CSP323L/CSP323A: Intro to Artificial Intelligence

• Intelligent Systems covered in CSP323L & CSP323A

Artificial Intelligence (AI)

• AI simulates human intelligence in machines programmed to think and act like humans
• AI involves creating algorithms and computer programs for tasks requiring human intelligence
• Examples of tasks include visual perception, speech recognition, decision-making, and language translation

Definitions of AI

• Rich and Knight define AI (1991) as making computers do things which, at the moment, humans do better
• Winston defines AI (1992) as studying computations that enable perception, reasoning, and action
• Charniak and McDermott define AI (1985) as studying mental faculties using computational models

Types of Artificial Intelligence

• AI is broadly classified based on capabilities and functionalities

Types of AI Based on Capabilities

• Narrow AI (Weak AI) is designed for specific tasks
• General AI (Strong AI) can understand, learn, and apply knowledge across tasks
• Superintelligent AI surpasses human intelligence in all aspects

Types of AI Based on Functionalities

• Reactive AI reacts to immediate conditions
• Limited Memory AI uses past experiences to inform future decisions
• Theory of Mind AI understands and predicts the behaviors of others
• Self-Aware AI has consciousness and self-awareness

Narrow AI (Weak AI)

• Narrow AI refers to AI systems designed to perform specific tasks within a limited context
• Narrow AI systems are highly specialized and excel at their programmed functions
• Narrow AI is task-specific
• Narrow AI provides no generalization
• Narrow AI provides human-like performance

General AI (Strong AI)

• General AI refers to AI systems that understand, learn, and apply knowledge across a wide range of tasks, akin to human cognitive abilities
• Unlike Narrow AI, General AI would have the capacity to perform any intellectual task and apply knowledge to different contexts
• General AI has broad intelligence
• General AI provides human-like reasoning
• General AI is self-learning

Superintelligent AI

• Superintelligent AI represents the most advanced AI form, surpassing human intelligence in creativity, problem-solving, and emotional intelligence
• This AI type would outperform the brightest human minds in any field
• Superintelligent AI surpasses human intelligence
• Superintelligent AI offers the ability of autonomous decision-making
• Superintelligent AI presents ethical and existential concerns

Types of Agents in AI

• Agents are entities that perceive their environment and act to achieve specific goals
• Agents have diverse behaviors ranging from simple reactive responses to sophisticated decision-making

Simple Reflex Agent

• Simple Reflex Agents act on current perception without considering past states
• Simple Reflex Agents follows predefined rules or condition-action pairs
• Simple Reflex Agents works in fully observable environments but fails in complex or partially observable situations

Model-Based Reflex Agent

• Model-Based Reflex Agents maintain an internal world model to handle partially observable environments
• Model-Based Reflex Agents track changes and making informed decisions based on past states

Goal-Based Agents

• Goal-Based Agents make decisions by considering goals
• Goal-Based Agents select actions to achieve those goals rather than only reacting to immediate conditions

Utility-Based Agents

• Utility-Based Agents choose actions based on a utility function that ranks different outcomes
• Utility-Based Agents ensures the best possible result rather than just achieving a goal

Learning Agents

• Learning Agents improve performance over time via learning from past experiences
• Learning Agents adjust behavior and are adaptable to new or changing environments

Rational Agents

• Rational Agents always select the most optimal action based on available information
• Rational Agents maximize performance while considering uncertainties

Reflex Agents with State

• Reflex Agents with State extends simple reflex agents by storing past information
• Reflex Agents with State react appropriately even when current perception is insufficient

Learning Agents with a Model

• Learning Agents with a Model uses an internal model of the world to predict outcomes
• Learning Agents with a Model refine their learning process enhancing decision-making

Hierarchical Agents

• Hierarchical Agents break down decision-making into multiple layers
• Hierarchical Agents makes complex tasks more manageable and efficient by handling different aspects separately

Multi-Agent Systems

• Multi-Agent Systems consist of multiple interacting agents that can collaborate, compete, or communicate
• Multi-Agent Systems assist achieve individual or collective goals in dynamic environments

