Branches of AI | 6 Branches of Artificial Intelligence

1. AI Robotics

The sci-fi movies telling a story about robot dominating the human world is no longer fiction but soon could turn into reality. With the advent of the Sophia robot becoming a legal citizen in Saudi Arabia, we can expect more such robots entering the human world. This branch of artificial intelligence is mind-boggling. To develop a robot, experts look after the robot’s design, building, and programming to undertake various tasks. Robots are artificially intelligent enough to move and perform actions with the ability to sense the environment and interact accordingly.

 A robot is a machine that may not necessarily require artificial intelligence to complete tasks. However, when such bots are embedded with artificial intelligence software, it allows this machine to act like humans, or maybe better than humans.

An artificially intelligent robot is controlled by AI programs like Machine Learning, Reinforcement Learning, and Computer vision. They are augmented with sensors like 2D and 3D cameras, Vibration sensors, proximity sensors, and Accelerometers that receive several types of real-time data. The AI program further analyzes this real-time data to produce an output.

Check out the latest developments goin on in the AI Robotics space in this video on a collab between OpenAI & Figure in the form of Figure 01 Robot:

Some other examples of AI-powered robots are Starship delivery Robots, Pepper humanoid robots, and Penny restaurant robots.  Experts have forecasted that by 2030, manufacturing jobs shall be handled by bots and 30% of jobs shall be automated.

2.Natural Language Processing

Ever wondered how Google is so efficient at predicting your search queries? Well, this branch of AI is the answer to it. Natural Language Processing is the type of AI that makes use of Linguistics and computer science to understand human language. For this, NLP makes use of techniques like Deep Learning and Machine Learning to infer the human languages.

For a machine, understanding human language is like moving mountains since there are more than 6500 languages currently spoken across the world, and each has its own linguistic rules. A computer tries to decipher the meaning of every word rather than the sentence or the phrase. However, an NLP enables the computer to read between the lines and understand the tone and the sentiment of the sentence. The use of NLP has made the business able to understand the unstructured data in the form of tweets, emails, or transcripts to be well understood and utilized well to serve the customers in a better way.

The four major steps to processing the language are:

• Morphology– word formation and its relationship with other words
• Syntax– the way the words are put together in a sentence
• Semantics– meaning and the interpretation of words and sentence structure
• Pragmatics– studies the language which is not directly spoken

Several techniques for NLP include Part-of-speech Tagging, Bag of Words, Stop word removal, Lemmatization, Stemming, Tokenization, and Semantics, among others. 

3. Machine Learning

How does Spotify recommend songs that you may like? It has over 82 million songs and over 4 billion playlists, and yet it never fails to please each of the 456 million users across the globe. How can it manage to do so? Well, it is like herding cats for a human being, but not for an AI. This branch of AI is what makes Spotify so good at it.

Machine Learning is the branch of Artificial Intelligence that makes it possible for the latest self-driving Tesla cars to function at par. Sci-fi is no longer fictional, but a reality now. The machines are programmed with AI tools to learn from the data and past experiences, enabling them to recognize patterns and make predictions without much human intervention. Check out this video by IBM on the relationship between AI & Machine Learning : 

For a machine to outperform, the more data points, the better the model, and the higher the accuracy. There are different ways in which a machine learns:

• Supervised Learning– It uses labeled data sets to train the algorithms for the classification of the data and predict the outcomes. The input data is provided to the model, and the model adjusts the weights to fit it appropriately. Such supervised learning helps to avoid real-world problems at a large scale. One such example is classifying mail as spam and separating it from your inbox. The common methods used in supervised learning include neural networks, linear regression, random forest, support vendor machine (SVM), and logistic regression.
• Unsupervised Learning– It makes use of machine learning algorithms for analyzing and clustering unlabeled datasets. Such algorithms are applied to discover hidden patterns without human intervention. Its ability to make out the similarities and differences in the information makes it an ideal choice for exploratory data analysis, customer segmentation, pattern recognition, image recognition, and cross-selling strategies. The commonly used approaches for unsupervised learning are Principal Component Analysis and Singular Value Decomposition. The use of algorithms like neural networks, K-means clustering, and probabilistic clustering methods is done.
• Semi-supervised Learning– For training the machine, a smaller labeled data set is fed to guide classification and feature extraction from the larger, unlabeled data set. It is helpful when it is too costly to label enough amount of data.
• Reinforcement Learning– as the name suggests, the model learns by trial-and-error method. The machine will be rewarded or punished as the outcomes are produced. This trains the machine to provide the best possible solution for a given problem.

