Collaborada

View Original

AI Primer for SMBs

Christine Darby
See this content in the original post

Feeling overwhelmed by the increasing buzz around AI and GPT in the business world? Our AI primer for small businesses breaks down key terminology to provide you with an overview understanding of complex concepts. We provide clear definitions to help you become fluent in generative AI conversations. 

This AI guide offers a step-by-step accessible introduction to essential topics in artificial intelligence, ranging from the broad concept of AI, to the basics of machine learning (ML), down to popular generative AI tools like ChatGPT and AI search engines. Gain insights into AI fundamentals and how generative AI will transform and impact your SMB.

See this content in the original post

What is AI?

Artificial Intelligence (AI) involves developing algorithms and systems that can perform tasks that would normally require human intelligence, such as learning, planning, reasoning, perception, and decision-making. It is a broad field that draws from the fields of computer science, philosophy, psychology, linguistics, mathematics, statistics, neuroscience, and even art and design.

AI encompasses areas such as computer vision, robotics, game playing, and natural language processing, or the ability to understand and generate human language.

What is machine learning (ML)?

Machine learning (ML) is a subset of AI that gives computers the ability to learn from data without being explicitly programmed. Think of it as teaching a computer to recognize patterns, similar to how we might learn to ride a bicycle: once we understand the pattern of pedaling and balancing, we can do it again and again.

Machine learning algorithms and models can be used to build systems that can make predictions, detect patterns, and classify data. ML has numerous applications in various industries, including healthcare, finance, retail, and more. For example, machine learning can be used to predict patient outcomes, detect fraud, recommend products to customers, and optimize supply chains.

There are many different techniques in machine learning, but two major categories include:

  • Traditional Machine Learning: These techniques have been around for a while. Some examples are linear regression (used to predict a continuous outcome, like house prices based on various features of the house) or decision trees (used to classify data, like determining whether an email is spam or not). These methods often need important pieces of data to be identified in advance.

  • Deep Learning: This is a newer subset of machine learning that’s all about neural networks with several to many layers, hence the term “deep.” These neural networks are modeled after the neural networks found in the human brain. These tools are designed to analyze large amounts of data by recognizing complex patterns. Deep learning powers many AI tools, enhancing automated systems that can perform both analytical and hands-on tasks without needing humans. It can be used for a variety of tasks, including natural language processing, speech recognition, and image recognition.

The broad field of machine learning encompasses Natural Language Processing and Language Modeling as subfields.

What is NLP?

Natural language processing (NLP) is a subfield of AI that specifically focuses on developing algorithms and models that can understand, analyze, and generate human language. This is a field that blends computer science and linguistics.

Examples of NLP tasks powered by AI algorithms and systems include:

  • Text Classification: Categorizing text into predefined categories, such as sorting emails into spam and non-spam folders.

  • Sentiment Analysis: Determining the emotional tone of a piece of text, such as whether a customer review is positive or negative.

  • Named Entity Recognition: Identifying and extracting important named entities from a piece of text, such as Google’s Knowledge Graph that extracts people, places, or things from text to provide a summary.

  • Machine Translation: Translating text from one language to another, such as Google Translate which can help small businesses communicate with international customers.

  • Question Answering: Automatically generating answers to questions based on a given context, such as chatbots, or Apple’s Siri and Amazon’s Alexa.

Other examples of cloud-based NLP services include Amazon Comprehend and IBM Watson—these platforms provide sentiment analysis, entity recognition, content classification and topic modeling.

NLP focuses on the understanding and analysis of human language, and language modeling is specifically concerned with generating human-like text or predicting the next word in a sentence. Language models are one of many tools used in AI-based natural language processing.

What is a language model (LM)?

A language model (LM) is a type of computer program that has been trained on large amounts of text data to analyze, understand, and generate human language or perform other NLP tasks.

Language models can be used to perform tasks like writing, identifying grammatical errors in text, translating, or summarizing text. LMs can also be used for tasks such as generating new text based on prompts or predicting the next word in a sequence.

It’s important to note that while language models can improve in ability over time, this is done through training on large amounts of data, not through individual experiences or interactions. The tools don’t have the capability to learn or remember information from one interaction to the next.

Examples of large language models (LLM) you might be familiar with include GPT, BERT, RoBERTa, XLNet, ELECTRA, and T5.

The BERT language model was added to the Google search system in 2019, making Google better at understanding natural language and the intent behind search queries. This had an SEO impact on websites with low-quality content. If your business was already producing high-quality content for SEO relevant to your target audience, then the impact of BERT on your optimization efforts was likely minimal.

It is worth noting, up to this point the focus has been on creating large language models, but now “baby” language models are being developed—the thinking is that teaching fewer words to language models might help the tools sound even more human.

Some language models, known as generative AI, are specifically designed to generate new text based on training data. These models can be used for a variety of natural language processing tasks.

What is Generative AI? GPT-3.5 and GPT-4?

