Artificial Intelligence Cheatsheet

Cheatsheet Content

What is AI? Definition: Artificial intelligence (AI) is the intelligence exhibited by machines or software. It's an academic field of study focused on creating computers and software capable of intelligent behavior. Core Idea: Making computers behave like humans, including areas like: Games playing Expert Systems Natural Languages Neural Networks Robotics Ability: For a computer to think, learn, and simulate human mental processes (perceiving, reasoning, learning). Distinction from Machine Learning: Machine Learning is a subset of AI. AI signifies general ability to mimic human thought, while ML implies technologies and algorithms that allow systems to recognize patterns, make decisions, and improve through experience and data. Academic Disciplines of AI (Foundations) Discipline Contribution to AI Philosophy Logic, methods of reasoning, mind as a physical system, foundations of learning, language, rationality. Mathematics Formal representation and proof, algorithms, computation, (un)decidability, (in)tractability. Probability/Statistics Modeling uncertainty, learning from data. Economics Utility, decision theory, rational economic agents. Neuroscience Neurons as information processing units. Psychology/Cognitive Science How people behave, perceive, process cognitive information, represent knowledge. Computer Engineering Building fast computers. Control Theory Design systems that maximize an objective function over time. Linguistics Knowledge representation, grammars. Types of AI by Capability Narrow AI (Weak AI): Designed for specific, limited tasks. Cannot perform tasks outside its defined scope. Examples: Voice assistants (Siri, Alexa), facial recognition, recommendation engines (Netflix, Amazon). General AI (Strong AI / Artificial General Intelligence - AGI): Hypothetical AI with human-level intelligence. Ability to learn, understand, and apply knowledge across a wide range of tasks, similar to a human. Examples (theoretical): Robots that learn new skills and adapt to unforeseen challenges, AI systems diagnosing complex medical issues across specializations. Superintelligence (Super AI / Artificial Superintelligence - ASI): Theoretical concept representing AI capabilities far surpassing human intelligence. Includes creativity, problem-solving, and awareness beyond human levels. Types of AI by Function Reactive Machines: Most basic type, no memory, react only to current stimuli. Make decisions based on current data. Example: Deep Blue (chess-playing computer). Limited Memory AI: Can store past data and experiences to inform future decisions. Capable of complex tasks but vulnerable to outliers. Example: Self-driving cars. Theory of Mind AI (Speculative): Future AI capable of understanding emotions, beliefs, and thoughts of others. Does not currently exist. Self-Aware AI (Theoretical): Most advanced form, possesses consciousness, sentience, emotions, and desires. Primarily a concept within science fiction. AI Techniques 1. Machine Learning (ML) Building algorithms to learn patterns in data and make predictions. Uncovers hidden patterns in datasets, predicts new data without explicit programming. Types of Machine Learning: Supervised Learning: Learns from labeled data (input-output pairs). Predicts an output variable based on input variables. Example: Email spam filters (distinguishing spam from legitimate emails). Unsupervised Learning: Analyzes unlabeled data without predefined outcomes. Uncovers inherent structures or patterns. Example: Grouping customer behavior data to identify market segments. Semi-supervised Learning: Uses a small amount of labeled data and a large amount of unlabeled data. Combines aspects of supervised and unsupervised learning. Reinforcement Learning (RL): Learns through trial-and-error, receiving feedback (rewards/penalties). Aims to maximize rewards by making optimal decisions. Example: Training agents for complex tasks like grasping objects or navigation. 2. Natural Language Processing (NLP) Programming computers to process human languages for human-computer interaction. Leverages algorithms to recognize and abstract language rules, converting unstructured human language into a computer-understandable format. Applications: IVR systems, call center applications, language translation (Google Translate), grammar checking (Microsoft Word), virtual assistants, chatbots. Common Variants of NLP: Lexical Analysis (Tokenization): Converts sequence of characters into tokens (words, units). Syntactic Analysis (Parsing): Analyzes string of symbols conforming to formal grammar rules (grammatical structure). Semantic Analysis: Attempts to understand the meaning of human language, considering context, logical structuring. Lexical Semantic Analysis Compositional Semantics Analysis Pragmatic Analysis: Extracts information from text, focusing on actual meaning and context, including social content. Discourse Integration: Uncovers underlying meaning of spoken or written text by considering social and historical contexts. 