# Encyclopedia of Artificial Intelligence --- ## Table of Contents 1. [Introduction](introduction) 2. [Major Concepts in AI](#major-concepts-in-ai) - [Artificial Intelligence (AI)](#artificial-intelligence-ai) - [Machine Learning (ML)](#machine-learning-ml) - [Supervised Learning](#supervised-learning) - [Unsupervised Learning](#unsupervised-learning) - [Reinforcement Learning (RL)](#reinforcement-learning-rl) - [Neural Networks (NN)](#neural-networks-nn) - [Deep Learning](#deep-learning) - [Symbolic AI](#symbolic-ai) - [Natural Language Processing (NLP)](#natural-language-processing-nlp) - [Artificial General Intelligence (AGI)](#artificial-general-intelligence-agi) 3. [History of AI: Decade by Decade](#history-of-ai-decade-by-decade) - [1940s](#1940s) - [1950s](#1950s) - [1960s](#1960s) - [1970s](#1970s) - [1980s](#1980s) - [1990s](#1990s) - [2000s](#2000s) - [2010s](#2010s) - [2020s](#2020s) 4. [Landmark Algorithms](#landmark-algorithms) - [Perceptron Learning](#1-perceptron-learning) - [Backpropagation](#2-backpropagation) - [ID3 Decision Tree](#3-id3-decision-tree) - [Q-Learning](#4-q-learning) - [A\* Search](#5-a-search) - [Monte Carlo Tree Search (AlphaGo)](#6-monte-carlo-tree-search-alphago) - [Support Vector Machine (SVM)](#7-support-vector-machine-svm) - [K-Means Clustering](#8-k-means-clustering) - [Naive Bayes Classifier](#9-naive-bayes-classifier) - [Transformer](#10-transformer) 5. [Philosophy of Mind](#philosophy-of-mind) 6. [AI Ethics and Alignment](#ai-ethics-and-alignment) 7. [The Next 100 Years: Predictions](#the-next-100-years-predictions) 8. [Index and Glossary](#index-and-glossary) --- ## Introduction Artificial Intelligence (AI) is the field devoted to creating systems capable of tasks that require intelligence when performed by humans. This encyclopedia provides a rigorous, comprehensive, and interconnected overview of AI, its history, its key concepts, landmark algorithms (with formal and code-level details), philosophical and ethical considerations, predictions, and glossary. --- ## Major Concepts in AI ### Artificial Intelligence (AI) **Definition:** AI is the study and design of agents that perceive their environment and take actions to maximize their chance of achieving goals. **Formalization:** An AI agent can be modeled as a function \[ \pi: \mathcal{O}^* \to \mathcal{A} \] where: - \(\mathcal{O}\) is the set of possible observations, - \(\mathcal{A}\) is the set of possible actions, - \(\mathcal{O}^*\) denotes sequences of observations (history). --- ### Machine Learning (ML) **Definition:** Machine Learning is a subset of AI focused on algorithms that improve their performance at tasks through experience. **Formalization:** Given a data distribution \(\mathcal{D}\) over input-output pairs \((x, y)\), a hypothesis class \(\mathcal{H}\), and a loss function \(\ell: \mathcal{Y} \times \mathcal{Y} \rightarrow \mathbb{R}\), ML aims to find \[ h^* = \arg\min_{h \in \mathcal{H}} \mathbb{E}_{(x, y) \sim \mathcal{D}} [\ell(h(x), y)] \] --- ### Supervised Learning **Definition:** Learning a function \(f: \mathcal{X} \rightarrow \mathcal{Y}\) from labeled data \(\{(x_i, y_i)\}_{i=1}^n\). **Formalization:** \[ \min_{f \in \mathcal{F}} \frac{1}{n} \sum_{i=1}^{n} \ell(f(x_i), y_i) \] --- ### Unsupervised Learning **Definition:** Learning patterns from unlabeled data \(\{x_i\}_{i=1}^n\). **Formalization:** Given a model \(p_\theta(x)\), find parameters \(\theta^*\) maximizing the likelihood: \[ \theta^* = \arg\max_\theta \prod_{i=1}^n p_\theta(x_i) \] --- ### Reinforcement Learning (RL) **Definition:** Learning how to act in an environment to maximize cumulative reward. **Formalization:** Modeled as a Markov Decision Process (MDP) \((\mathcal{S}, \mathcal{A}, P, R, \gamma)\): - \(\mathcal{S}\): states - \(\mathcal{A}\): actions - \(P(s'|s,a)\): transition probability - \(R(s,a)\): reward function - \(\gamma\): discount factor Find policy \(\pi(a|s)\) maximizing \[ \mathbb{E} \left[ \sum_{t=0}^{\infty} \gamma^t R(s_t, a_t) \right] \] --- ### Neural Networks (NN) **Definition:** A class of parameterized functions inspired by biological neurons. **Formalization:** A feedforward neural network with \(L\) layers: \[ h^{(0)} = x\\ h^{(l)} = \sigma(W^{(l)} h^{(l-1)} + b^{(l)}) \quad \text{for } l = 1,\ldots, L \] where \(\sigma(\cdot)\) is a (nonlinear) activation function. --- ### Deep Learning **Definition:** ML using neural networks with multiple layers (depth \(L > 2\)), enabling hierarchical feature learning. --- ### Symbolic AI **Definition:** AI systems based on explicit rules, symbols, and logic. **Formalization:** Knowledge is encoded as a set of logical statements (e.g., in