Understanding AI technologies helps companies choose the right tools for their needs and smoothly integrate them into their current operations.

AI technologies can be best understood as a kind of nesting of layered technologies that all fall under the overarching category of IPA.

Intelligent process automation

IPA is a grouping of technologies used to manage and automate digital processes and is designed to assist human workers by augmenting human labor and performing tasks that are typically repetitive processes. IPA includes the following technologies:

• Robotic process automation (RPA): Seeks to automate tasks within existing processes
• Digital process automation (DPA): Uses low-code tools to automate processes that can span multiple applications with a focus on automating or partially automating tasks involved in a variety of business practices that typically require some form of human interaction
• Business process automation (BPA): Seeks to automate and streamline multi-step processes
• AI: Systems with the ability to learn, perform and reason like a human

Moving into the AI realm, IPA begins to make decisions based on the provided data and identified trends to augment human decisions or even make decisions independently. IPA utilizes AI to drive innovation, optimize and streamline workforce efficiency.

Artificial intelligence

AI is an overarching term that encompasses the field of developing computer systems to perform tasks that require human intelligence, such as reasoning, decision-making and pattern learning and includes any technology that falls within the scope of ML and automation.

Additional subfields of AI include:

• Natural language processing (NLP): Seeks to enable machines to understand, interpret and respond to human language in a valuable way
• Large language model: Trained on extensive datasets to understand and generate human-like text and can perform various language-related tasks
• Computer vision: Seeks to enable computers to derive information from images, videos and other data
• Deep learning: ML in which multi-layer artificial neural networks adapt and learn from vast amounts of data

With the continued advancements in autonomous vehicles and the ability to tackle complex, nuanced challenges, artificial intelligence continues to evolve, inching closer to human-like cognitive abilities.

Machine learning

Machine Learning (ML) enables AI systems to learn from data by recognizing patterns and making predictions or recommendations based on statistical analysis. It can adapt over time with new data, but its learning is generally guided by specific features and rules defined by developers.

There are four types of ML:

• Supervised learning: In supervised learning, models are trained on labeled data. Each training example is paired with an output label. The algorithm iteratively makes predictions on the training data and is corrected by the teacher, allowing the model to learn over time.
• Unsupervised learning: Machine learning models that operate without the need for labeled data. Unlike supervised learning where models are trained on a labeled dataset, unsupervised learning algorithms work with datasets that have not been annotated or categorized. Unsupervised learning is crucial in scenarios where obtaining labeled data is impractical or too expensive. It provides a method to extract meaningful information and insights from raw, unstructured data, paving the way for deeper understanding and better decision-making.
• Semi-Supervised Learning: Semi-supervised learning falls between supervised learning and unsupervised learning. In semi-supervised learning, the algorithm is trained on a dataset that contains both labeled and unlabeled data. Generally, a small amount of data is labeled while a large amount of data is unlabeled.
• Reinforcement learning: When the model learns from its own actions and feedback from the environment, such as a robot navigating a maze or a game agent playing chess, before using an algorithm to find a policy that maximizes a reward function with the goal of learning optimal behavior for complex and dynamic situations.

ML is widely used in various applications, such as speech recognition, image recognition, spam filtering, fraud detection and self-driving cars.

Deep learning

Deep learning is a subset of ML that uses artificial neural networks to enable digital systems to learn and make decisions based on unstructured, unlabeled data. A neural network is a series of algorithms that can learn from input data to discern features, such as distinguishing characteristics among various images, to make independent decisions without explicit programming.

ML enables AI systems to learn from data by recognizing patterns and making predictions or recommendations based on statistical analysis. It can adapt over time with new data, but its learning is generally guided by specific features and rules defined by developers.

On the other hand, deep learning employs neural networks with multiple layers (hence why it's often referred to as deep) to analyze various factors of data. Unlike traditional ML, deep learning autonomously extracts features from raw data, eliminating the need for manual feature extraction. It learns from data in a way that is somewhat analogous to human learning, through a hierarchy of concepts where each layer of the network extracts and refines features from the input data.

Generative models, like those used in generative adversarial networks (GANs), are a notable application of deep learning. They learn to generate new data that resembles the training data and can be used to improve decision-making capabilities over time as they are exposed to more data. However, it's not just generative AI that benefits from deep learning; other forms of AI, like those used in image and speech recognition, natural language processing and autonomous systems also leverage deep learning to enhance their performance and capabilities over time.

Generative AI

Generative AI refers to a category of algorithms that are capable of generating new data that resembles a given set of training data. While it's not exclusively about language models, it indeed encompasses them. These algorithms can create a variety of content types including images, text, computer code or audio based on the patterns they learn from the input data, aiding in accelerating the creative process.

Large language models like ChatGPT are instances of Generative AI applied to text generation. They are trained on vast datasets to produce human-like text based on the input prompts they receive. These models can be fine-tuned to perform a myriad of tasks, enhancing their versatility and utility across different domains.

It's important to note that generative AI also includes models like GANs and variational autoencoders (VAEs) which are used for generating images, audio and other types of data beyond text. Hence, while large language models are a subset of generative AI, the term generative AI encompasses a broader range of models and capabilities.

Generative AI has evolved into an advanced search and content generation tool integrated across various industries to augment daily tasks and produce new content. With 55% of organizations piloting or in production mode with generative AI [1], the growth of AI in the market highlights the need for risk management as the creation of new AI-driven content accompanies inherent risks.