Deciphering the Evolution of chatbots: A closer look at Microsoft Copilot.

Clients frequently inquire about the advantages and rationale for upgrading to Copilot from their existing rule-based or AI chatbots used for their customers. Also at times its difficult for organizations to decide which one is to use when?

In this article, I aim to demystify these questions and offer a comparative analysis between three primary types of bots:

At a high level, we can identify three primary types of bots in use.

1. Chatbot (Rule based, mostly single turn)
2. Chatbots with AI (Conversational Chatbot, multi turn)
3. Copilots (Generative AI, natural conversation)

Chatbots (Rules Based, mostly single turn)

• These chatbots rely on predefined rules and predetermined content, limiting their ability to generate innovative responses.?
• They operate using keywords, rules, and decision trees, providing predefined answers to specific queries.?
• Limitations include dependency on live support agents, as they struggle to understand various user queries outside their programmed scope.?

These chatbots are mainly dependent on three components.??

1. Keywords?
2. Rules??
3. Decision tree (if/then else statements)?

Limitations and dependency on Live support Agents:??

Given the impossibility for conversation designers to predict and pre-program the chatbot for every conceivable user query, rules-based chatbots often gets stuck because they can’t grasp the user’s request which often leads the chatbot to transfer the user to a live support agent.?

?Example:

Usage:

Rule based chatbot can be used in scenarios requiring standardized or system-generated responses, such as booking restaurant tables, furnishing delivery times, tracking codes for orders, FAQs and other straightforward instances.

Chatbots (AI, multi turn)

AI based chatbots refers to AI-driven communication technology such as chatbots and virtual assistants (e.g., Siri or Amazon Alexa).

• AI-driven chatbots leverage machine learning and natural language processing to detect contextual information shared by users.
• Trained on specific datasets, they self-learn and improve their knowledge base over time.
• While proficient, they have limitations in understanding complex emotions, handling intricate requests, and are reliant on accurate data for optimal performance.

The components mainly it uses:?

1. Data on which its trained (Specific Datasets),??
2. Machine learning (ML),??
3. Natural Language Processing to recognize vocal and text inputs, mimic human interactions, and facilitate conversational flow.?

1. Specified Datasets:?

These bots can be trained on specialized data sets. They will not be likely to answer questions outside their domain. Such as, If these bots are trained on data about Cars then they would be able to answer questions about cars but if user ask questions about any other topics they would not likely to answer.

2. Machine Learning:

• Machine learning is subset of AI. It allows AI bots to self-learn and develop an increasingly intelligent knowledge base of questions and responses that are based on user interactions.
• One of the most important elements that they can predict future behaviors by automation; Such as Amazon use machine learning to recommend products to a specific customer based on what they’ve looked at and bought before.

3. Natural Language Processing:

AI chatbots can understand user’s questions, no matter how they’re phrased. With AI and natural language understanding (NLP) capabilities.

• Natural Language Processing (NLP) empowers chatbots to engage with user inputs, encompassing spelling and grammatical errors.?
• Also distinguishing between intentions and questions.??
• Additionally, NLP captures emotional content and emphasis, mirroring the nuances of a face-to-face conversation.?

Limitations:

• AI chatbots lack the ability to understand complex human emotions and nuanced responses.?
• Difficulty in handling complex requests???
• Reliance on accurate data??
• only as good as the data and algorithms they're trained on, so if the data is flawed, the chatbot's responses will be too.??

• They also can't answer every question or handle every situation, so there are still limits to what they can do.?

Example:

Microsoft Copilots(Generative AI, natural conversation)?

Microsoft Copilot is based on Generative AI which uses a much wider range of Data to answer almost any question in any category. These are trained on more diverse dataset than AI chatbots.?

It creates new written, visual, and auditory content by way of existing data or input by humans. ChatGPT from OpenAI is a generative AI.?

In this article we are not going into details the Copilot design, but main focus is how generative AI makes it different and advanced than any other rule based and AI chat bots.

The key component of Copilot, as with other generative AI tools, is the??

1. NLP (defined above)?
2. LLM?
3. Deep Learning?

LLM:?

A large language model (LLM) is a type of artificial intelligence (AI) algorithm that uses deep learning techniques and massively large data sets to understand, summarize, generate and predict new content. Training LLMs to use the right data requires the use of massive, expensive server farms that act as supercomputers.

The Microsoft 365 Copilot “system” consists of Microsoft 365 apps, Microsoft Graph, which includes data across the Microsoft 365 environment; and the OpenAI LLM models that process user prompts: OpenAI’s ChatGPT-3, ChatGPT-4, DALL-E, Codex, and Embeddings. These models are all hosted on Microsoft’s Azure cloud environment.

OpenAI is the company which released their largest LLM model (GPT-3) in June, 2020 with 175 billion parameters and the company’s latest model – GPT-4 – is purported to have 1 trillion parameters.??

The development of Microsoft 365 Copilot is the result of a close collaboration between Microsoft and OpenAI.??

Deep learning:

LLMs use a type of machine learning called deep learning. Deep Learning is a subset of Machine Learning that involves training neural networks to process vast amounts of data.

• It enables the use of large data sets, enabling them to generate more personalized and relevant responses.??
• A deep learning model cannot actually conclude anything from a single sentence. But after analyzing trillions of sentences,?
• ?It could learn enough to predict how to logically finish an incomplete sentence, or even generate its own sentences.?
• As example, In the most basic sense, deep learning uses neural networks which are circuits of neurons (or artificial neurons). Neural networks solve artificial intelligence problems.

• LLMs are built on neural networks.??
• Just as the human brain is constructed of neurons that connect and send signals to each other, an artificial neural network is constructed of network nodes that connect with each other.??
• They are composed of several "layers”: an input layer, an output layer, and one or more layers in between. The layers only pass information to each other if their own outputs cross a certain threshold.?

Limitations:?

AI-produced content and outputs may contain inaccuracies, biases, or sensitive materials because they were trained on information from the internet, as well as other sources. AI may not know about recent events yet, and struggles to understand and interpret sarcasm, irony, or humor.? Please remember that it's not a person.?

It's important that you review any content the AI generates for you to make sure it has accurately produced what you wanted.?

Privacy:

Copilot for Microsoft 365 is built on Microsoft's comprehensive approach to security, compliance, and privacy. Your data (including prompts, responses, and the business data Copilot uses to formulate its response) isn’t used to train the foundation large language models (LLMs) that Copilot uses.?

Conclusion:

In conclusion, the evolution of chatbot technology has led to distinct categories: rule-based bots, AI-driven chatbots, and Generative AI bots like Microsoft Copilot. While rule-based bots operate within predefined parameters, limiting their flexibility, AI-driven chatbots show improved adaptability but struggle with complex emotions and nuanced responses.

In contrast, Microsoft Copilot, leveraging Generative AI, stands out for its ability to handle diverse queries across vast categories, thanks to its training on extensive datasets.

Microsoft Copilot's utilization of Large Language Models (LLMs) powered by Deep Learning allows it to understand, summarize, generate, and predict content with exceptional accuracy and contextuality. Unlike conventional AI bots, Copilot's vast dataset enables it to offer more personalized, relevant, and dynamic responses. However, challenges persist, such as potential inaccuracies, biases, and limitations in interpreting sarcasm or recent events.

Nevertheless, the security, compliance, and privacy measures inherent in Copilot for Microsoft 365 underscore Microsoft's commitment to safeguarding user data and ensuring responsible AI utilization. As organizations navigate the landscape of chatbot technologies, Generative AI bots like Copilot offer unparalleled potential for handling diverse and complex user queries across various domains.