AI as goal-directed intelligent behavior means an AI system performs tasks associated with human intelligence to defined goals.

Context: AI as goal-directed intelligent behavior

The general problem of simulating intelligence is broken into traits like reasoning, knowledge representation, planning, learning, language, and perception; this is called decomposition of intelligence.

Context: Subproblem decomposition of intelligence

Reasoning under uncertainty is difficult partly because exact search can suffer from , where the number of possibilities grows exponentially.

Context: Combinatorial explosion

A is a representation of knowledge as concepts within a domain and the relationships between them.

Context: Ontology

A is a body of knowledge represented in a form that a program can use.

Context: Knowledge base

In decision-making, an agent chooses actions by maximizing expected , which averages utilities over possible outcomes weighted by probabilities.

Context: Utility and expected utility

A is a decision model that includes a transition model (probabilities of state changes) and a reward function (utility/cost).

Context: Markov decision process (MDP)

Machine learning improves performance automatically, including supervised, unsupervised, and reinforcement learning; deep learning is a major approach within paradigms.

Context: Machine learning paradigms

Modern NLP increasingly uses embeddings and transformer-based models; the core model family is the architecture.

Context: Transformer architecture

A local optimization method that iteratively adjusts parameters to minimize a loss function is called .

Context: Gradient descent

Perception from sensors aims to infer aspects of the world; is the subset of perception focused on visual analysis.

Context: Perception and computer vision

AI techniques include state space search, local optimization (e.g., gradient descent), and formal (propositional/predicate) to solve problems and reason.

Context: AI techniques: search, optimization, and logic

Cause → effect: GPUs began being used to accelerate neural networks (around 2012) which led to outperformed previous AI techniques, increasing funding and interest substantially.

Context: Cause→effect chain: GPUs → deep learning performance

Cause → effect: Transformer architecture emerged and was adopted (after 2017), which caused AI growth to further.

Context: Cause→effect chain: Transformers → accelerated growth

Affective computing systems simulate or recognize emotions in interaction, which can cause naïve users to develop an unrealistic conception of the of computer agents.

Context: Affective computing and trust limitations