Although history has not always been kind to it, artificial intelligence is definitely here to stay. People continue to conduct groundbreaking research in the field of AI on a daily basis. They also try to find new ways to integrate it into other areas of our lives.
Data from PricewaterhouseCoopers indicates that business leaders expect artificial intelligence to be fundamental in the future, with 72% of them actually calling it a business advantage. So, it is obvious that many more will want to jump in the AI bandwagon and help revolutionize the world. This also means we will continue to hear more about machine learning, algorithms, supervised learning, deep learning, and many other concepts we might not comprehend. And let’s be honest, researchers and technical experts are not helping either. Most of the times they talk about neural networks and deep learning and how great they are in image recognition, but they fail to explain what exactly the terms mean. Today, understanding the main terminology and concepts in AI is essential for everyone in the business world.
The idea of a device being ‘intelligent’ has been embedded in the human mind since antiquity. Yet, the term artificial intelligence has been defined and founded as an academic discipline much later, in 1956.
Artificial intelligence refers to a specific field of computer engineering that focuses on creating systems capable of gathering data, making decisions and/or solving problems. In plain English, the term refers to intelligent behaviour displayed by a machine. This means that computers can now perform more tasks that were done by humans. If in the past, computers could conduct simple logical tasks such as solving certain math problems, today they can also carry on conversations, identify objects or people in pictures and even recognize core emotions.
Algorithms are the cornerstone of artificial intelligence. They represent a set of rules (math formulas and/or programming commands) that help a regular computer learn how to solve problems with AI. Although they are initially set by programmers, these rules allow computers to learn without any further human interaction. The algorithms are tweaked using machine learning, which means the programs start to adapt these rules for themselves.
One of the most important subsets of artificial intelligence is machine learning (ML). They are so connected that, sometimes, people even substitute the terms for one another. But this should not be the case. Machine learning is the part of artificial intelligence that allows computers to improve themselves over time as a result of experience and practice. According to computer scientist Arthur Samuel, who coined the term in 1959, machine learning enables computers to learn without being explicitly programmed.
To work accordingly, machine learning requires constant access to varied data as well as properly-defined algorithms and accurate data sets. Machine learning uses two types of techniques, namely supervised and unsupervised learning.
Supervised learning (SL) is based on a known set of input data and known output (responses to the data). A supervised learning algorithm trains a model to generate reasonable predictions for the response to new data. This means that supervised learning can only be used if you have known data for the output you are trying to predict.
The technique can be used in a wide range of domains to predict future behaviours based on existing data that can be combined into a model. For instance, in healthcare, they use SL to predict heart attacks and in fintech to determine fraudulent loan applications.
Unlike SL, unsupervised learning (UL) does not rely on known outputs, but only on unlabeled input data. An unsupervised learning algorithm needs to find patterns or intrinsic structures in the data. The final goal is to model the underlying structure or distribution in the data in order to learn more about it.
Unsupervised learning problems can be further grouped into clustering and association problems. Clustering is the most common unsupervised learning technique. It is used for exploratory data analysis to find hidden patterns or groupings in data. Applications for cluster analysis include gene sequence analysis, market research, and object recognition. For example, Airbnb uses UL algorithms for grouping the housing options into neighborhoods to ease users’ navigation.
Association is used when you want to discover rules that describe large portions of the data, such as people that buy product X also tend to buy product Y.
Some of the recent major breakthroughs in the field of machine learning have been related to one of its main sub-fields, called deep learning (DL). DL teaches computers to learn by using examples. Deep learning is inspired by how the human brain works and uses a structure called an artificial neural network which is arranged into many layers, between an input and an output. Deep learning systems assimilate large quantities of data and are able to determine the features related to that data through supervised or unsupervised learning. In image recognition, for instance, DL is able to work out for itself what a certain animal looks like, based on the input data. In contrast, machine learning needs to be told what features make up that specific animal for it to be able to recognize it.
Similar to deep learning, neural networks tend to imitate the processes of the human brain. Also known as artificial neural network, neural net, deep neural net, and other similar terms, the neural network is able to divide and analyze the data into different levels. Breaking down the data allows them to solve complex problems by using the results from a certain level to understand data from the next one. Let’s take for example the case of image recognition. Instead of looking at an animal and deciding it’s a monkey, the neural network takes into account different features of the image and monkeys in general, analyzes them and assigns different levels of importance. This process continues at every level of a neural network. In the end, we get a much more accurate result.
Generative Adversarial Networks
Generative adversarial networks (GANs) are deep neural net architectures comprised of two nets, pitting one against the other. GANs’ main ability is that it can learn to mimic any distribution of data. Generative algorithms try to predict features given a certain label or category, as opposed to discriminative algorithms that are concerned solely with mapping features to labels.
GANs work by simultaneously training two different models: a generative model, the generator (G), which generates new data instances and a discriminative model (D) which reviews them for authenticity. The discriminator decides whether each instance of analyzed data belongs to the actual training dataset or not.
Natural Language Processing
Natural language processing (NLP) is an application domain for machine learning techniques. NLP is about training an AI software to interpret human communication. This can be used for AI chatbots and translation services as well as AI assistants like Siri and Alexa.
Also called bots, interactive agents or conversational interfaces, AI chatbots are artificial intelligence systems that use NLP capabilities to carry on conversations. AI chatbots can be used in different industries such as sales, marketing, human resources, finance, insurance and many others.
In general, we do not need to know everything in a conversation, but we should at least know the majority of the terms used in that conversation. The glossary of artificial intelligence is much more comprehensive than what we covered in this article, and we understand it can be a bit scary at first. But perseverance and a steady interest wins the race in the long run. As the great Albert Einstein once said:
“Any fool can know. The point is to understand.”
— Albert Einstein