What is artificial intelligence really, and how does it work?
If you are interested in AI, you’ll undoubtedly know that a lot of the concepts are a bit overwhelming. There are plenty of terminologies to understand, such as machine learning, deep learning, neural networks, algorithms, and much more.
With the world of AI continually evolving, it’s good to go over some of the basic concepts to better understand how it’s changing.
In this episode of Short and Sweet AI, I address some of the questions that I get asked a lot: what is AI? How does AI work? I also delve into some of the limitations of AI and the possible solutions.
Listen to the episode below or keep reading to learn more.
How does AI work?
Artificial intelligence works by using computers programmed with algorithms. Algorithms are step-by-step instructions that tell a computer what to do and how to solve a problem. Think of it like a recipe. A recipe has specific steps you need to follow in a particular sequence to bake a cake, for example.
Computer scientists write algorithms using a programming language that the computer understands. There are a few different programming languages, which typically have strange names like Python or C++.
These computers can also perform math calculations or computations to analyze information to give an answer. This is called computational analysis.
Together, the programming language and math calculations make computer software. By using this software, the algorithms can come up with an answer from data sets fed into the computer.
Machine Learning is a type of AI
The main type of AI used today is something called machine learning. Machine learning is carried out by artificial neural networks, also known as “neural nets,” for short.
Neural networks underpin the most advanced artificial intelligence used today. They get their name because they are, in part, based on the way neurons in the brain function.
In the brain, neurons receive information (inputs) and process it to give a result (outputs).
Artificial intelligence uses digital models of brain neurons based on the computer binary code of ones and zeroes. The digital neurons process information and pass it along to other higher layers of processing – just like within the brain. By “higher levels,” it means that the results become more specific.
Deep Learning is a type of Machine Learning
Before a computer can give us all the answers, it must be trained using large amounts of data. As the computer processes more and more data, it starts to learn from it. This is known as training the machine.
After some time, when you give the computer completely new data, it knows what to do with it and can give you a correct answer to a specific question. In a way, the machine learns from what you have already fed it.
If you have many, many layers of neural networks, each processing and passing on information to another layer, this is called a deep neural network. When machines learn from deep neural networks, we call this deep learning.
Present-day AI has limitations
AI has been truly revolutionary and has surpassed expectations in many ways. However, it’s not without its limitations.
All of the software and computer calculations used in machine learning, especially deep learning, require absurd amounts of data and computer power. Neural nets can be hundreds of data layers deep with billions on parameters, like AlphaFold or GPT-3, which I’ve discussed in previous episodes.
These systems are gargantuan machine learning algorithms that require supercomputers to run them. As a result, this limits who can use deep learning. Only large tech companies and corporations have the resources to harness this type of deep learning.
The next major limitation to know about is that AI systems are limited by the training they receive. As mighty as the neural nets may appear, they are very narrow at a core level.
AI is designed to recognize specific things and act accordingly, such as identifying an image, steer a car left or right, or translating a language. As they are trained to do specific things, they struggle when something deviates from its training. It often acts brittle and breaks when it is presented with something it wasn’t programmed to do.
New AI neuroplasticity
While AI has its limitations, there is something new that adds an interesting element to artificial intelligence. AI has reached the point where it’s becoming less artificial and more biological.
Like a human brain, AI has started to develop “neuroplasticity.” This means that AI could indeed start to adapt and learn from change, just like a brain does.
Now that we’ve covered the basics of artificial intelligence, I will be discussing something called “liquid AI” in my next episode. This form of AI could solve a lot of these limitations through a type of neuroplasticity. Watch out for my next episode to learn more.