Artificial Intelligence (AI): Types & Algorithms

AI Types

Expert Systems

Despite the fact that some experts claim that Expert Systems (ES) is not real AI, as they lack the ability to learn independently from external data, ES can capture knowledge in a very specific and limited domain of human expertise. It usually performs limited tasks such as diagnosing malfunctioning machines and determining whether to grant credit for loans.

Machine Learning (Main Focus)

To make decisions, Machine Learning (ML) primarily uses algorithms and data. In B2B marketing, ML has outstanding value. It can, for example, produce rich insights from consumer purchasing behavior data and help make informed decisions (Cortez & Johnston, 2017) (Wright, Robin, Stone, & Aravopoulou, 2019). The key methods are supervised learning, unsupervised learning, and reinforcement learning as we can see in the below graph.

Machine Learning Methods (Source: KDnuggets News)

Supervised learning utilizes labeled knowledge. Labeled data refers to data where a mark is properly applied to a known data attribute (or feature). Unsupervised learning clearly isn’t the same! Scientists do not track the model; the model learns on its own during the unsupervised study as there is no examples of attributes and labels of objects. Reinforcement learning is based on behaviorism and is a psychological interpretation of human behavior (Lison, 2015).

Neural Networks and Deep Learning Networks

Neural networks try to find patterns and search for relationships by building models and correcting them over and over in massive amounts of data which is too complicated for humans to analyze. In medicine, neural networks are used for pattern classification, prediction, financial analysis and optimization issues.

Genetic Algorithms

Genetic Algorithms (GAs) originally inspired by the Darwinian theory of evolution by the (genetic) selection, are a heuristic solution-search or optimization technique. To create solutions to given problems, a GA uses a highly abstract version of evolutionary processes. On a population of artificial chromosomes, each GA operates in a finite alphabet, and each chromosome is a solution to a problem that has a measured real number to evaluate how good a solution to the specific issue is.

Natural Language Processing

An area of Artificial Intelligence that gives computers the ability to interpret, understand and infer meaning from human languages is Natural Language Processing or NLP. It is an area that focuses on the relationship between data science and human language and applies to a broad variety of fields, including healthcare, media, finance and human resources (Yse, 2018). A good example of NLP is Google Translate and digital assistances: Siri, Alexa, Cortana, Google Assistant.

Computer Vision

Computer vision is a digital image system that generates a digital image map (like a face or a street sign) and recognizes this image in near real-time in large image databases where each image has a specific pixel pattern. Some important applications of computer vision can be:

• Facebook’s DeepFace can identify friends in photos across their system and the entire web
• Autonomous vehicles can recognize signs, road markers, people, animals, and other vehicles with good reliability
• Industrial machine (robot) vision
• Passport control at airports
• Identifying people in crowds

Robotics

NASA identifies robotics as the study of robots. Robots are machines that can be used to do jobs. Some robots can do work by themselves. Other robots must always have a person telling them what to do. Design, construction and operation of machines that can replace humans in many applications for factories, offices, and homes.

Robotics are generally designed in restricted domains to perform precise and comprehensive acts, e.g. robots spray paint cars and assemble certain parts, heavy assembly movement or hazardous circumstances such as bomb disposal and surgical capabilities.

Bionics & Cybernetics living systems, Multi-agent Systems (air traffic control) and Semi-Fully Autonomous Driving (Examples range from autonomous helicopters to Roomba, the robot vacuum cleaner) and Vehicular Automation (such as Tesla’s “Autopilot” system) are technologies that fall under the robotics of AI.

Intelligent Agents

They work without direct human intervention to carry out repetitive, predictable tasks like deleting junk e-mail and finding the cheapest airfare, such as typical chatbots and Siri.

Common Machine Learning Algorithms

Machine learning algorithms range from linear regression and logistic regression to deep neural networks and ensembles in complexity (combinations of other models). However, some of the most widely used algorithms are as follows:

*X means is extending K-means with efficient estimation of the number of clusters

** Are ensemble algorithms that create a series of models where each new model tries to correct errors from the previous model

?Categories:

1.????Classification: We may use classification to deal with a wide variety of issues. It helps us make better choices, filter spam, predict whether a borrower can repay a loan, and tag friends in Facebook photos.

2.????Regression: A standard regression task is to determine the relationship between two or more continuous variables for prediction of prices for example.

3.????Clustering: Find similarities between objects and then group them together based on what they have in common.

4.????Optimization: Use product utilization analytics to figure out how new product features impact demand.

5.????Anomaly Detection: Helps identify and deter fraudulent transactions in real-time, even for forms of fraud that were previously unknown.

There are other algorithms that can be used for ranking (such as search engines) and recommendation systems to motivates user to buy more or explore more content with valuable suggestions. Choosing the right algorithm can be based on the Task-Based Learning Approach (Input/ output), understanding the data and trial & error approaches (Gavrilova, 2020).

Many companies are working with consumers these days, gathering information on where the consumer is anxious, and introducing services with an upgraded version of the previous version, which was offered by the old telecom industry, and gaining customers through their advertisements, resulting in huge losses to other telecom industries (Buckinx & Poel, 2005) (Shyry, 2014). Many experiments have been conducted to identify several unusual types of predictors in order to create a customer stress model, where stress here means losing a customer by moving to another industry. This model can take into account demographics, environmental shifts, and other variables (Srinivas, Manikanta, Jacob, Nagarajan, & Pravin, 2021).

Business to customer (B2C) interactions are becoming more individualized and pervasive, resulting in heavily digitized footprints. Companies have been investing heavily in machine learning to improve their marketing capabilities as a result of the abundance of data. According to BCC Reseach, by 2022, the global demand for machine learning-enabled solutions would have grown at a 43.6 percent annual pace, reaching $8.8 billion.

Deep learning engines analyze and tag billions of photos on social media sites like Facebook, and sophisticated ML algorithms power recommender systems at e-commerce websites and content channels like Amazon and Netflix. To assess the best bid for ad distribution, automated bidding algorithms test a web surfer's profile in milliseconds. Human-like communications are carried out by chatbots. AI agents operated by machine learning algorithms have demonstrated their effectiveness in processing large-scale and unstructured data in real-time, producing accurate predictions to aid marketing decisions, through applications such as social media mining, sentiment analysis, and consumer churn prevention. All aspects of corporate success have significantly improved as a result of these measures.

However, despite the increasing interest, the use of machine learning approaches in marketing is still in its early stages, and existing research is quite scattered. To date, there is no clear vision or a coherent structure for incorporating machine learning approaches into marketing research.