Artificial intelligence

Artificial intelligence, or AI, is the capability of computational systems to perform complex tasks.

These tasks are typically associated with human intelligence. Examples include learning, reasoning, problem-solving, perception, and decision-making. AI is a massive field of research. It combines engineering, mathematics, and computer science. Researchers develop software that allows machines to perceive their environment. These machines use learning to take actions that maximize their chances of achieving specific goals. AI is no longer just a theoretical concept. It is a fundamental part of modern technology.

To understand how AI works, we must look at its different subproblems. One major area is reasoning and problem-solving. Early researchers built algorithms to imitate step-by-step human reasoning. These were used for puzzles and logical deductions. By the late 1980s, scientists developed ways to handle uncertain information. They used concepts from probability and economics to solve these problems. Another core area is knowledge representation. This involves creating a knowledge base, which is a body of information a program can use.

Researchers use an ontology to represent knowledge. An ontology is a set of concepts and relationships within a specific domain. This allows AI to answer questions and make deductions about the real world.

Another vital component is planning and decision-making. In this context, an "agent" is any entity that perceives and acts in the world. A rational agent has specific goals or preferences. In automated planning, the agent works toward a specific goal. In automated decision-making, the agent manages preferences. It assigns a number, called "utility," to different situations. The agent calculates the expected utility for every possible action. It then chooses the action with the highest expected utility. In the real world, things are often not deterministic. This means the agent cannot be certain about the outcome of an action. It must make probabilistic guesses and then reassess the situation.

Machine learning is perhaps the most famous way to achieve intelligence.

It is the study of programs that improve their performance automatically. There are three main types of machine learning. Supervised learning uses training data that is labeled with the correct answers. This includes classification and regression. Unsupervised learning analyzes data to find patterns without any guidance. Reinforcement learning involves an agent that is rewarded for good responses and punished for bad ones. Finally, there is deep learning. This uses artificial neural networks to process inputs. These networks are biologically inspired by the human brain.

Natural language processing, or NLP, is another specialized field. It allows programs to read, write, and communicate in human languages. NLP includes tasks like speech recognition and machine translation. Early work in NLP struggled with word-sense disambiguation. This was due to the difficulty of teaching machines common sense. Modern NLP uses deep learning and the transformer architecture. This architecture uses an attention mechanism to process language. In 2019, generative pre-trained transformer (GPT) models began to produce coherent text. By 2023, these models achieved human-level scores on tests like the SAT and the bar exam.

The history of AI is a series of cycles. It was founded as an academic discipline in 1956. The field has experienced many periods of great optimism. However, it also faced periods of disappointment and lost funding. These low points are known as "AI winters." Interest and funding increased significantly after 2012. This was because graphics processing units helped accelerate neural networks. Deep learning began to outperform older techniques during this time. The arrival of the transformer architecture in 2017 accelerated this growth even further. In the 2020s, we entered an AI boom driven by generative AI.

Today, AI is used in many high-profile ways. You see it in advanced web search engines like Google Search. Recommendation systems on YouTube, Amazon, and Netflix use AI. Virtual assistants like Siri and Alexa are common examples. Autonomous vehicles, such as Waymo, use AI to navigate. Some companies, like OpenAI, Google DeepMind, and Meta, seek to create artificial general intelligence (AGI). AGI would be an AI capable of completing virtually any cognitive task as well as a human.

Because of this power, researchers also focus on AI safety. They discuss ethics, unintended harms, and potential existential risks. This has led to global discussions about AI regulation.