What's the difference between AI, Machine Learning and Deep Learning?
AI has been around for over 60 years. It has been a bumpy road, with inflated hypes and lost hopes along the way. There are several reasons for why it has taken off right now. It is only now we have the data, the computing power and tools needed to really get it to work.
And then we have Deep Learning. Deep Learning is the rocket fuel of AI.
There are many different concepts in the world of AI. You hear about Computer Science, Data Science, Machine Learning, Deep Learning, Supervised Learning, Unsupervised Learning, Reinforcement Learning, Convolutional Neural Networks, Recurrent Neural Networks. The list just goes on and on. It’s easy to feel lost, at least if you aren’t and expert yourself.
I thought it could be good to break down some of the basic concepts and see how these relate.
First of all, let’s look at the difference between Artificial Intelligence and Machine Learning.
Artificial Intelligence is a branch of Computer Science, and Machine Learning is a subfield of Artificial Intelligence. Machine Learning uses algorithms that allow computers to learn from examples without being explicitly programmed. Within the full concept of AI there is room for other kind of algorithms as well, programmed in more traditional ways. In the early days of AI, you programmed rules into the system. That was cumbersome and inefficient, and one of the reasons for why AI didn’t take off. When Machine Learning entered the stage in the 1980:ies, it helped boosting AI into a second gold age.
In the field of Machine Learning, Deep Learning is what everyone is talking about. The theories behind Deep Learning have been around for a long time, but it’s only during the last 10 years or so that Deep Learning has come into full use. Deep Learning algorithms need massive amounts of data and powerful computers, and that is something we didn’t have before.
Deep Learning uses deep Artificial Neural Networks as models and does not require feature engineering. Feature engineering is when you use the knowledge you have around a particular question and make sure the system gets this. It is of course good to give the system as much information as possible, but apart from that it takes time to figure out what to tell the computer in a way it understands, there are still many situations where you don’t have all information. To be honest, most situations. Like when it comes to driving. You don’t know when someone will hit the brakes, or if anything or anyone will just jump out in front of you. So, to get a self-driving car to work it has to be able to act on unforeseeable situations. That is what Deep Learning can do, and one of the reasons for why it’s so hot.
In Deep Learning you allow the data to flow through a mathematical network, from one side to the other. Deep Learning is inspired by how the brain works, which is why it is called an Artificial Neural Network. There are many different algorithms in the Deep Learning family, like for example Convolutional Neural Networks and Recurrent Neural Networks. The algorithms work in different ways, so depending on what kind of problem you want to solve, you will have to choose the best algorithm to do the job.
We also have the training algorithms. All AI systems have to be trained on data that are relevant to the task they are supposed to handle. If you for example want a system to spot dogs in a crowd, you first have to feed it with a lot of images where you have labeled the ones representing dogs. That is called supervised training, and it works a bit like a teacher in elementary school. You already know the answer, you just have to make sure to train the student until she can perform the task well enough on her own. Supervised learning is really good if you want to classify stuff or make predictions about the future.
Unsupervised learning works in a different way. Here you don’t know the answer and instead you let the system try to find patterns or structures in the data by itself. Unsupervised learning is good if you want to find hidden structures, like correlations between different factors you didn’t know of. In other words, you can gain new insights.
And then finally, there is reinforcement learning. In reinforcement learning the system learns how to react to the world around it. You give the system an end goal, and then the system looks for the best actions to take in order to achieve the goal or maximize the reward. In a way reinforcement learning resembles how humans learn. In most situations we learn by experience. We observe, test, reflect and then try again until we reach our end goal.
So, in summary, depending on what you would like to accomplish, there are different algorithms that will do the job. If you want to do a customer segmentation you choose unsupervised learning to create a model that works for you. And if you want to have autonomous and adaptive systems, like self-driving cars, you look for something in the Deep Learning family and train the system using reinforcement learning.
As you can see, AI is really powerful. And in the wake of what the technology can do, calls are now being made around ethics, security and privacy in relation to AI. Let’s have a look at that in the next episode.
If you want to know more about AI and can spare 5 minutes a day, then you should watch this 7-days series I've created on what AI is, how it can be used and how to get started.