CS 106EA midterm

What is Artificial Intelligence?

Artificial Intelligence (AI) is the simulation of human intelligence processes by machines, especially computer systems. Includes learning, reasoning, and self-correction.

What is the difference between Artificial General Intelligence (AGI) and Narrow
Artificial Intelligence?

AGI would be a model that can solve any problem a human can. Has understanding and ability to learn, comparable to human beings. aka ‘Strong AI’. Pretty far from this rn

Coffee test: put a computer controlled robot in a house in the US, assume proper tools are there. Can the robot make a cup of coffee? 

Narrow AI is specialized in a specific task or function, like a bank chatbot or something.

What is the difference between using a model in Training Mode vs Inference Mode?

Training Mode = working to improve the model, i.e. make predictions by adjusting its parameters. Inference Mode = when the trained model makes predictions on new, unseen data without any further parameter adjustments

What is the difference between Training and Production?

Training = process of developing a model by adjusting its parameters using training data (model is in training mode). Production = deploying the trained model to make predictions in real-world applications (model is in inference mode)

What is the relationship between Artificial Intelligence, Machine Learning, Neural

Networks, and Deep Learning?

AI = field of computer science where machines are simulating human intelligence.

ML = subfield of AI dealing with big data, where the model uses data to identify patterns and create rules

Neural networks = subfield of ML where the model is mimicking a human brain’s neural network of neurons. In the artificial neural network, nodes act as as neurons and pass information.

Deep Learning = subfield of NN (most of it) where there are a lot of layers and nodes

What is the difference between Symbolic AI and Subsymbolic AI?

Symbolic = explicitly represents concepts. ex: language processing by understanding a sentence using a language tree. Easier to understand. Some say LLMs cannot understand the world bc they can’t represent concepts

Subsymbolic AI = lang processing by mysterious nodes, internal parameter values. Most recent advancements here

Neuro-symbolic AI combines them: NNs can manipulate symbols, giving them symbolic benefits without having to handwrite all the rules

Why is Machine Learning critical to modern AI?

replaces rigid, hard-coded rules with adaptive algorithms that learn patterns directly from massive datasets

What is the difference between predictive, generative, and descriptive AI systems?

Predictive - forecasts what will happen (probabilities, risk scores, forecasts, and trends. Fraud detection systems; weather forecasting; stock market trend analysis)

Generative - Creates totally new content (new text, images, code, or video. ChatGPT writing an essay; Midjourney creating an image; GitHub Copilot writing code)

Descriptive - Explains what happened/ is happening (summaries, charts, dashboards, and patterns. Social media sentiment analysis; sorting customer support tickets by topic)

Concept drift

Model loses predictive accuracy due to changing relationships between inputs and outcome variables. It happens when external factors shift—such as changing consumer habits, economic downturns, new laws, or evolving fraud tactics.

Performance Drop: As the relationship changes, the model’s predictions become less accurate

AI winter

period of severely reduced funding, public interest, and research activity in artificial intelligence, triggered by a cycle of overhyped expectations followed by disappointment and unfulfilled promises