AI is the new buzzword of the 21st century. Artificial intelligence is coming in to our lives faster than we had anticipated. It’s helping us in shopping. It’s at the other end when you’re talking to customer service centre. It’s driving our cars and even recommending you videos on YouTube. It’s making sure that it knows about you more than you do.
Alan Turing defined AI as:
“Artificial Intelligence is the science of making machines do things that would require intelligence if done by man.”
That means, AI is not specifically related to computer science. This is a field of study that encompasses human behaviour, biology, psychology, and even language and linguistics. There’s still not a common consensus among academicians about its definition.
In this blog post, we try to give a broader picture of AI. How it is organized and its various areas and fields of study. First we will discuss the terminologies associated with AI and then we will discuss the techniques used in implementing AI.
Key AI terminologies
1. Machine Learning
First thing to understand about machine learning is that it is an application of AI. It is the process by which we create systems that have ability to learn with experience.
For instance, the systems which automatically identify spam emails are trained to do so by exposing them to millions of e-mails which are spammy and non-spammy in nature. With more data, the program is able to understand and learn what makes an email.
Machine learning is different from a fully-AI system in the way that it does not know what it was trained to do. For instance, the above mentioned AI system can identify spams but it will never be able to tell why it is doing that. Or if a new type of email comes in, it will fail to understand it.
2. Cognitive Analytics
As the name suggests, this field of study deals with the ‘thinking’ part of an AI. This system can not only analyse data to create obvious results, but it can train itself to deduce and synthesize new thoughts. For instance, if this system reads a Harry Potter book and is asked who were the two best friends of Harry, it can reply with Hermoine and Ron. To put it more clearly, this system can make sense of unstructured data.
This system is being used a lot in chat bots and customer service platforms. Many recommender systems on the internet are using this type of systems to give custom content recommendation.
Although a more general term that existed before the onset of the AI wave, robotics is getting a new push with the rise of AI. In the recent years, various companies have launched AI backed robotic systems. For example, MIT’s Cheetah II and IPsoft’s Amelia have taken AI based robotics to another level.
Robotics, academically, is defined as a programmed machine that can perform rule-based tasks. With AI, the ‘rule-based’ part broadens as the robot learn with its experiences.
Techniques Used in AI
Myriads of AI techniques have emerged in the past decade for implementing and building AI systems.
1. Natural Language Processing
In a one-liner, natural language processing is the study of how a computer interacts with a human language. Broadly, in application sense, it refers to speech recognition and speech synthesis in human language.
This field of study is already in application phase and companies are using it in their voice assistants. Apple’s Siri, Google Assistant, Microsoft’s Cortana, and Amazon’s Alexa relies a lot on natural language processing.
Natural language processing further uses different techniques for implementation like parsing techniques, text recognition, and part-of-speech tagging.
2. (Artificial) Neural Networks
Neural networks are available in living beings. Humans and animals uses a complex network of billions of neurons (which makes neural systems) to take decisions in day-to-day life and learn new things to do. Building artificial neural networks is an attempt to create neural networks modelled on our own brains!
These networks can identify patterns in inputs as it processes a lot of data and learn from it. It uses different learning methods: supervised learning, unsupervised learning, and reinforced learning. Neural networks have wide applications in pattern recognition, machine learning, and deep learning.
3. Vector machines
Vector machines are really capable in solving classification problems. For instance, an email system like Gmail for classifying an email as ‘Social’ or ‘Promotion’ or ‘Personal’ in nature and categorizing them in their respective categories.
The fundamental of vector (or sometimes called support vector) machines is to create parameters that draws the line between do distinct objects dividing them into two classes. This technique of AI has wide applications in image recognition, face recognition, and text recognition systems.
This is probably the most basic technique in AI. And it totally comes from our understanding of human behaviour in learning processes. We make mistakes and learn. That’s how heuristics work. Heuristics work on the principle of trial-and-error. You keep doing things until you chance upon the answer or the right solution.
Heuristics are good for solving problems where it is hard to come upon a solution definitely. For instance, telling which route is shorter on a map. The best way to solve this problem is by identifying all possible routes and then identifying the shortest one or the one which has least traffic.
Rise of AI has given way to five new technologies which are bound to change the way we live our lives. They are:
- Cloud computing
- Internet of things
- Big data
- Application programming interfaces (or APIs)
- Open source technologies
If you want to develop your career or create an outstanding product, you can choose any specific field of AI rather than trying to cover them all. You cannot have it all. AI is such a large and new field that it is practically impossible for a single company to cover everything.
If you have any questions or queries related to this article or you want to add anything in the anatomy, please feel free to write down them in the comments below.