📙 The Siena AI Glossary

Welcome to the exciting world of Siena! We understand that some of the terms we use might sound like a foreign language to you. That's why we've created this handy glossary to help you navigate through our AI jargon.

This glossary is your go-to guide for understanding the AI terms that we frequently use here at Siena. It's like your personal AI dictionary, helping you decode the language of artificial intelligence.

User

Also known as *shopper* or *chatter,* is the person with whom the AI assistant engages in a conversation. The user is interacting with the AI system through a chat interface, typically using text or voice input, and receives responses from the AI system in the same format.

Conversation

Also known as a ticket in legacy help desks is a series of interactions between a user and the agent or a conversational AI system.

Learning phrases

Examples of user input that are used to train a conversational AI system to recognize a specific intent or entity. Think of learning phrases also as different expressions of a person that might invoke a specific intent or entity.

Natural language processing (NLP)

A field of artificial intelligence that focuses on enabling computers to understand, interpret, and generate human language.

Intent

The goal or purpose behind a user's spoken or written message. In a customer service chatbot, the intent is often determined by mapping the user's input to one or more predefined categories. For example, Siena might identify the intent of a customer's message as "request refund," "request discount," or "get product information.”

Prompt Engineering

The practice of designing and optimizing the instructions or 'prompts' that we give to an AI model. The aim is to help the model produce the most useful and relevant responses. It's a bit like guiding a conversation in the right direction.

Hallucination

When the model generates information that wasn't in the input or the training data. In other words, it's when the AI makes stuff up. This can happen when the model is asked to fill in gaps in information and gets it wrong.

Knowledge Base

A collection of information organized in a structured way, that a conversational AI system can use to answer customer inquiries or provide assistance. For example, Siena AI uses a knowledge base to resolve tier-1 inquiries.

Entity

A specific piece of information that is relevant to the user's intent. For example, in a customer service chatbot, "order number" might be an entity representing a specific order that a customer is inquiring about.

Slot filling

The process of extracting specific pieces of information (entities) from a user's input and storing them for later use.

Dialog flow

The sequence of steps that a conversational AI system follows in response to user input.

Context

The information that the conversational AI system has gathered so far in the current dialog, which can be used to inform its responses.

Fallback

A response that the conversational AI system provides when it is unable to understand the user's input or fulfill their request.

Human-in-the-loop

A design approach in which a human analyst is involved in the decision-making process of a machine learning system, either by providing input or reviewing and approving output.

Chatbot

A software application that uses artificial intelligence to conduct a conversation with human users, typically through a chat interface.

Rule-based system

A conversational AI system that relies on a fixed set of rules to determine how to respond to user input.

Machine learning-based system

A conversational AI system that uses machine learning algorithms to improve its responses over time based on feedback from users or human analysts.

Deep learning

A type of machine learning that uses artificial neural networks with multiple layers to learn and make decisions.

Neural network

A machine learning model inspired by the structure and function of the human brain, consisting of interconnected "neurons" that can process and transmit information.

Evaluation

The process of assessing the performance of a machine learning model using metrics such as accuracy, precision, and recall.

Bias

A systematic error or deviation from the true value in a machine learning model, often due to the inclusion of biased data or inadequate representation of certain groups.

Ethical AI

The principles and practices for designing and implementing artificial intelligence systems that are fair, transparent, and respectful of human rights and dignity.

Supervised learning

A type of machine learning in which the model is trained on labeled data, with the correct output provided for each example.

Unsupervised learning

A type of machine learning in which the model is not given any labeled examples and must discover patterns and relationships in the data on its own.

Semi-supervised learning

A type of machine learning in which the model is trained on a mix of labeled and unlabeled data.

Reinforcement learning

A type of machine learning in which the model learns through trial and error, receiving rewards or punishments for its actions. Think of a feedback loop between humans and the AI system.

Self-learning

A design approach in which a conversational AI system is able to learn and improve its responses over time based on user interactions.

Dialog management

The process of controlling the flow of a conversation between a user and a conversational AI system. This can involve managing multiple turns, handling interruptions, and maintaining context.

Natural language generation (NLG)

A field of artificial intelligence that focuses on enabling computers to produce human-like language often used to generate responses in a conversational AI system.

Language model

A machine learning model that is trained to predict the likelihood of a sequence of words, often used in NLP tasks such as language translation or text generation.

Large language model (LLM)

A large language model is a machine learning model that is trained on a very large dataset of text and is able to generate human-like language or perform other natural language processing tasks. These models are typically based on neural networks, which are able to capture complex patterns and relationships in the data.

One example of a large language model is GPT-3 (Generative Pre-trained Transformer 3), which was developed by OpenAI and is currently one of the largest and most powerful language models available. GPT-3 is trained on a dataset of over 8 billion web pages and is able to perform a wide range of natural language processing tasks, including language translation, text generation, and question answering.

Other examples of large language models include BERT (Bidirectional Encoder Representations from Transformers), developed by Google, and RoBERTa (Robustly Optimized BERT Pretraining Approach), developed by Facebook. These models are also trained on large datasets and are able to perform a variety of NLP tasks.

Pre-training

The process of training a machine learning model on a large dataset in order to learn general features that can be useful for a wide range of tasks. For example, a language model might be pre-trained on a dataset of Wikipedia articles in order to learn the statistical patterns of natural language.

Fine-tuning

The process of adapting a pre-trained machine learning model to a new task or dataset by adjusting the model's parameters using additional training data. For example, a pre-trained language model might be fine-tuned on a dataset of customer service conversations in order to create a chatbot that can handle a specific domain.

Tier-1 inquiries

Basic customer inquiries that an automated system, such as Siena AI can easily and accurately answer. These questions can be related to commonly asked questions, frequently encountered problems, or simple tasks that can be easily completed using a knowledge base or other data sources.

**Example of tier-1 inquiries**

- What is the return policy for this product?
- How do I use this coupon code?
- What are the payment options available?

Tier-2 inquiries

Also known as workflow-based questions are customer inquiries that require additional data or analysis to be accurately answered. These questions may be more complex or less common and may require human intervention or more sophisticated artificial intelligence capabilities to resolve.

**Example of tier-2 inquiries**

- I will be traveling so skip my delivery this month
- My order arrived damaged, I need a replacement
- How do I cancel my order? I placed it by mistake

Tier-3 inquiries

Also known as consultative inquiries, these are customer inquiries that require a high level of expertise or personalized assistance to be accurately fulfilled. These questions may involve complex or specialized topics or may require a deep understanding of the customer's specific needs or circumstances.

Examples of tier-3 inquiries:

- Can you recommend a product that meets my specific needs and it’s between $100 and $200?
- How does this product compare to similar products regarding features or performance?
- Can you help me choose the best shipping options for my order?
- Can you provide guidance on how to use this product for a specific application or task?
- Can you provide some product recommendations based on my recent purchases?

Generally, tier-3 inquiries are more difficult for a traditional chatbot to handle accurately and may require human intervention. However, Siena is specifically designed to fulfill complex and personalized conversations using LLM.