This article is the first in our series on AI Literacy, where we explain basic AI concepts so that everyone can understand what AI is and how it works.
If you are new to AI, there are three things you should know to start off with:
What is AI?
Artificial Intelligence (or AI) is the umbrella term used for a large group of technologies that have one thing in common - their goal is to help computers do things that come easily to the human brain. Tasks like recognizing images, making plans or strategies, forecasting the future, detecting unusual or anomalous situations, all fall under the umbrella of AI.
Within AI there are many different types. You may hear about Machine Learning, which is a subset of AI and is a set of algorithms (program approaches) that help machines learn from data. Within Machine Learning there is also a very popular subset called Deep Learning (also called Neural Networks or Deep Neural Networks). These focus on a type of AI that has a superficial similarity to how the human brain works. In recent years, Deep Learning has been shown to be very effective with rich data like images, sound, video, etc.
Learning and Prediction
The next important thing to know is that all AIs need to learn. They learn in a variety of different ways, depending on the type of AI. Some learn from historical data, some learn by observation, and some learn by experimenting with their environment. However, virtually all AIs need to learn.
This is where data comes in. The more data the AI has access to, the better it learns. One reason why AIs have become so popular in recent years is that the internet, smart devices, and connectivity have made a very large amount of data available for AIs to learn from.
Once an AI has learned (a process called training), it stores its learnings in a structure called a model. This model can then be used to answer new questions, via a process known as prediction or inference.
How AIs are built in real life
The figure below shows the AI lifecycle that most real-life AIs follow. It starts with the problem to be solved. Then the data scientists or engineers gather up data about the problem and experiment with different types of AIs to find one that works well. Once they have one, they deploy it (which means to bring it to life outside the lab) and connect it to their application. Once this is done, the AI can be used to solve the problem. Then the team monitors the AI and repeats the cycle as often as necessary to improve the AI.
In real-world companies, this cycle can take as long as a year if done manually or can happen as often as every 15 minutes if fully automated. How frequently a company runs an improvement cycle depends on many factors - how quickly they need the AI to adapt to new information, how expensive it is to run the cycle, whether they need to get legal approval for updating the AI, and so on.
Hope this has helped you get an insight into the world of AI. If you would like to learn more, please subscribe to our AI literacy newsletter here.