Artificial intelligence:
Artificial intelligence (AI) is the display of intelligence by machines especially computers. It’s a branch of computer science that crafts methods and software to enable machines to sense their surroundings and use learning and intelligence to act in ways that increase their success in achieving specific objectives.
AI is prevalent across various sectors, including industry, government and science. Notable uses are in sophisticated search engines like Google Search, recommendation systems like those on YouTube, Amazon and Netflix voice interactions through Google Assistant, Siri and Alexa, self-driving cars like Waymo creative tools like ChatGPT and in mastering strategic games like chess and Go. Often, AI blends into everyday applications so seamlessly that it’s no longer recognized as AI.
Machine intelligence:
The term “machine intelligence” was first researched by Alan Turing. AI became an official field of study in 1956 and has since experienced cycles of high hopes and setbacks, including periods of reduced funding known as “AI winters.” Interest surged post-2012 with breakthroughs in deep learning and again after 2017 with the development of transformer architecture leading to the AI boom of the early 2020s primarily driven by the United States.
AI’s rise in the 21st century is steering a shift towards more automation data-centric decision-making and embedding AI systems in various sectors affecting jobs healthcare, governance, industry and education. This prompts discussions on the long-term impact, ethical considerations, and risks of AI, leading to calls for regulatory measures to ensure its safety and benefits.
AI research focuses on specific goals and employs particular tools. Traditional objectives include reasoning, knowledge representation, planning, learning, natural language processing, perception, and aiding robotics. A major long-term goal is general intelligence which is the capability to perform any human task at least as well as a human.
To achieve these goals AI researchers utilize a variety of techniques such as search and mathematical optimization, formal logic, artificial neural networks and approaches based on statistics operations research and economics. AI also incorporates insights from psychology, linguistics, philosophy, neuroscience, among other disciplines.
Reasoning and problem-solving:
AI algorithms mimicked human step-by-step reasoning used in puzzle-solving and logical deductions. methods to handle uncertain or incomplete information were developed using probability and economics.
These algorithms often struggle with large reasoning problems due to a “combinatorial explosion,” where they slow down exponentially as problems grow. Humans typically rely on quick intuitive judgments rather than step-by-step deduction making accurate and efficient reasoning a challenge for AI.
Knowledge representation:
This involves encoding knowledge in a way that AI programs can make intelligent deductions and answer questions about the real world. It’s used in various applications including content indexing, scene interpretation, clinical decision support and knowledge discovery.
A knowledge base contains information usable by a program, while an ontology defines the objects, relations, concepts and properties within a knowledge domain. Representing common knowledge and its often non-verbal nature poses significant challenges as does acquiring knowledge for AI use.
Planning and decision-making:
An “agent” in AI is anything that perceives and acts in the world. Rational agents have goals and take actions to achieve them. In automated planning, agents aim for a specific goal, while in decision-making, they have preferences and choose actions based on expected utility.
Classical planning assumes agents know the outcomes of their actions but real-world scenarios often involve uncertainty. Agents may need to make probabilistic guesses and reassess situations after acting. Preferences can be learned or refined and agents must navigate vast possibilities under uncertainty.
A Markov decision process models action outcomes and rewards, with policies guiding decisions for each state. These can be calculated, heuristic, or learned.
Game theory: This studies rational behavior among multiple interacting agents and informs AI decision-making involving other agents.