Generative AI is a specific subset of artificial intelligence focused on creating new, original content. These AI systems can produce a wide variety of outputs, such as text, images, music, and even complex entities like 3D models. A generative AI model can write text that mimics the style, tone, and structure of the training data used.

An example of generative AI is the Generative Pre-trained Transformer (GPT) large language model developed by OpenAI. The GPT-3.5 and GPT-4 language models generate highly coherent and realistic text. GPT-3.5 is an earlier model, whereas GPT-4 and GPT-4o are advanced models solving more complex problems with greater accuracy. The most robust GPT features are available via a subscription plan or the OpenAI API.

The GPT API (Application Programming Interface) allows developers to access the functionality of OpenAI’s latest GPT language model. With the API, developers can integrate GPT’s capabilities into their own applications, products, or services. This allows businesses to leverage GPT without needing to develop their own language models.

Generative AI (and ChatGPT specifically) has garnered significant attention for its capabilities, but keep in mind, while generative AI can mimic human-like text, the tools don’t understand the text in the same way humans do—the language models generate responses based on patterns learned from the training data.

What is ChatGPT?

ChatGPT is a tool based on OpenAI’s GPT language model that generates responses mimicking human-like patterns based on its the training data. Though called a “conversational bot,” the tool does not comprehend conversation in the human sense. ChatGPT brought AI into the spotlight for most small businesses.

AI has been integrated into widely-used everyday applications for a number years, including popular consumer platforms such as Siri, Alexa, Netflix, and Spotify, and common business programs such as Salesforce, Quickbooks, HubSpot, and Tableau. But the introduction of ChatGPT—a free and easy-to-use tool—made AI capabilities accessible to a much wider audience. For example, ChatGPT can help small business generate content, such as blog posts, social media captions, and product descriptions, saving time and effort in content creation.

OpenAI continues to release new GPT models. The free version of ChatGPT allows limited access to the most advanced model (currently GPT-4o), while a paid subscription to ChatGPT Plus offers access to more functionality.

Chatbots versus AI Search

After OpenAI released ChatGPT to the public, Google raced to release their own generative AI chatbot, Gemini (previously called Bard). Around the same time, Microsoft and Google added AI-powered results to their search engines, resulting in Microsoft’s Copilot (formerly Bing Chat) and Google’s AI Overviews (formerly SGE).

There is crossover in the function and use of these tools, but in general, chatbots are better for creative tasks, while AI search engines will be better suited for finding up-to-date information. That said, these distinctions continue to blur with every passing day.

Chatbots

Chatbots can be used to generate content such as stories, marketing campaigns, and even code. But the “creativity” displayed is based on patterns recognized from existing data, not genuine creativity as exhibited by humans. The tools can provide summaries or outlines of factual topics and help with more complex tasks that are difficult to explain in words, such as math problems.

The chatbot experience is designed to mimic a conversation with a human who understands questions and provides thoughtful answers. Examples include:

  • ChatGPT: The various models have different knowledge cutoffs. For example, GPT-4o training data is up to October 2023 and GPT-4 Turbo is up to December 2023.

  • Google Gemini: Previously known as Bard, Gemini was Google’s answer to ChatGPT. Gemini’s knowledge cutoff is November 2023.

The models above can now search the internet and read webpages when generating responses.

AI Search

AI-powered search engines can help users find information quickly on the web and get more comprehensive answers to questions. These search tools aim to help users discover a variety of perspectives on a topic or learn about a new product. Examples include:

  • Google AI Overviews or AIO (formerly SGE, Search Generative Experience): Powered by Gemini AI. These AI-generated summaries surface at the top of Google’s search results. Note, if you prefer the former decluttered format of simple text links, look for the new “Web” filter under the “More” tab in your search browser.

  • Microsoft Copilot (formerly Bing Chat): Powered by GPT-4.

Chatbots and AI-based search are very powerful tools, but have limitations. The tools cannot discern what is true or false in the information they process. And the tools can produce biased or inaccurate responses based on training data, so it is important to carefully evaluate AI output. That said, AI model developers are actively working on reducing biases to make AI outputs more reliable and fair.

Overall, the output quality of any generative AI tool is highly dependent on the quality of your prompting instructions.

AI and Your Small Business

The use of AI, including NLP and language models, raises ethical considerations, such as privacy, transparency, and accountability. The field of AI ethics is growing, with many organizations and researchers dedicated to creating guidelines and best practices to ensure the responsible use of AI.

Even very small businesses will want to develop guidelines that dictate how the business can use AI responsibly. For now, you want to regard generative AI output as something created by an assistant rather than an expert. So for example, if AI is used in content creation, commit to having a human review the content before publishing.

There are also legal issues surrounding the generation and use of AI content. Some creatives and publishers are blocking AI bots and taking legal action against AI companies because intellectual property and copyrighted material was used to train chatbots and AI-powered search.

AI tools are rapidly evolving with new advances, techniques, and concerns. It’s important to keep up with the latest developments to ensure your small business is using the most effective and accurate AI applications and considering AI best practices.

Additional Reading