The terms Artificial Intelligence (AI), Machine Learning (ML) and Deep Learning (DL) have become very common these days. They are often used interchangeably, especially while dealing with Big Data, analytics, and searching and indexing. Although these 3 terms are very much related, they’re not the same thing.
In today’s story, we’ll explain you what exactly is AI, ML and DL, and how they are different from each other. Don’t worry, we won’t use any complicated scientific term – we’ll keep it short and as simple as possible.
AI is the computer science subdivision, first coined in 1956 by John McCarthy. The computer usually performs tasks that we command. However, AI is a way of building a machine, or software that can think intelligently and perform itself, just like humans. This includes things like understanding language, recognizing sounds and visuals, learning, planning and solving problems.
For instance, a conventional program can answer only specific questions, but AI program can answer the generic questions.
There are Four Types of AI –
- Completely Reactive: Basic type, can’t draw conclusions. Example – Google’ AlphaGo, IBM’s DeepBlue
- Limited Memory: Can make proper decisions and take actions. Example – Chatbots, Self-driving vehicles
- Theory of Mind: Can understand thoughts, emotions and interact socially – Not built yet*
- Self-Aware: Can form representations about themselves, aware of self state and can predict feeling of others – Not built yet*
*Although third and fourth type of machine don’t really exist, they are demonstrated in sci-fi movies – R2D2 (type-3) from Star Wars and Eva (type-4) from Ex Machina.
ML is a subset of AI that provides system the ability to learn, act and improve from experience without being explicitly programmed. So instead of explicitly writing all methods with specific instructions to achieve a particular task, ML is a technique of training a program so that it can learn from past experiences. Here, training refers to feeding a large amount of data to program and allowing the program to configure itself and improve.
For example, if you provide thousands of cat pictures to ML algorithm, it will start recognizing how a cat looks like – their height, color, face shape, etc. Eventually, it can identify and automatically tag cats in the pictures. Once the accuracy level is high enough, the algorithm has now learned what a cat looks like.
Types of Machine Learning:
- Supervised: Makes machine learn explicitly through data with defined output.
- Unsupervised: Machine understands data (pattern/structure) and draws inferences from datasets.
- Reinforcement: An approach to AI, learn from positive and negative reinforcement, and reward the positive results.
Deep Learning is a subfield of machine learning that deals with algorithms inspired by the structure and function of human brain, or interconnection of many neurons. These algorithms are known as Artificial Neural Networks (ANNs) that mimic the biological structure of the brain.
The neurons have discrete layers and connections to other neurons. One can visualize these layers as a nested hierarchy of related concepts or decision trees. Each layer is capable of selecting particular feature to learn or follow a specific path. Depth is built by multiple layers – the more layer a network has, the deeper/complex it is.
In order to be well-trained, deep learning networks require large quantities of items. Rather than writing code for each edge that defines items, the system learns from exposure to millions of data points.
Google brain is a perfect example of deep learning to recognize cats after taking over ten million image-samples. These networks don’t need to be coded with certain criteria that define items, they can identify edges after being exposed to a large quantity of samples.
In October 2017, Google Brain chief, Jeff Dean said at VB Summit, Berkeley, ‘If you’ve 10 samples of something, it’s going to be very difficult to make deep learning work. However, if you have 100,000 records or whatever you care about, that’s the kind of scale where you can expect deep learning technique to work.’
Today, image recognition system developed on deep learning is better than humans in some cases – this ranges from recognizing cats to identifying indicators of blood cancer and tumors in MRI scans. Google AlphaGo trained on the game of Go (much more complex than chess), and it turned its neural network by playing against itself over and over. In March, 2016, it became the first computer program to defeat a professional human Go player
Visualizing AI ML and DL
Image credit: Nvidia
The simplest way to think of AI, ML and DL relationship is to visualize them as concentric circles, in which Artificial intelligence comes first, then machine learning, and finally deep learning which is driving present AI explosion.
From Bust To Boom
Artificial intelligence has been part of human imagination and simmering in research labs since 1956. We have made more progress in 5 years since 2012 than we have done in preceding 25 years on numerous key AI problems such as text understating, signal processing, voice and image recognition (a tough job).
The main reason of explosion of AI in the past few years is wide availability of GPUs that make parallel processing even faster and cheaper. It also has to do with practically infinite storage and the whole Big Data movement – text, image, transactions, you name it.
Today, all tech giant companies are heavily investing in AI projects, and billions of people interact with AI software on daily basis through web search engines, social media and eCommerce platforms, and one of the types (or you can say the only type) of AI that we interact with most is Machine Learning.
According to statista, global revenue of AI market would cross $59 trillion by 2025.
AI Is The Future, Thanks to Deep Learning
Deep Learning has enabled several practical applications of Machine Learning, by breaking down tasks in manners that make all types of machine assists seem possible. Better product recommendations and story suggestions, better preventive healthcare, Driverless vehicles – today, all these things are possible. With the help of Deep Learning, AI may even get to that sci-fi state humans have so long imagined.