first-order logic): \[ \forall x, \text{Cat}(x) \implies \text{Mammal}(x) \] Inference proceeds via deduction (e.g., Modus Ponens). --- ### Natural Language Processing (NLP) **Definition:** AI dealing with the interaction between computers and natural languages. **Formalization:** Tasks include: - Language modeling: estimate \(p(w_t \mid w_{<t})\) - Sequence-to-sequence mapping: \(f: \mathcal{W}^n \to \mathcal{W}^m\) --- ### Artificial General Intelligence (AGI) **Definition:** A hypothetical AI with the ability to understand, learn, and apply intelligence broadly as humans do. --- ## History of AI: Decade by Decade ### 1940s - **Key Figures:** Alan Turing, Warren McCulloch, Walter Pitts, Norbert Wiener - **Breakthroughs:** - 1943: McCulloch & Pitts model neuron as binary threshold device - 1948: Turing's "Intelligent Machinery" memo - 1948: Wiener's "Cybernetics" --- ### 1950s - **Key Figures:** Alan Turing, John McCarthy, Marvin Minsky, Allen Newell, Herbert Simon - **Breakthroughs:** - 1950: Turing Test ("Computing Machinery and Intelligence") - 1956: Dartmouth Conference (birth of AI) - 1957: Rosenblatt invents the perceptron - 1956-59: Logic Theorist, General Problem Solver (Newell & Simon) --- ### 1960s - **Key Figures:** Marvin Minsky, John McCarthy, Joseph Weizenbaum - **Breakthroughs:** - 1965: Robinson's resolution principle for automated theorem proving - 1966: ELIZA chatbot (Weizenbaum) - Development of LISP and SHRDLU --- ### 1970s - **Key Figures:** Allen Newell, Herbert Simon, Ed Feigenbaum - **Breakthroughs:** - 1972: MYCIN expert system (rule-based medical diagnosis) - 1976: Knowledge representation frameworks --- ### 1980s - **Key Figures:** Geoffrey Hinton, David Rumelhart, Judea Pearl - **Breakthroughs:** - 1986: Backpropagation algorithm (Rumelhart, Hinton, Williams) - 1982: Hopfield networks - 1988: Bayesian networks (Pearl) --- ### 1990s - **Key Figures:** Yann LeCun, Vladimir Vapnik, Richard Sutton - **Breakthroughs:** - 1997: IBM Deep Blue defeats Garry Kasparov - 1995: Support Vector Machines (Cortes & Vapnik) - 1992: Q-learning (Watkins) --- ### 2000s - **Key Figures:** Geoffrey Hinton, Yoshua Bengio, Yann LeCun - **Breakthroughs:** - 2006: "Deep Learning" resurgence (Hinton et al., deep belief nets) - 2009: ImageNet dataset (Fei-Fei Li) - 2001: Random forests (Breiman) --- ### 2010s - **Key Figures:** Demis Hassabis, Ilya Sutskever, Ian Goodfellow - **Breakthroughs:** - 2012: AlexNet wins ImageNet - 2014: Generative Adversarial Networks (Goodfellow et al.) - 2017: Transformer architecture (Vaswani et al.) - 2016: AlphaGo defeats Lee Sedol --- ### 2020s - **Key Figures:** Sam Altman, Yann LeCun, Dario Amodei - **Breakthroughs:** - 2020: GPT-3 (OpenAI) - 2022: ChatGPT and widespread deployment of LLMs - 2021: AlphaFold2 for protein folding (DeepMind) - 2023: Open-source LLMs (LLaMA, Falcon, etc.) --- ## Landmark Algorithms ### 1. Perceptron Learning **Mathematical Formulation:** Given labeled data \(\{(x_i, y_i)\}_{i=1}^n\), \(y_i \in \{-1, +1\}\): \[ \text{If } y_i(w \cdot x_i + b) \leq 0: \quad w \leftarrow w + \eta y_i x_i, \quad b \leftarrow b + \eta y_i \] **Pseudocode:** ```python # Perceptron Learning Algorithm initialize w = 0, b = 0 for epoch in 1..N: for x, y in data: if y * (dot(w, x) + b) <= 0: w += eta * y * x b += eta * y ``` --- ### 2. Backpropagation **Mathematical Formulation:** Given NN parameters \(\theta\), inputs \(x\), targets \(y\), loss \(\ell\): \[ \theta \leftarrow \theta - \eta \nabla_\theta \ell(f_\theta(x), y) \] **Pseudocode:** ```python # Backpropagation for one step forward propagate x through network to get output y_hat compute loss L = loss(y_hat, y) compute gradients dL/dtheta via chain rule (reverse mode autodiff) update theta = theta - eta * dL/dtheta ``` --- ### 3. ID3 Decision Tree **Mathematical Formulation:** At each node, select feature \(A\) maximizing information gain: \[ \text{Gain}(S, A) = H(S) - \sum_{v \in \text{Values}(A)} \frac{|S_v|}{|S|} H(S_v) \] where \(H(S)\) is the entropy of sample set \(S\). **Pseudocode:** ```python def ID3(S, features): if all same label in S: return leaf(label) if features empty: return leaf(majority label) best = argmax_A information_gain(S, A) tree = Node(best) for v in values(best): Sv = subset(S, best == v) if Sv empty: tree.add_branch(v, leaf(majority label)) else: tree.add_branch(v, ID3(Sv, features - {best})) return tree ``` --- ### 4. Q-Learning **Equation:** \[ Q(s, a) \leftarrow Q(s, a) + \alpha \left[ r + \gamma \max_{a'} Q(s', a') - Q(s, a) \right] \] **Pseudocode:** ```python # Q-Learning initialize Q(s, a) arbitrarily for each episode: s = initial_state while not terminal: a =