PART 3

This set of flashcards covers the historical timeline of AI, key definitions of AI types (Symbolic vs. Connectionism), and the cognitive science principles underlying AI development as discussed in Week 5.

Symbolic AI (1950s-1970s)

A period in AI history focused on logic, rules, and the theory of minds as symbol manipulators.

Connectionism (1980s-1990s)

Brain-inspired learning systems using neural networks and parallel distributed processing.

Data-driven machine learning (2000s)

A shift in AI where statistical models learn patterns from large datasets instead of hand-coded rules.

Deep learning (2010s-Present)

An era characterized by huge neural networks, breakthroughs in vision and language, and the rise of LLMs.

Turing Test (1950)

A method proposed by Alan Turing to evaluate machine intelligence through conversation.

Dartmouth conference (1956)

The event recognized as the birth of Artificial Intelligence as a field.

Deep Blue

The first major symbolic AI to defeat a chess world champion (Kasparov) in 1997.

ImageNet breakthrough (2012)

A moment when deep neural nets outperformed humans in object recognition tasks.

AlphaGo

A system using reinforcement learning and deep learning that defeated Lee Sedol in the game of Go in 2016.

Artificial Intelligence (AI)

A science and set of computational technologies inspired by the ways people use their nervous systems and bodies to sense, learn, reason, and take action.

Top-down model approach

An approach to AI that moves from abstraction to reasoning, often logic-based and using production rules (if-then phrases).

Bottom-up model approach

An AI approach that consists of 'building a brain' and letting the machine learn to solve problems and recognize patterns from data.

Universal computing machine

A concept from Turing (1937) stating that if an algorithm can be formulated for a problem, a machine can be built to implement the solution.

Symbolic Representation

A way information is stored using explicit symbols and rules where meaning is clear, such as $\text{red} = \text{red}$.

Distributed Representation

Patterns across many units in a neural net where no single unit holds the meaning, making it not human-readable.

Hebb’s rule

The principle that repetition leads to stronger associations in human learning.

Generalization

The ability of an intelligence system to apply past experience to new inputs.

Predictive AI

A machine learning technique using statistical models and algorithms to output quantities of distinct categories or future outcomes.

Loss

A metric in AI decision-making that measures how far an output is from the correct answer.

Reward

A value used in AI optimization that grades the quality of an action.

Large Language Models (LLMs)

Models like GPT, Gemini, and Claude designed to generate and understand text.

Multimodal models

AI systems that integrate multiple forms of data such as text, vision, and audio.

Flashbulb memory

A memory created during an emergency or emotional event that is remembered well.

Violation of expectation method

A way to study infant cognition based on the observation that infants look longer when they are surprised by a scene.

Phonology, morphology, syntax/semantics, pragmatics

The identified components of language according to the Quiz 4 materials.

Turing’s State Variable Approach

An approach involving mutual interaction where signals from the environment determine transitions between machine states, and state signals control the environment.

Mood-dependent memory

A phenomenon where memory retrieval is affected by being in the same mood as when the memory was formed.

ImageNet

A large-scale dataset that led to a breakthrough in 2012 when deep neural nets outperformed humans in object recognition.