2.2 Key AI Technologies

There are many technologies and concept terms that are thrown around in the AI discipline. Some are used interchangeably; some are used correct; and many are used incorrectly. Let's identify some distinct concepts, techniques, and technologies to clarify this field.

But first, let's revisit a basic definition of AI. It's the ability to replicate human decision making with computer science and engineering solutions.

Machine Learning

Machine learning (ML) is a subset of AI that involves the use of algorithms and statistical models to enable computers to improve their performance on a specific task through experience (i.e., data) without being explicitly programmed.

Just like how a person learns to ride a bike through practice and experience, an ML model learns from data. The more data it gets, the better it becomes at making predictions or decisions. Or, an ML model works the same way our preferences are formed for products as we get experience purchasing. The more experience we get trying out new products, the more accurate our next purchase will be; meaning the more likely that product will fulfill our needs or desires. ML is used in all kinds of applications to recommend products, filter out spam email, order our social media feeds, and much more.

Neural Networks

Neural Networks (NN) are a type of ML algorithm that attempt to recognize underlying relationships in a set of data through a process that mimics the way the human brain operates. They are the foundation of deep learning which we'll talk about next. They are the most sophisticated type of algorithm used in ML.

NN work the same way that neurons in the human brain connect and communicate. They consist of interconnected nodes (neurons) that process information and learn from it. Similar to how a spider web has interconnected threads that catch prey, neural networks have interconnected nodes that capture patterns in data. NN are used in applications like facial recognition systems and voice assistants. The are also commonly used to predict stock prices and economic trends based on historical data.

Deep Learning

Deep Learning (DL) is a subset of neural networks that require many layers (hence "deep") to analyze various factors of data. It is particularly effective for large datasets and complex tasks with many variables (a.k.a. "features").

DL works the same way a person recognizes a face by identifying various features (eyes, nose, mouth) and putting them together, a deep learning model uses multiple layers to process data and identify patterns. It is also similar to how a person learns to understand a language by recognizing words, sentences, and context. DL models learn to understand data by processing it through multiple layers of abstraction. DL is used in self-driving cars to identify objects and in Facebook to tag friends in photos. It is also used by Google Assistant and Siri for speech recognition and understanding.

Computer Vision

Computer Vision (CV) is a sub-field of AI, and ML in particular, that enables computers to interpret and make decisions based on visual data from the world, such as images and videos. It requires a trained ML model to help recognize what it sees in images and video.

CV works no differently than the way humans look at a photo and recognize objects. CV systems analyze images to identify and understand what's in them. Just as a driver reads and interprets traffic signs to navigate, CV enables autonomous vehicles to recognize and react to road signs and signals. Security systems and doorbell cameras use CV to identify individuals based on facial features. CV also helps with medical imaging to analyze X-rays and MRIs to detect anomolies.

Natural Language Processing

Natural Language Processing (NLP) is a branch of AI that focuses on the interaction between computers and humans through natural language. It involves understanding, interpreting, and generating human language. NLP includes a wide variety of tasks to break down unstructured text to extract meaning as well as generating new text based on prompts. Some of these tasks use ML algorithms while some do not.

Customer service chatbots that understand and respond to customer inquiries are built upon NLP models. Google Translate uses NLP to translate text from one language to another.

Generative Adversarial Networks

Generative Adversarial Networks (GANs) are a class of machine learning frameworks where two neural networks (a generator and a discriminator) are pitted against each other to produce new, synthetic instances of data that can pass as real. While GANs are often considered as another type of neural network algorithm, they are really more of a method or technique that uses other deep learning algorithms to generate high quality images.

GANs work like a forger creating counterfeit art and an expert detecting fakes; or a student writing an essay and a teacher grading them. GANs are often used to generate realistic images of imaginary people. They also are great for generating synthetic data to train other AI models--a critical task in both predictive and prescriptive analytics.

Expert Systems

Expert Systems (ES) are AI programs that mimic the decision-making abilities of a human expert. They use a knowledge base of human expertise and an inference engine to solve specific problems. ES are much less sophisticated than all six of the AI technologies discussed previously. The only technology required for ES is a database of rules and facts that guide the ES. Although they still require data, it is significantly less data and more simple than the data required for ML, NN, DL, CV, NLP, and GANs.

Think of ES like asking a specialist for advice on a complex issue or similar to how a doctor uses medical knowledge to diagnose patients. ES use stored expertise to diagnose problems and suggest solutions. In fact, ES are often used for those exact tasks. WebMD.com is a perfect example of an ES. ES are also often used in the financial industry to offer investment recommendations based on market data.

Robotic Process Automation

Robotic Process Automation (RPA) refers to the use of software robots or "bots" to automate highly repetitive, routine tasks usually performed by a human interacting with digital systems.RBA is the most basic form of AI and even less sophisticated than ES. It requires no data whatsoever and only a set of rules that never change once they are initially defined whereas ES rules and often updated with the latest knowledge.

RPA is often used in assembly line manufacturing to perform routine tasks that were once performed by a human. RBA is also used to automate data input from invoices into accounting or similar systems. When you use the password reset feature of an app or website, the logic used by the system to help you is a basic form of RPA.

Swarm Intelligence

Swarm Intelligence (SI) is an AI paradigm inspired by the collective behavior of social insect colonies and other animal groups. It involves decentralized, self-organized systems.

SI systems use simple agents working together to solve complex problems much like how ants work together without a central command to find food efficiently or how birds flock together and change direction in unison. SI systems involve agents coordinating their actions based on local interactions. In practice, SI systems are great for solving complex logistical problems, like route planning for delivery trucks. They also help coordinate multiple robots to achieve a common goal

Summary

To be clear, this list above is not a comprehensive set of techniques, algorithms, and technologies in the AI field. But these are some of the most significant ones that will get you started on becoming "AI-literate".