3. Computer Vision Equips machines with the ability to interpret visual information. Revolutionized industries like healthcare, automotive, robotics. Enables tasks such as facial recognition, object detection, autonomous driving. Key Concepts: Sensitivity: AI's ability to pick out small details in visual information. High sensitivity detects fine details. Resolution: Level of detail a computer vision system can capture and process. High-resolution images are vital for correct identification. Applications: Screen Reader Code & Character Reader (OCR) Computer Vision + Robotics for Bin Picking Defect Detection Metrology Intruder Detection Assembly Verification 4. Robotics & Automation Enables machines to perform boring, repetitive jobs, increasing productivity. Employs machine learning, artificial neural networks, and graphs. Applications: Fraud prevention during online payments (CAPTCHA), high-volume repetitive tasks (Robotic Process Automation). AI in Robotics: AI helps create intelligent robots that can perform tasks based on experience, not just pre-programming (e.g., Humanoid Robots like Erica and Sophia). 5. Deep Learning (DL) Branch of machine learning based on artificial neural network (ANN) architecture. ANNs use layers of interconnected nodes (neurons) to process and learn from input data. In a fully connected Deep Neural Network, input and hidden layers transform data through non-linear transformations to learn complex representations. Main Applications: Computer Vision: Identify and understand visual data (object identification, locating objects in images/videos). Natural Language Processing: Understand and generate human language (essay generation, language translation, sentiment analysis). Reinforcement Learning: Train agents to take actions in an environment to maximize rewards (robot grasping, navigation, manipulation). 6. Data Mining Process of extracting knowledge or insights from large amounts of structured, semi-structured, or unstructured data. Uses statistical and computational techniques. Primary Goal: Discover hidden patterns and relationships for informed decisions or predictions. Techniques: Clustering, classification, regression analysis, association rule mining, anomaly detection. Applications: Marketing (customer segments, target campaigns), healthcare (risk factors, personalized treatment), finance, telecommunications. Concerns: Ethical and privacy issues, especially with personal data. Requires safeguards to protect privacy and prevent misuse. Applications of AI Finance: Automation, chatbots, adaptive intelligence, algorithmic trading, machine learning (e.g., Mastercard's Decision Intelligence). Data Security: Enhance safety and security against cyber-attacks (e.g., AEG bot, AI2 Platform for bug detection). Travel & Transport: Travel arrangements, hotel/flight suggestions, route optimization, AI-powered chatbots for customer interaction. Automotive Industry: Virtual assistants (Tesla Bot), self-driven cars. Entertainment: Content recommendations (Netflix, Amazon) using ML/AI algorithms. Agriculture: Soil and crop monitoring, predictive analysis for better yields. E-commerce: Competitive edge, product recommendations (size, color, brand). Education: Automated grading, AI chatbots as teaching assistants, personal virtual tutors. Social Media: Organizing and managing massive user data, identifying trends, hashtags. Robotic Vehicles: Autonomous driving (e.g., STANLEY in DARPA Grand Challenge). Speech Recognition: Automated systems for booking flights, managing dialogues. Autonomous Planning & Scheduling: On-board planning for spacecraft (NASA's Remote Agent). Game Playing: Defeating human champions (IBM's Deep Blue in chess). Spam Fighting: Classifying spam messages, adapting to new tactics. Logistics Planning: Planning and scheduling for large-scale operations (e.g., 1991 Gulf War logistics). Machine Translation: Automatic translation between languages (e.g., Arabic-to-English). Benefits of AI Reduction in Human Error: AI systems can reduce errors in tasks like weather forecasting. Takes Risks Instead of Humans: AI can perform dangerous tasks in hazardous environments (e.g., Chernobyl disaster scenario). Available 24x7: AI systems can operate continuously, handling queries and issues (e.g., educational institutes, helpline centers). Helping in Repetitive Jobs: Automates mundane tasks, increasing efficiency (e.g., document verification in banks). Digital Assistance: Provides customer support via voice bots or chatbots. Faster Decisions: AI algorithms can make optimal decisions rapidly (e.g., AI in chess games). Daily Applications: