#TECH
4 MINUTES
Chatbots and artificial intelligence,
what if we made the difference?
You’ve probably already come face to face with one of those little digital assistants that nestle in websites or Facebook Messenger (and other instant messaging systems) to converse with you. Known as “chatbots”, or “webchats” (when they’re installed on websites), they enable users to chat instantly by answering questions or clicking buttons. But how do they work? What type of artificial intelligence (AI) is used in chatbots?
What is artificial intelligence?
The ubiquitous use of the term artificial intelligence blurs our understanding of the real contours and possibilities of this technology. A simple definition of artificial intelligence is “computer programs and techniques capable of simulating certain traits of human intelligence”.
In concrete terms, there are two main mechanisms that enable a chatbot to mimic a conversation with its users.
Rule-based chatbots: "less is more"
The most frequently encountered chatbot system is the rule-based chatbot. This involves setting up a conversation scenario or “decision tree” that mimics a discussion, through which the user can navigate, notably by clicking on buttons. These chatbots are often considered the basic level of artificial intelligence, as they are developed according to a simple process. Contrary to popular belief, the most intelligent chatbots are not necessarily the most technically advanced.
Depending on the usage scenario and functionalities mobilized, a decision tree can sometimes suffice, provided it delivers quality content and experience. This is particularly true of entertainment formats such as games. Their architecture is based on a decision tree that allows users to follow a path or answer questions based on the quiz principle.
For the “Le Monde Nouveau de Charlotte Perriand” exhibition, for example, a “Create with Perriand” game module was added to Twelvy, the Fondation Louis Vuitton chatbot.
Let's talk about natural language recognition
The second mechanism involves automatic natural language processing (NLP). It enables users to interact with a chatbot by sending it sentences written in free text. These are then subjected to more or less complex processing to understand the user’s query.
It is possible to extract entities from the sentence (name, value, date…), but also to try to understand the user’s intention or to interpret his feelings (positive, negative or neutral) using the formulation of his message. The notion of learning comes into play here insofar as the artificial intelligence has learned to interpret user requests from a dataset of training sentences. It performs this action while continuing to train and improve as the queries come in.
In fact, natural language processing comes under the heading of machine learning, a subset of artificial intelligence often referred to when talking about AI, as Yann Lecun reminds us.
It’s a good idea to mobilize automatic natural language processing for all chatbots that offer practical information to their users. In fact, web users are more likely to be satisfied with receiving immediate, precise answers to their questions, rather than navigating blindly through the tree structure of a website.
In fact, this is the chatbot’s main advantage over the website. Note that additional buttons can also point users in the direction of information they can obtain from the chatbot. According to our observations, users are split on average 50/50 between those who prefer to click on buttons and let themselves be guided, and those who prefer to ask their questions directly by writing a free text.
In parallel, other types of artificial intelligence can be mobilized depending on the chatbot’s functionalities, such as visual recognition to identify photos or points of interest shared by web users in conversations.
At Ask Mona, technology is definitely not an end in itself, but rather a means to an end. We are keen to mobilize the most appropriate technologies according to the needs of each cultural institution. That’s why we’re constantly coming up with new projects. Tell us what yours is!