Problem Solving in AI

• Problem-solving is a fundamental AI aspect, involving the design and application of algorithms to solve problems
• AI systems use various techniques to find solutions efficiently

Search Algorithms in AI

• Search algorithms navigate through problem spaces to find solutions
• Search algorithms are categorized into uninformed and informed searches

Uninformed Search Algorithms

• Uninformed Search Algorithm explores the search space without domain-specific knowledge beyond the problem's definition
• These algorithms do not use additional information like heuristics

Breadth-First Search (BFS)

• BFS explores all neighbor nodes at the current depth before moving to the next depth level

Depth-First Search (DFS)

• DFS explores as far as possible along each branch before backtracking

Uniform Cost Search (UCS)

• UCS is similar to BFS, but considers the cost of reaching nodes
• UCS expands the least-cost path first

Depth Limited Search (DLS)

• DLS is a DFS with a depth limit to prevent infinite recursion

Iterative Deepening Depth-First Search (IDDFS)

• IDDFS repeatedly applies DFS with increasing depth limits until a solution is found

Bidirectional search

• Instead of searching from the start node to the goal, bidirectional search runs two simultaneous searches
• One search proceeds forward from starting node
• The second search backtracks from the goal node

Informed Search Algorithm

• Informed Search Algorithm uses additional information (heuristics) for decision-making
• Informed Search Algorithm efficiently finds solutions by guiding the search process towards promising paths

Greedy Best-First Search (GBFS)

• GBFS finds the shortest path between two points or solving problems with multiple possible solutions

A* Search Algorithm

• The A* (A-star) algorithm is a powerful search method used in computer science
• A star finds the most efficient path between nodes in a graph

Simplified Memory-Bounded A* (SMA*)

• SMA* solves the problem of high memory usage by limiting stored nodes while still finding an optimal path
• If memory is full, SMA* removes the least promising node, remembers its cost, and re-expands if needed

Large Language Model (LLM)

• LLM is an AI model trained on textual data to understand, generate, and process human language
• LLMs use deep learning and transformer architectures (GPT, BERT, LLaMA) to perform natural language processing (NLP) tasks
• NLP tasks include text generation, translation, summarization, and answering questions
• In the context of Problem Solving in AI, LLM refers to an artificial intelligence model trained on vast amounts of textual data to understand and generate human-like text
• This models, like GPT-4, LLaMA, and Claude, leverage deep learning techniques, particularly transformers, to process and solve problems across various domains

Role of LLMs in Al Problem Solving

• LLMs contribute significantly to AI problem solving through natural language understanding, knowledge retrieval, automating decisions, assisting with code, and optimization/planning

Natural Language Understanding (NLU)

• LLMs can analyze and interpret human language, helping in tasks like text summarization, question-answering, and dialogue systems

Knowledge Retrieval & Reasoning

• LLMs recall vast amounts of information and apply reasoning skills to provide solutions

Automating Decision-Making

• By processing complex data, LLMs assist in making informed recommendations

Code Generation & Debugging

• LLMs help programmers solve coding problems by generating and optimizing code

Optimization & Planning

• LLMs are used in AI systems to generate step-by-step solutions for problems like scheduling, logistics, and game strategies

Challenges of Using LLMs for Problem-Solving

• LLMs lack true reasoning as they do not "think" as humans do
• LLMs sometimes generate false or misleading information (hallucinations)
• Running large models requires significant hardware resources (computational cost)

Ollama

• Ollama is an open-source framework for running and interacting with large language models (LLMs) locally
• Ollama is designed to provide an easy way to load, fine-tune, and serve LLMs without cloud-based infrastructure

Key Features of Ollama

• Ollama has local execution; it Runs models entirely on local hardware
• Ollama has model management; it Allows easy downloading, switching, and updating of LLMs
• Ollama supports custom model fine-tuning for specific tasks
• Ollama is has a focus on privacy; No data is sent to external servers
• Ollama has lightweight APIs; It Provides a simple API for integration into various applications