Have you ever tried to get your feet wet into the world of AI (Artificial Intelligence)? If so, I’m sure you stumbled upond different concepts that seem to be talking about the same thing, although you are also under the impression that there are some differences between them. This stands true for the following concepts: Artificial Intelligence, Machine Learning and Deep Learning.
In this article, I’m going to give you some illustrations that will, I hope so, make things simpler.
Simply put, this is a set of techniques that gives some kind of intelligence to machines. By intelligence, we mean the ability to analyze and take decisions or make predictions based on a set of data that have been provided to the “system”. I’m talking about “system” since, most of the time, the “machine” is a set of hardware and software. Sometimes, the hardware comes with some sensors that capture datas. Think about those self-driving cars that come with a lot of sensors and data so that it can take you from point A to point B without human interaction.
Historically, the term “artificial intelligence” first appeared in 1956. Not that new isn’t it? In fact, a lot of research has already been done in this field. If you discuss with some mathematicians, they’ll surely fool you with a lot of models and techniques that all converge into Artificial Intelligence, or at least the use case that we are now aware of.
But why did it take so long before Artificial Intelligence got some popularity? One of the reasons is that the computing power for treating all those datas wasn’t available, and it was still so costly to use. Moreover, it was hard to gather a lot of data. Now, with the popularity of cloud computing as well as big data, the ingredients are there so as to play with this artificial intelligence. The exponentantial adoption of social media also generates a lot of data to compute. Moreover, even single laptops are now so powerful that they can deliver astounding calculation power. Just check those gamer laptops that are available right now and you can see how realistic the rendering are.
In our daily lives, we have seen a lot of application of Artificial Intelligence: from the youtube video recommendation, to the GPS your car is using to suggest to you the best route. With all those self-driving cars, as well as drone-delivery, there’s a lot of image-processing algorithms working behind that appeals to AI techniques. The use cases are endless.
Machine Learning
As Artificial Intelligence got more mature and got hold of more computing power as well as enormous data, researchers logically went into imitating human brain. Indeed, as human, we always learn new things: from baby steps to becoming olympic athletes. We also get better at some activities. The question was, what’s the mathematical model that allows “something” to have the ability to learn and get better at one activity. As humans, we call it experience. But what about machines?
With Machine Learning, the system gets better as it is fed with new data. All along the way, the system keeps track of thousands (if not millions) of properties (or variables). As the user adds more input, those properties got reevaluated. Then some algorithm tries to find a meaning, some correlations with all those datas, thereby identifying some patterns or trends. This allows the system to give some recommendations or predictions.
You have surely been faced with “recommendation” services. Think about your shopify or deezer music service recommendation. Have you seen how accurate it becomes as you keep on using the service. Unconsciously, we are using online services that have this machine learning feature embedded into it. Think about your social media feed algorithm, whether it is facebook, twitter, instagram or whatever social media platform you may be using. From whatever input you feed the platform (likes, retweet, comments, …), the algorithm is learning about your preferences, and then, in return, it shows you more content that is more relevant to you.
Getting back to our music recommendation system: in fact, the first time you create your account, the system doesn’t know anything about your music preferences. As you start following some artists, creating some playlist and listening to music, that system creates some kind of attributes of music you like: the tempo, the year the music aired, your friends’ music taste, the time you’re listening to your music, … In fact, that same system collects thousands of attributes about you and your tastes. Then comes the big data analysis (and learning). So the more you interact with the system, or reacts to the recommendation it makes, the more it is accurate.
Usually, Machine Learning have found its way to our daily lives as most apps and web-based applications are now leveraging the pozer of our smartphones and laptops to bring tailored experience to us. One of the biggest users of this Machine Learning technique is the online advertisement industry. But the same can also be applied to other business aspects: let’s think about the inventory process for example: based on different variables (market, trend, internal use, weather, market prices, …), a Machine-Learning based system can recommend the buyer on what items need to be restocked at some period of time, thereby helping for a better and optimized buying planning.
As we dig deeper into the techniques of Artificial Intelligence, we get into more conceptual approach. The deep learning approach is an exercise where we teach the AI some things we take for granted. For example: once we’re shown a square compared to a circle, instinctively, we know how to differentiate them. But it takes a lot of learning and exception handling so as to help an AI-powered system to differentiate them. Apart from removing the noises that may exist around the forms, we need to teach the system to identify angles. This is an explanation at its basic level, but this kind of processing is the same when it comes to differentiating a zebra from a cow.
One of the most used techniques in this field is the neural network. Simply put, it’s a technique that tries to refer to our brain’s neuronal network. Each neurons is “attached” to some properties/values. Those nurons are then interconnected and interact with each other, changing status depending on the input that external datas will feed them. Imagine a sea that will react differently depending on the different external stimuli: boats, wind, seism, high tide/low tide, whales, pollution, …
Usually, as opposed to Machine Learning, Deep Learning is more efficient when applied to large amount of datas. Moreover, Deep Learning usually referes to the Neuron Network. This usually means there are a certain number of layers that interact each other with each other, and that at the simplest form, a neuron has some attribute. The “interaction” between these neurons will help make informed decision. But clearly, the learning phase of Deep Learning is way longer compared to the “simpler” Machine Learning technique.
Topics that evolves around this AI topic easily gets very technical. In this article, I tried to explain the core fundamentals in simple terms, by using analogies. Most of the times, all the technicalities are handled by software editors, as a user (business user or simple user), you can consider integrating those solutions with the business-end in mind. Think about what you need to achieve, those AI solutions will be on eternal learning mode so as to deliver the best experience. But remember: the more you feed the system with data, the more accurate it can be. Feel free to share in the comments, or in the social network posts where this article is shared, the insights you have gotten from this article. Or even better, suggest some more topics you’d be interested in reading, wile using that same easy-to reand and understand format.
Growing, sharing in the digital era 