Artificial Intelligence (AI) is reshaping industries and daily life, powering innovations from virtual assistants to medical diagnostics. For ServiceNow customers and developers, AI plays a vital role in automating workflows, enhancing decision-making, and driving efficiency across the Now Platform.
This article aims to provide a high-level overview of AI, its core subsets, and how these technologies intersect to support and enhance ServiceNow products. By understanding the different branches of AI and their applications, you will gain insight into how these tools are embedded into ServiceNow solutions to transform business operations.
What is Artificial intelligence (AI)?
The AI Alliance's Guide to Essential Competencies for AI defines Artificial Intelligence as a collection of technologies designed to emulate human intelligence—computational systems that perform tasks previously requiring human intervention.
These tasks can include:
-
Natural language processing (NLP)
-
Pattern recognition
-
Problem-solving
-
Decision-making
-
Content generation
To learn more about AI terminology, I encourage you to explore the AI Alliance's glossary of algorithms and learning techniques.
Subsets of AI in ServiceNow
AI consists of several branches, each contributing to different functionalities within ServiceNow. This article focuses on the following key subsets:
Machine Learning (ML) is a subset of AI that uses algorithms or models to learn from data to make decisions or predictions and improve performance on tasks over time without explicit programming for each task. ML powers several capabilities within the Now Platform including:
-
Predictive Intelligence – Automates ticket categorization, anomaly detection, and task routing in IT Service Management (ITSM) and Customer Service Management (CSM).
- Task Intelligence - Uses machine learning to train solutions with your data to achieve important outcomes, such as automating task creation, triaging, and investigation.
- Document Intelligence - This is an AI solution that enables any organization to automate and accelerate the process of extracting data from documents.
- Process Mining - Helps analysts and process owners quickly analyze and optimize business processes. One example is process mining using integrated machine learning clustering to identify and remediate process inefficiencies faster.
-
Recommendation Framework – Suggests relevant knowledge base articles, accelerating resolution times in Employee Center and HR Service Delivery (HRSD).
Deep Learning (DL) is a form of ML that uses neural networks that are inspired by the human brain network of neurons with multiple layers to analyze complex data patterns, hence the term "deep." Examples of Now Platform applications of DL:
-
Optical Character Recognition (OCR) in Document Intelligence – Uses OCR and AI to identify, understand, and extract text and data from documents.
-
Virtual Agent—Q&A Genius Results display top search results extracted from HTML fields of records on the Knowledge table and tables that extend it. They are based on deep neural networks that are continually improving.
-
Security Operations (SecOps) – AI for security uses machine learning and deep learning techniques to understand cybersecurity risks.
Natural Language Processing (NLP) is a branch of AI that enables machines to understand, interpret, and generate human language. Examples of where NLP is used in the Now Platform:
-
Virtual Agent- This allows for natural, conversational interactions. Now, Assist in Virtual Agent uses LLM topic discovery instead of NLU topic discovery.
- NLP techniques are used in multiple ServiceNow products. Here's a quick search of where you can see them being used.
Generative AI focuses on creating new content—text, images, code, and more—by learning from vast datasets. Examples of generative AI in ServiceNow:
-
Now Assist – Uses generative AI that is designed to enhance user productivity and efficiency through conversation and proactive experiences.
-
Now LLM Service - Provides access to specialized large language models (LLMs) that are developed by ServiceNow.
It's important to understand these branches and the technologies used, as not all AI is generative AI, and you may need to incorporate a mixture of AI solutions for various business outcomes. As you embark on your AI journey within the Now Platform, you will come across the need to use different AI solutions. @Lener Pacania1 posted a great article on this: Not all AI is Generative AI. Solve new problems by using both!
@dangrady510 has also published a great compilation of articles: Performance Analytics, Virtual Agent, Process Mining: A Now Intelligence blog carnival if you want to dive deeper into those areas.
You may be asking, what does any of this have to do with LLMs?
Understanding that Large Language Models (LLMs) are advanced forms of ML that utilize neural networks and transformer architecture to understand and generate language. ML concepts such as training, evaluation, and deployment techniques apply to generative AI. As you and your organization explore AI solutions in the Now Platform, having a basic understanding of the differences between ML, DL, and GenAI models will aid you in the outcomes of the applications, the types of data and training methods needed, and the types of models being used.
Here's a relationship diagram of the AI technologies we use in our products today, you can see there are layers to the branches of AI and how our products have been using AI technologies prior to generative AI. This is not an exhaustive example but a general overview:
Training Methods for AI in ServiceNow
Now that you have an overview of the branches of AI that pertain to ServiceNow currently, hopefully you have a better understanding of where GenAI falls into the larger picture.
Unsupervised Learning:
- Goal: The model learns from data without labeled examples or instructions. The model learns the underlying structures and patterns within the data. For example, an ML model designed for clustering will group similar data points into clusters. An LLM will be exposed to very large amounts of unlabeled text data to learn the properties of each language, such as grammar, syntax, and semantics in order to predict the next word or token in the sentence. An LLM that has undergone unsupervised learning may also be referred to as a pre-trained model.
Supervised Learning:
- Goal: The model learns from labeled datasets to predict known outcomes such as input/output pairs. These datasets may be smaller sample datasets that include demonstration data, prompts (for LLMs), and corresponding responses. For example, an ML model such as a classification model uses supervised learning techniques where data is labeled with a label such as category (think spam/not spam). An LLM may use the technique of Instruction Fine-Tuning (IFT) with input (prompt)/output (response) pairs to train a model. IFT is a form of fine-tuning that transforms the model from a general-purpose model by adding control over its behavior. It aims to create an LLM that understands cues as instructions rather than text. An LLM that has undergone supervised techniques may be referred to as a fine-tuned model.
Reinforcement Learning
- Goal: To improve model performance by incorporating human feedback into the learning process. Reinforcement Learning from Human Feedback (RLHF) trains models incrementally to align with human feedback across multiple iterations. This learning method aims to improve instruction following, content safety, and alignment with human preferences. Over time models become more aligned with user needs, ensuring higher accuracy, relevancy, and improved user satisfaction.
There are other types of learning/training methods that aren't covered here, but these are key concepts to know about LLMs. You may hear of foundation models or pre-trained LLMs being used interchangeably. I'll cover foundation models in another article; at a high level, these are models that have been trained on that larger unlabeled dataset to learn patterns, syntax, etc. using unsupervised training. When you hear of fine-tuned or domain-specific LLMs, normally what is being referred to are pre-trained models that are trained on smaller, domain-focused (healthcare is a good example; for ServiceNow, we focus on product domains such as ITSM, CSM, etc.) labeled datasets with input/output examples using a supervised learning technique, and other optimization techniques in order to deploy to production as fine-tuned LLM. An example here at ServiceNow is our Now LLMs.
Understanding the building blocks of AI—including ML, DL, NLP, and training techniques —provides insight into how ServiceNow leverages these technologies to enhance workflows and improve operational efficiency. As AI continues to evolve, ServiceNow's integration of AI-driven features will play a key role in shaping the future of enterprise automation.
