Last Updated on October 7, 2020 by Sean B
The answer to the question “what is an AI chatbot?” can be a complicated one. First, as I explain in What is a Chatbot? Chatbots fall into either the rule-based or AI-powered categories.
What is a Rules-Based Chatbot?
Rules-based chatbots are also called decision-tree bots, and they follow a series of defined rules to solve a problem or answer a question. As the name suggests, every interaction these chatbots have is guided by the set of rules created, they do not learn from their interactions, which makes them limited in their application. These are great for simpler tasks, but they can’t answer questions that are outside your predefined set of rules.
What is an AI Chatbot?
An AI chatbot will typically rely on machine learning or natural language processing or NLP to understand the context and intent of a question and then formulate a response based on what they know. Some of these chatbots rely on a set of rules within their program to respond, while others can generate their own answers to fairly complicated questions using NLP. Some AI chatbots are capable of learning from their interactions. The more these chatbots are used, the more they learn and the better they become at answering questions.
When looking at the Interaction Category, all AI chatbots are Transactional Chatbots. But there are actually two types of chatbots in the Technology Category that belong to the category of AI chatbots.
Keyword Based Chatbots
Keyword chatbots work by identifying predefined keywords in a user’s query and then replying based on a set of programmed rules. They utilize AI in order to find their keywords and to help select the responses, but they are not able to learn from their interactions. These chatbots take a page from the Button-Bot playbook and mix in a little AI to make the chatbots more interactive.
Contextual chatbots use machine learning and NLP to interact with users using natural language. This is accomplished by interpreting the user input and then re-configuring it in a way that allows them to respond. I talk about this process more in the Technology section below, but it is the combination of Machine Learning and NLP that puts chatbots like Mitsuku, or Kuki, in this category.
What Technologies Are Used in an AI Chatbot?
Now let’s take a look at some of the AI technologies that are used to build these chatbots. These include processes of both Machine Learning and Natural Language Processing that are combined to create the end result. Think of these as your ingredients and the planning and specifications you write as your Recipe for creating a true AI Chatbot.
Machine learning is the ability of computer systems to learn from experience without human intervention or interference, and then use what they have learned. The algorithms used in machine learning build a mathematical model based on sample data or “training data” in order to make predictions or decisions without being programmed to.
In chatbots, machine learning refers to the ability of the chatbot to learn from the inputs is experiences, one of the ways they achieve this is through natural language processing or NLP, which I’ll discuss a little further on.
But machine learning allows the chatbot to learn from every interaction, even the negative ones, which is where Microsoft’s Tay (NSFW Link) went off the rails. The Microsoft AI Twitterbot became, in the words of The Verge “a racist asshole,” in less than a day. It began spouting racist tweets within hours of being launched and was shutdown within 16 hours. It was of course learning from the worst of the interactions that Twitter trolls were throwing at it, but Microsoft should have better understood this possibility before launching Tay. They launched Zo a “politically correct” version of their chatbot later, and due to the over-correction, they made it seems as bad as Tay.
But if managed correctly, and fed a solid diet of training data, machine learning can help you create a well-rounded chatbot that is a model “employee” that sounds a little more human and not anything like Tay or Zo.
Deep learning is a type of machine learning that uses layered algorithms called an artificial neural network or a deep neural network. Instead of focusing on task-specific algorithms, deep learning involves techniques in which the system discovers representations of the data to allow it to make sense of that data.
Deep learning with neural networks can be supervised, unsupervised, or somewhere in between and involves multiple layers that are used to extract higher-level features from the raw input. A good example of this is an image processing application that is trying to decide whether pictures are of an animal or a building. Lower layers might look for outlines and edges while higher layers might look for details like faces or windows.
Natural Language Processing
Natural language processing or NLP is a field within linguistics, computer science, and artificial intelligence that deals with the interactions between computers and humans, or more precisely it deals with the interactions between computers and human or “natural” languages. The primary goal is to learn how to create programs or algorithms that can process and analyze large amounts of natural language data.
A simpler way of explaining it is that NLP helps computers understand, interpret, manipulate, and then ultimately respond to human languages as they are spoken or written.
The steps for handling this process are easy:
- Input: A user types something into a chatbot (the input),
- Interpret: The chatbot applies the algorithms for NLP to turn the data into something it can understand.
- Analyze: Then it analyzes the translation to obtain meaning of the input.
- Manipulate: Depending on what the input was, the program may use part of the input to help formulate a response of find an answer to a question.
- Respond: The chatbot posts a response to the user.
As I said, the steps for NLP are easy, but the actual process itself is incredibly involved.
NLP requires the chatbot to analyze the syntax and the semantics behind everything that is entered to truly make it a perfect translation for the computer. Syntax Analysis checks the text against the standard grammar rules to help the computer understand what is being said by identifying the parts of speech. Semantic Analysis is more difficult and requires the program to decide what elements of the input are the most important to pull the real intent behind the input out. The problem is that often what is said is not what is meant, which can be common in slang and in any more poetic use of the language. The easiest example is when you say someone is “hot” you’re not usually talking about their temperature.
Predictive analytics includes the use of different statistical techniques from data mining, predictive modeling, and machine learning to analyze current and historical facts and make predictions about future or unknown events. In business, predictive analytics exploits patterns in transactions and other data to identify areas of risk and opportunity.
This area of computer science is being applied to a variety of different areas including financial markets, healthcare, insurance, telecommunications, retail, travel, and even social networking. Right now the most interesting use of it is in the development of new vaccine candidates for the Coronavirus.
Sentiment analytics is also known as opinion mining or emotion AI. It refers to the use of NLP, text analysis, and biometrics to identify, extract, and study the emotional states of individuals by examining their text data. These tools are used to help a business understand how their customers feel about the products and brand.
Where Are AI Chatbots Used
AI chatbots are used for a wide variety of applications. There are AI chatbots designed for companies handling everything from Sales and Marketing, to Human Resources and Financial applications. Industries in the FInancial Sector all the way to Hospitals are finding new ways to use chatbots every day.
One of the more interesting AI chatbots in use right now is WoeBot, a therapy chatbot that was created by Stanford trained Psychologists. WoeBot isn’t intended to replace human therapists, but to act as a supplement to direct therapy.
So, what is an AI chatbot? They are a mix of complicated ingredients from the Technology Category that you use to build your chatbot. These are combined using the plans you made based on the chatbots place in the Purpose Category.
AI chatbots have the potential to be great tools for businesses, governments, and other organizations to interact with their customers. With the addition of advanced algorithms including NLP, Predictive Analytics and Sentiment Analytics they become even more powerful allowing them to pull hidden details out from each sentence typed to a chatbot.
The addition of Machine Learning allows the most advanced AI chatbots to learn from each interaction with a user, making them more effective the more they learn.
But you have to be careful when you deploy an AI chatbot not to let it develop along the wrong path. Remember the lesson of Microsoft’s Tay and spend the time to train your chatbot before releasing it into the wild, and once you’ve set it loose make sure you monitor it closely so it doesn’t pick up any bad habits.