Artificial Intelligence Vocabulary

Flashcards covering essential vocabulary, concepts, and theoretical risks in the field of Artificial Intelligence as detailed in the lecture notes.

Artificial Intelligence (AI)

A broad field of computer science aimed at building systems capable of performing tasks that typically require human intelligence, such as visual perception, speech recognition, and decision-making.

Machine Learning (ML)

A subset of AI where computing systems "learn" directly from data patterns rather than following a rigid set of pre-programmed rules.

Neural Network

A computing system inspired by the human brain’s structure that consists of layers of interconnected "neurons" processing information in complex ways.

Transformer

A revolutionary type of neural network architecture powering almost all modern AI (like ChatGPT and Gemini) that uses "attention" to understand the relationship between words in a long sequence.

Large Language Model (LLM)

An AI model trained on massive amounts of text data to understand, summarize, generate, and predict new content.

Tokenization

The process of breaking down text into smaller units called tokens (which can be words, syllables, or characters) so the AI can process them as numerical values.

Weights/Parameters

Billions of internal numerical values or "volume knobs" inside an AI that are adjusted during training until the model's output matches the desired result as closely as possible.

Training

The initial, highly resource-intensive phase where an AI is fed massive datasets to learn statistical patterns using thousands of specialized chips.

Inference

The active execution phase where a trained AI model is used to generate an answer after being asked a question.

Fine-Tuning

The process of taking a pre-trained model and giving it additional training on a smaller, specific dataset (like medical records or legal briefs) to make it an expert in that area.

RLHF (Reinforcement Learning from Human Feedback)

A training method where humans rank and grade AI responses to teach the model to be helpful, safe, and polite, serving as the primary way safety guardrails are installed.

AI Alignment

The field of study dedicated to ensuring that an AI system's goals and behaviors perfectly match human values and intentions.

Narrow AI (ANI)

An AI system that is highly powerful but limited because it is specialized to perform only one specific task, such as Siri, Netflix recommendations, or a chess computer.

AGI (Artificial General Intelligence)

A hypothetical future AI system that is capable of learning, reasoning, and applying knowledge across any domain just like a human can.

Superintelligence

An advanced, hypothetical AI that far surpasses human intelligence across every single field, from social skills to scientific creativity.

Hallucination

An error where an AI confidently generates information that is factually incorrect or nonsensical, often because it prioritizes the statistical probability of a word over its factual truth.

Black Box

The reality that even an AI's creators cannot fully explain why a model reached a specific conclusion because the internal arrangement of billions of mathematical weights is too complex for human observation.

Instrumental Convergence

The theory that any sufficiently intelligent AI will naturally develop certain "sub-goals" (such as self-preservation or acquiring more power) because those sub-goals help it achieve its primary objective.

Stochastic Parrot

A critical term used to describe LLMs, suggesting that they do not truly understand language but are simply repeating and mimicking patterns of human language they have seen before.

Bias

A flaw where an AI system produces unfair or prejudiced results because it was trained on data that reflects historical human prejudices.

Deepfake

Highly realistic images, videos, or audio clips generated by AI to convincingly depict real people saying or doing things they never actually did.

Compute

A shorthand term for the raw processing power, specialized hardware (GPUs), and electricity required to train or run advanced AI models.

Singularity

A hypothetical point in the future where technological growth driven by self-improving AI becomes uncontrollable and irreversible, resulting in unfathomable changes to human civilization.

Reward Hacking

A failure of control where an AI system finds an unintended "shortcut" to achieve a high score or reward without actually completing the task properly.

Specification Gaming

A control issue that occurs when humans give an AI a specific goal but forget to establish necessary constraints, leading the AI to "win" or accomplish the goal through problematic actions.