4. Neural Networks

Did you simply like a product page on Instagram and now you see its advertisement across all the social media platforms? How did it happen with just a like?

The answer is Neural Networks! This branch of artificial intelligence makes use of data points like gender, age, location, likes, and dislikes, and shows you ads that target these categories. A neural Network can also be called an Artificial brain since it uses interconnected nodes or neurons in a layered structure, mimicking a human brain. These nodes work together to solve a complicated problem.

A simple neural network has three layers of interconnected neurons:

• Input Layer– It receives information from the outside world. The function of the input nodes in this layer is to process the data received, analyze it, categorize it, and then pass it to the next layer.
• Hidden Layer– A Neural network can have one or many hidden layers of nodes. A deep Neural network has many hidden layers of nodes that can theoretically receive any input and produce any output. Every hidden layer functions to analyze the output of the previous layer, process it, and then pass it on to the next layer.
• Output Layer– The function of the output layer is to present the result of the data processed by the Neural network. It may have single or multiple nodes.

5. Experts System

According to Pan American Health Organization, there are over 20 million new cases of cancer and 10 million deaths from cancer. In such a scenario, an Expert System like CaDet can help the Health Line workers detect cancer at the early stages.

Such an Expert system makes use of five components: A knowledge base, an inference engine, a knowledge acquisition and learning module, a user interface, and an explanation module, making it trustworthy and interactive. The job of an Expert System is to solve the trickiest problems in a target field. It incorporates the knowledge from multiple human Experts, thus improving the effectiveness of the answers.

The components and their function are discussed here:

• Knowledge base: It is a database containing facts and regulations that include norms for solving a problem and formulation of methods about the domain and knowledge in the specific discipline.
• Inference Engine: The primary job of an inference engine is to gather pertinent information from the Knowledge base. After analyzing, it identifies the solution to the problem submitted by the user. An inference engine can also possess troubleshooting skills as well.
• Knowledge acquisition and learning modules: This component aids the Expert System in gathering more information from varied sources. This information is stored in the Knowledge base.
• User Interface: This element allows a non-expert user to interact with the Expert system and develop solutions.
• Explanation module: It justifies the conclusion to the user.

Watch these 10 benefits of expert systems in healthcare

6. Fuzzy Logic

As said by Nelson Mandela “Nothing is black or white”. Fuzzy logic is the branch of AI that focuses on this aspect of the human world where language cannot be easily translated into the absolute terms of 0 and 1 only. Fuzzy logic is based on the “degrees of truth” instead of the usual “true or false” Boolean logic used by modern computers.

Fuzzy logic works with the set of rules used to reach a logical conclusion from the fuzzy data set. As the human mind works, the fuzzy logic is fed with the state between 1 and 0 as well. It enables the AI to imitate human reasoning and cognition. It has a wide range of applications like facial recognition, air conditioning, washing machines, vacuum cleaners, and idle speed regulation. As Fuzzy logic shares similarities with natural language, the algorithm is easier to code than the standard logical programming. Since it requires few instructions, it saves requirements for memory storage. However, there are substantial drawbacks which include the requirement of broad validation and verification, and dependency on human expertise and knowledge. If the drawbacks are ignored, it may produce inaccurate results.

A fuzzy logic is made up of four components:

• Fuzzification: It is the process involving the conversion of specific input values to some degree of membership of the fuzzy data sets depending on how well they fit.
• Fuzzy rules: It is also called the Knowledge base of fuzzy logic. It has If-then rules to be followed which are often derived from the opinions of the expert or through quantitative approaches.
• Inference method: It is the way to obtain the conclusion following the degree of membership of the input variables to the fuzzy set and the fuzzy rules.
• Defuzzification: It converts the fuzzy conclusion to detailed output values.