Throwing around terms like AI, GenAI, and ML without understanding what they mean is like ordering from a menu in a foreign language—you might end up with something you really don’t want! To avoid this undesirable outcome, the following post reviews AI fundamentals and clarifies the terms we use in roadmaps, announcements, and other product-related documentation. We hope that you’ll find it informative, and that it will facilitate clearer, more productive conversations and communications.

Artificial intelligence (AI)

AI is an inclusive term that refers to the use of data (information) and algorithms (rules) to allow computers to learn, act, and perform functions that are normally associated with human intelligence. Here are few examples of AI capabilities:

Common-sense reasoning: creates the appearance of human-like understanding and reasoning, bridging the gap between data-driven patterns and the judgment, intuition, and context-sensitive decision making that humans do routinely Abstract thinking: enables computers to engage in more complex problem solving like workflow augmentation or automation

GenAI and ML (discussed below) are two of the more well-known branches of AI. There are important differences between them, and while it is correct to say that GenAI and ML are AI, it is not correct to say that all AI is GenAI or ML.

Generative AI (GenAI)

GenAI refers to a specific subset of AI that uses programs to process large data sets, detect patterns, and then create new works of text, imagery, video, and even computer code based on the instructions it’s given (known as “prompts”). GenAI relies on artificial neural networks, which are methods for processing information that mimic biological neural networks . It’s limited by the data it's fed to train its models, so everything it produces is derivative of the data it learns from. (Related, the push to train models with bigger and bigger training sets is one of the factors driving demand for AI compute power.)

Here are few examples of GenAI capabilities:

Creativity: generate artistic creations, write stories, compose music, and even create visual art Novel data creation: produce new data that wasn't present in GenAI’s training set, enabling the generation of synthetic data (see below) which can be useful for financial planning, workforce planning, supply chain planning, treasury planning, etc. Synthesis: create new summaries, arguments, and viewpoints based on synthesized information Machine learning (ML)

ML systems learn and improve based on the data they consume. There are two major types of learning algorithms—supervised learning and unsupervised learning—which refer to the way the model uses data to improve its performance. With supervised learning, the ML algorithm is presented with inputs plus the desired outputs to help it discover a general rule that relates them; with unsupervised learning, the algorithm is presented unstructured data and left to discover relationships and patterns on its own. (Since precision matters in the scope of this post, it’s worth noting that GenAI often uses ML techniques, in addition to natural language processing.)

Here are examples of ML capabilities:

Predictive analytics: predict outcomes based on historical data patterns (e.g., using actual financial results to create forecasts) Anomaly detection: detect anomalies or outliers in data, usually used to aid in fraud detection or quality control Recommendation systems: suggest products, content, or services based on user preferences and behavior Regression analysis: analyze relationships between data to determine their strength (e.g., you can use it to analyze the relationship between advertising spend and potential revenue increases due to advertising) Further reading on other terms

There are other common terms that can be helpful:

Deep learning: a type of ML that uses artificial neural networks to learn from data; it’s used to solve a wide range of problems, including image recognition, natural language processing, and machine translation Hallucinations: refer to false or inaccurate outputs that are produced by an AI model, usually because of incomplete or noisy data, or when it is used to make predictions outside of its domain of expertise Prompt engineering: a technique used to optimize and fine-tune language models for specific tasks and desired outputs (also known as prompt design) Synthetic data: used as testing samples to evaluate the performance of GenAI models; it is created by generating new data that shares the mathematical properties of real-world data

In conclusion

We may be tempted to use terms interchangeably, but we shouldn’t, because AI is not the same as ML, and GenAI diverges from traditional AI. These differences are important. And understanding them will help you navigate Fusion Apps enabling technologies, product enhancements, and roadmaps.