Integrated into everyday tools (Siri, Cortana, Google Assistant) for various tasks. New Inventions: Powers innovations across domains to solve complex problems. Risks of AI High Costs of Creation: Requires significant investment in hardware, software, maintenance, and complex development. Making Humans Lazy: Automation can lead to over-reliance and decreased human engagement in tasks. Unemployment: Automation of repetitive tasks by AI and robots can lead to job displacement. Lacking Out of Box Thinking: Machines are limited to programmed tasks; they lack creativity and adaptability outside their defined scope. No Emotions: AI lacks human emotions and connection, which can be a drawback in roles requiring empathy or interpersonal skills. Ethical Issues in AI Bias and Fairness: Problem: AI systems can perpetuate or amplify existing biases from training data, leading to unfair decisions (e.g., hiring, loans, criminal justice). Impact: Discrimination against certain groups (race, gender, age). Solution: Use diverse/balanced data, apply fairness techniques, regularly check system results. Privacy and Data Security: Problem: AI requires vast amounts of personal data, leading to privacy concerns. Impact: Misuse of personal data, violation of privacy rights. Solution: Strong security measures, transparency in data use, obtaining user consent. Accountability and Transparency: Problem: Complex AI systems can be "black boxes," making their decision-making opaque. Impact: Difficulty holding AI responsible for errors, especially in critical fields (healthcare, law enforcement). Solution: Make AI systems explainable, create rules for accountability. Job Displacement: Problem: AI automation can take over many jobs (manufacturing, transportation, customer service). Impact: Job losses, increased economic gaps, social problems. Solution: Training programs for new skills, considering universal basic income. Autonomy and Control: Problem: Concern about losing control as AI systems become more independent (e.g., self-operating weapons). Impact: AI making major decisions without human oversight can lead to serious problems. Solution: Strict limits on AI control, ethical rules, human involvement in critical decisions. AI in Warfare: Problem: Ethical concerns with AI in military, especially self-operating weapons making targeting decisions without human control. Impact: Lack of responsibility, unethical uses, violations of international laws. Solution: International agreements on rules, adherence to humanitarian laws. Long-Term Impact and Existential Risk: Problem: Fear that superintelligent AI could become too powerful, acting against human values. Impact: Threats to human safety, freedom, or survival if AI development is uncontrolled. Solution: Researching safe AI, ensuring alignment with human goals, regulation. Manipulation and Misinformation: Problem: AI can create fake content (deepfakes), spreading misinformation and scams. Impact: Erodes trust in media, politics, public discourse. Solution: AI for detection, rules to control misuse, public education on spotting misinformation. Ethics of AI in Healthcare: Problem: Questions about patient consent, trustworthiness of AI decisions, potential replacement of doctors. Impact: Concerns about privacy, patient rights, reliability, loss of trust. Solution: Strict rules for AI in healthcare, human involvement, patient consent. AI in Society AI's integration into society profoundly impacts various sectors: Workforce Transformation: Impact: Automates tasks, reshaping industries. Benefits: Increased efficiency, productivity, enhanced creativity. Challenges: Job displacement, need for new skills, potential for widening inequality. Healthcare: Impact: Advancements in diagnostics, personalized medicine, drug discovery. Benefits: Improved outcomes, faster diagnostics, efficient resource use, remote monitoring. Challenges: Data privacy, over-reliance on AI, responsibility for AI decisions. Education: Impact: Personalized learning, intelligent tutoring, automated grading. Benefits: Greater access to education, assistance for teachers. Challenges: Equity issues, depersonalization of education, algorithmic bias. Social Interaction and Communication: Impact: AI-powered chatbots, virtual assistants, social media algorithms. Benefits: Enhanced customer service, personalized content, accessibility for people with disabilities. Challenges: Social isolation, reinforcement of filter bubbles, distorted views. Governance and Public Services: Impact: Improves delivery of public services, optimizes traffic, crime prediction, administrative processes. Benefits: More efficient government, better resource allocation, urban planning. Challenges: Privacy concerns, algorithmic bias, need for transparency. Ethics and Accountability: Impact: AI raises questions about ethics, accountability, fairness