-time awareness, physical embodiment, or subjective experience, which current AI lacks.

• A. It deliberately lies to users when it is uncertain
• B. It has a personal bias against certain topics
• C. It generates text based on learned patterns and can produce plausible-sounding but incorrect answers ✓
• D. It has access to a database of verified facts that occasionally gets corrupted

Why: AI language models predict likely text based on patterns in training data. This can yield fluent, confident answers that are factually wrong, often called hallucinations.

• A. AI understands the world exactly as humans do, just faster
• B. AI has common sense equal to an average adult
• C. AI can process and produce language impressively but lacks many forms of real-world understanding and grounded experience ✓
• D. AI has no capabilities beyond simple calculations

Why: Current AI can handle language and patterns remarkably well, but it lacks lived experience, physical grounding, and reliable common sense, and it does not understand the world the way humans do.

• A. The AI may be too fast to process applications accurately
• B. The AI may make errors or reflect biases in its training data, and its decisions may be hard to explain or appeal ✓
• C. The AI will automatically become biased against all applicants
• D. The AI cannot process numerical data

Why: High-stakes decisions require accountability. AI systems can inherit biases from training data and make mistakes, and their reasoning may be opaque, so human oversight and appeal processes are important.

• A. Generating text in multiple languages
• B. Recognizing patterns in large datasets
• C. Reliably remembering and learning from every individual conversation after it ends, unless specifically designed to do so ✓
• D. Translating simple sentences between languages

Why: Many AI chatbots do not retain memories across separate conversations by default. Pattern recognition, translation, and multilingual text generation are generally strengths of current systems.

• A. They always verify facts against a live database before answering
• B. They can generate plausible-sounding but incorrect information, sometimes called hallucination ✓
• C. They understand the meaning of text the same way humans do
• D. They never reflect biases present in their training data

Why: AI language models predict likely text based on patterns in training data, so they can produce fluent but false statements. They do not inherently verify facts, and users should check important claims.

• A. Because AI systems have personal opinions they form over time
• B. Because the training data may reflect historical or social biases ✓
• C. Because AI systems are programmed with deliberately unfair rules
• D. Because AI systems only work correctly in English

Why: AI systems learn patterns from data created by humans. If that data contains stereotypes or imbalances, the model can reproduce or amplify them, even without any intent to be unfair.

• A. Accept the AI's answer without question since it is computer-generated
• B. Use AI output as one input and verify it with qualified human experts ✓
• C. Share personal details freely so the AI can give more accurate advice
• D. Rely on AI only if it sounds confident in its answer

Why: AI can be useful for gathering information, but it can be wrong or lack context about a specific situation. Responsible use means checking outputs and consulting qualified professionals for high-stakes decisions.

• A. Anything typed into an AI tool is always automatically deleted immediately
• B. Information you enter may be stored or used depending on the provider's policies, so sensitive data should be handled carefully ✓
• C. AI tools cannot store any user information by design
• D. Privacy only matters when using AI for financial transactions

Why: Data handling varies by provider and settings. Responsible users avoid entering sensitive personal, confidential, or proprietary information unless they understand how it will be stored and used.

• A. It cannot process any language at all
• B. It identifies statistical patterns in data without necessarily grasping real-world meaning or context the way humans do ✓
• C. It only understands mathematical problems
• D. It can only answer questions it has been explicitly programmed to answer

Why: Many AI systems excel at pattern recognition and generating text, but this does not guarantee human-like comprehension. Recognizing this limit helps users avoid overtrusting outputs.