AI Basics

Donna AI Resource Hub

What is Artificial Intelligence

Artificial Intelligence (AI):

"In its simplest form, AI is the overarching description for technologies that use computers and software to create intelligent, humanlike behavior. If you have ever used Siri or Alexa, or conducted a Google search, you have used AI. If you have ever received recommendations for products or services based on past purchasing or browsing history, you have used AI." - [Bloomberg Law](https://pro.bloomberglaw.com/insights/technology/ai-in-legal-practice-explained/#whatAI)*

Intelligence, derived from Latin inter and Legere, refers to the ability to draw distinctions between different things, understand or comprehend oneself and the world around us. AI therefore, assumes that human intelligence can be replicated by constructing computer programs that can understand, sort, and comprehend given states of affairs.

Artificial Intelligence is a branch of computer science that aims to create systems capable of performing tasks that would normally require such intelligence. This involves learning, reasoning, problem-solving, perception, and language understanding. AI can learn somewhat like humans through a process called machine learning.

The field of AI research includes several sub-fields, such as machine learning, deep learning, natural language processing explained below.

Machine Learning (ML):

ML involves feeding data to algorithms which then use statistical techniques to predict outcomes. As more data is fed into these algorithms, they become better at predicting. This mimics the human ability to learn from experience.

An algorithm is a set of step-by-step instructions or rules to be followed in calculations or problem-solving operations, especially by a computer. It's a detailed series of actions to perform to accomplish some task or solve a problem.

Generative AI (Gen AI):

Generative AI is a subfield of artificial intelligence that involves the use of machine learning models to produce content.

"A generative AI tool (e.g. ChatGTP) generates “output,” typically in response to instructions from a user, referred to as the “input” or “prompt.” The output is based on an algorithmic model that has been trained on vast amounts of data – which could be text, images, music, computer code, or virtually any other type of content."

"What makes generative AI different from more familiar algorithm-based machine learning (ML) technology is that it draws on enormous sources to almost instantaneously create seemingly new, task-appropriate rich content: essays, blog posts, poetry, designs, images, videos, and software code." - [Bloomberg Law](https://pro.bloomberglaw.com/insights/technology/ai-in-legal-practice-explained/#whatAI).

Deep Learning (DL):

Deep learning is a subset of machine learning that operates structures known as neural networks. These neural networks are designed with numerous layers, hence the term 'deep'. This complexity and depth allow the neural networks to learn and make decisions in a way that is eerily similar to how a human brain would.

The 'deep' in deep learning isn’t a reference to any kind of deeper understanding achieved by the approach; rather, it stands for this idea of successive layers of representations. How many layers contribute to a model of data is called the depth of the model.

These layers are made of nodes. A node is just a place where computation happens, loosely patterned on a neuron in the human brain, which fires when it encounters sufficient stimuli. A node combines input from the data with a set of coefficients, or weights, that either amplify or dampen that input, thereby assigning significance to inputs for the task the algorithm is trying to learn.

Deep learning models are built using large amounts of data and can automatically learn representations from the data. This ability to use raw data makes deep learning models highly accurate and efficient. The more data they are fed, the better they perform.

Natural Language Processing (NLP):

Natural Language Processing (NLP) is a branch of Artificial Intelligence (AI) that deals with the interaction between computers and humans via natural language. It aims to read, decipher, understand, and make sense of the human language in a valuable way. This includes tasks such as translation, sentiment analysis, speech recognition, and topic segmentation.

NLP can utilize both ML and DL techniques to understand and generate human language. For instance, ML algorithms can be used in NLP for tasks like sentiment analysis and topic segmentation. On the other hand, DL—specifically Recurrent Neural Networks (RNN) and Transformer models—are commonly used for tasks like text generation, machine translation, and speech recognition. Hence, NLP is a field that intersects with both ML and DL under the umbrella of AI.

Large Language Models (LLMs):

LLMs are advanced machine learning models that have the capability to comprehend and generate human language text by analyzing massive datasets of language. They undergo training on diverse datasets, enabling them to detect patterns, relationships, and structural nuances in language.

During their training, these AI models are exposed to a variety of content, including books, articles, and myriad forms of digital texts. This exposure equips them to grasp a multitude of subjects, writing forms, and literary styles, effectively granting them a broad insight into collective human knowledge.

Fundamentally, LLMs operate as predictive tools. They assess sequences of words to infer the subsequent words, selecting the most probable continuation based on their prior learning. This prediction process is iterative, continuing until the AI constructs a fully formed concept.

The operation of LLMs can be likened to the childhood game of 'completing the sentence,' where they apply their linguistic knowledge to generate relevant and cohesive responses.

Occasionally, they may generate convincing yet inaccurate assertions. It's crucial to understand that LLMs create responses predicated on predictive logic, which is why errors can occur or the output may appear correct but lacks factual accuracy.

Retrieval-Augmentation Generation (RAG):

RAG is a technique in AI that combines a LLMs with an external knowledge base to generate more accurate and contextually relevant responses.

LLMs are often surprisingly knowledgeable about a wide range of topics but they are limited to only the data they were trained on. This means that clients looking to use LLMs with private or proprietary business information cannot use LLMs 'out of the box' to answer questions, generate correspondence, or the like.

RAG is an architectural pattern that enables foundation models to produce factually correct outputs for specialized or proprietary topics that were not part of the model's training data. By augmenting users' questions and prompts with relevant data retrieved from external data sources RAG gives the model 'new' (to the model) facts and details on which to base its response. [IBM](https://www.ibm.com/architectures/hybrid/genai-rag#:~:text=Retrieval augmented generation (RAG) is,of the model's training data.)

RAG works in two steps:

Finding Information: The AI searches through a large database to find relevant documents or pieces of information based on a specific question or query. This ensures the AI has the right data.
Creating Text: The AI then uses this information to create accurate and relevant responses.

A typical RAG system includes an LLM, a vector database (to conveniently store external data) and a series of commands or queries. When a user makes a query, the RAG enriches the prompt or question with data and facts obtained from an external database containing information relevant to the query. The RAG then sends the enriched prompt to the LLM, which is responsible for generating a natural language response based on the full power of the LLM, but also with the specific data provided in the retrieval stage.

Source

Other Terminology You May Come Across

ChatGPT: ChatGPT is a tool made by OpenAI. It uses Machine Learning (ML) to create text that sounds like it was written by a person. It's based on a technology called Generative Pretrained Transformer (GPT). This technology helps it make sentences that are not just right in grammar, but also make sense in the context.

Other alternatives to ChatGPT:

Google Gemini (formerly BARD) - Google's advanced conversational AI with real-time web info retrieval and conversation export.
Microsoft Copilot (formerly Bing Chat) - Microsoft's AI code assistant integrating with Microsoft 365 for coding autocomplete suggestions.
Claude-3 - Anthropic's highly advanced chatbot with exceptional reasoning and persuasive capabilities.

Hugging Face: Hugging Face is a platform with more than 350,000 models, 75,000 datasets and 150,000 demo applications, all open source and publicly available online where people can easily collaborate and build artificial intelligence models.

Further Sources

https://www.newsletter.theaidiscovery.com/p/what-is-generative-ai-exploring-its