of algorithms. Benefits: Ethical frameworks guide responsible development. Challenges: Determining responsibility for AI mistakes, ensuring fairness/transparency, concentration of power. Economic and Social Inequality: Impact: Potential to bridge or deepen societal divides. Benefits: Democratize knowledge, improve productivity, support economic growth if implemented equitably. Challenges: Exacerbate inequalities if benefits are concentrated, leading to job losses and unequal opportunities. Privacy and Surveillance: Impact: AI enables vast surveillance capabilities (facial recognition, behavioral tracking). Benefits: Identify security threats, enhance public safety, prevent crime. Challenges: Balancing security and privacy, potential for misuse, discrimination, violations of rights. Environmental Sustainability: Impact: AI optimizes energy use, waste management, aids climate change research. Benefits: Reduced environmental footprint, efficient resource use, better climate response. Challenges: Significant energy consumption of large-scale AI models, environmental impact of training. AI in Media and Content Creation: Impact: Enables creation of realistic computer-generated images, music, writing, deepfakes. Benefits: More creative expression, democratizes content creation. Challenges: Misinformation, disinformation, ethical implications of AI-generated content, manipulation. Agentic AI Definition: Artificial intelligence systems designed to act as autonomous agents , capable of making decisions, taking actions, and pursuing goals without constant human intervention. Characteristics: Perceive their environment (via sensors or data input). Make decisions based on goals, constraints, and context. Take actions to influence their environment or achieve objectives. Adapt through learning from outcomes or changing conditions. Example: In advertising and marketing, Agentic AI decides who sees which ad and generative AI creates ad content based on user preferences. Features of Agentic AI: Autonomy: Can act independently. Goal-oriented behavior: Can set, pursue, and revise goals. Planning and reasoning: Capable of multi-step planning. Memory and statefulness: Retains information about past actions and current tasks. Interaction with environments: May manipulate digital or physical environments. Self-reflection (emerging): Can reflect on performance and adjust strategy. Turing Test Concept: Proposed by Alan Turing in 1950. A method to determine whether a machine can demonstrate human intelligence. How it Works: Three participants: a human interrogator, a human respondent, and a machine respondent. The interrogator communicates via text with both the human and the machine, isolated from them. The interrogator's goal is to identify which participant is the human and which is the machine. If the interrogator cannot reliably distinguish the machine from the human, the machine is said to have passed the Turing Test. Transparency in AI Systems Transparency in AI systems is crucial for several reasons: ethical, technical, legal, and practical. Trust and Accountability: Why it matters: Users and stakeholders need to trust AI, especially when it impacts lives (healthcare, finance, justice). How it helps: Shows how decisions are made, allowing accountability for errors or biases. Understanding and Explainability: Why it matters: Many AI systems (deep learning) are "black boxes," making their workings hard to understand. How it helps: Enables researchers, developers, and users to interpret model behavior, explain decisions, and detect problems. Bias Detection and Fairness: Why it matters: AI can replicate or amplify societal biases if unchecked. How it helps: Allows inspection of data sources, model training processes, and decision-making criteria to identify and mitigate unfair outcomes. Compliance with Laws and Ethics: Why it matters: Regulations (EU AI Act, GDPR) require explanations for automated decisions and data use. How it helps: Ensures AI systems comply with legal standards and ethical norms (privacy, consent, non-discrimination). Debugging and Improvement: Why it matters: Complex systems need ongoing improvement. How it helps: Makes it easier to identify why and how systems fail, facilitating effective fixes or retraining. Public and Stakeholder Engagement: Why it matters: AI affects large groups of people. How it helps: Encourages dialogue among developers, regulators, and the public to guide responsible development. Examples: Medical AI: A transparent system explains its diagnosis recommendation, building trust for doctors. Hiring AI: Transparency allows candidates and regulators to see if the system discriminates. Transparency is key to building trustworthy, fair, and human-aligned AI systems. Without it, systems risk being opaque, biased, unaccountable, and harmful.