However, the idea of AI is not a 21st-century one. Highly intelligent and visionary men and women have been thinking about it since at least the 1950s. The core idea started at the same time computers were first being developed to aid human beings with routine tasks like business calculations or spreadsheet preparation. AI pioneers (such as Alan Turing and John McCarthy) started envisioning a computer system (combination of hardware and software) that does not just carry out repetitive instruction but can develop a capability of thinking for itself. They thought, that if such a machine can be developed, then it would be a faithful companion to human beings for all kinds of intellectual tasks – helping us in discovering new scientific facts, designing better products and services, and solving great challenges that the world faces, making everybody’s lives a lot better. In that sense, AI is supposed to be a human being’s ‘intelligent companion’ who may look and feel different (a computer program or a robot) but think, converse, and perceive the world just like us and provide us all the help to make it better. There has been a lot of research and development in the AI field since those early days. Today, the largest organizations (both private and Governmental) around the world are investing huge amounts of money and human resources in this field. Some of the brightest minds in the academia and industry are working on solving tough problems related to AI.

However, the current focus is on something called ‘Narrow AI’ that does not aim to become the general-purpose intelligent companion of human beings that we talked about but focuses on solving a narrow, specific task at hand. This ‘Narrow AI’ is a specialized tool, and there will be many such AI tools for all kinds of tasks, such as making a movie recommendation, playing video games, helping with automatic driving, helping industrial operations in factories, assisting doctors with healthcare delivery, even discovering new medicines and vaccines.

In contrast, Artificial General Intelligence (AGI) is supposed to have the true thinking and analysis faculties of a human mind, sufficient for handling all general-purpose everyday tasks. Such an AGI system may not be the master of a single task but can handle any task at least at the level that a human can do. Naturally, we also expect an AGI system to understand the world around us - through vision, audio, and language communication - as well as a human being is expected to. Unfortunately, there is no such AGI system in sight today. Understanding visual clues and natural languages (e.g., English, not some computer programming language like Python) has turned out to be the toughest task for AI tools to handle. A lot of progress is being made and researchers are hoping to get to some level of AGI within this century.
Modern AI systems mostly work by learning from examples. Careful programming allows humans to feed a huge amount of data into these systems autonomously. Generally, this data is in the form of scenarios and their responses. AI programs analyze that data, find the hidden patterns, and construct a mathematical model of how that data could have been generated and also what response goes with what scenario. Next time, a new data point or scenario is presented to the AI, the model can predict the correct response without explicit programming or instruction. In that sense, it becomes a prediction machine and helps humans solve many challenging problems.

For example, given medical test results, an AI system can predict what probable ailments a patient may have, thereby aiding a doctor. Or, it can predict what movie you may like given the preference data about your friends and what movies they have liked in the past. Please note, however, that the AI tool that makes the movie recommendation is completely different (in inner workings and programming) than the one that helps a doctor assess medical ailments. That’s why they all are individual, Narrow-AI systems.