Chatbots Are Not AI

Conditional code execution is a key concept in programming, tracing back to the inception of programming as we know it today. If you're not a programmer and that didn't make sense, it means we can instruct a computer to perform actions based on certain conditions. This has been a core feature since the inception of computers.

Chatbots, in their most basic form, are computer programs that utilise conditional logic. They're limited in what they can reply to, which is why they often don't let you type anything unless necessary; instead, they offer a set of response options to choose from.

For users, these chatbots can sometimes feel like a time-waster compared to filling out a form, which might be quicker. However, it's not the fault of the marketing or sales teams who push for their development. The reality is that chatbots often keep people more engaged than forms do, leading to higher conversions, which essentially means more profit for the business.

The Misconception: "It Talks to Me, So It Must Be AI!"

Chatbots, with their ability to simulate human conversation, often lead users to believe they possess genuine understanding and intelligence. The interactive nature of chatbots, where they respond to user queries and guide them through processes, makes them feel "smart." Marketing materials frequently use the term "AI chatbot" as a catch-all, further blurring the lines between simple scripted bots and those powered by more advanced AI.

Peeling Back the Layers – What It Is: Pattern Matching and Decision Trees

Many basic chatbots, particularly older versions or those designed for simple tasks, operate primarily on rule-based logic, keyword matching, and decision trees. These systems are programmed with a set of predefined responses linked to specific user inputs or conversational paths. For instance, a significant number of chatbots are essentially automated FAQ responders, pulling answers from a company's knowledge base or using word recognition; they function as a database lookup rather than demonstrating cognitive understanding. Some rule-based chatbots don't even allow free text, instead offering users a series of buttons or "click strings" to navigate the conversation.

How It Works (The Nitty-Gritty):

The mechanics of basic chatbots can be broken down as follows:

Keyword Matching: The chatbot scans the user's input for specific words. If a recognised keyword is found, it triggers a corresponding pre-programmed response. For example, if a user types "opening hours," the bot is programmed to retrieve and display the store's operating hours.
Decisions (Flow-based): The conversation follows a branching structure, much like a flowchart. The user's responses to questions or the user's selection from a menu of options guide the chatbot down a specific path in the decision tree, leading to a relevant answer or action. An example could be a bot asking, "Are you a new or existing customer?" and then, based on the answer, "What product are you interested in?"

These basic systems often struggle with variations in language, such as typos, synonyms, or complex sentence structures. If a query falls outside their predefined script or uses unrecognised keywords, they may provide a generic, irrelevant response or state that they don't understand. Their effectiveness is not reliably reliant on how comprehensively their rules and keyword lists have been designed.

When I worked on a project that aimed to be a lightweight alternative to Intercom a few years ago, the “chatbot” we had built at the time was essentially a configurable graph-like schema, with each node representing either a user action or a chatbot response. Each node contained information about the next node based on the user’s selection. You could also add or modify “edges” for any of the nodes, which was a fancy way of storing another linked node.

This is, for the most part, a representation of how most chatbots were implemented.

Why It's Not (Necessarily) AI: Lack of True Intention and Learning

The fundamental reason basic chatbots are not considered true AI is their limited ability to comprehend genuine language and their inability to learn. They do not understand the meaning or intent behind a user's words in the same way a human does; they merely match patterns or follow scripted paths. Furthermore, these basic bots do not learn from their interactions to improve their responses or handle novel queries over time unless a human developer explicitly reprograms them. Their perceived intelligence is an illusion crafted through careful scripting and conversation design, not an emergent property of cognitive ability.

The evolution of chatbot technology from these simpler rule-based systems to more sophisticated models powered by Natural Language Processing (NLP) and Machine Learning (ML) has created a spectrum of capabilities. This progression can make it easier for vendors of simpler systems to benefit from the marketing buzz around advanced AI chatbots, contributing to widespread mislabeling.

The AI Connection: NLP and ML-Powered Chatbots

In contrast to their basic counterparts, advanced chatbots genuinely leverage AI. They employ Natural Language Processing (NLP) to parse and understand the nuances of human language, including intent, sentiment, and context. Machine Learning (ML) models, often trained on vast datasets of conversations, enable these AI chatbots to generate more human-like, relevant, and dynamic responses. Some modern AI agents are even pre-trained on extensive customer experience (CX) interactions, allowing them to anticipate customer needs, resolve complex inquiries, and integrate seamlessly with backend systems for personalised support. These AI-driven systems can continually gain insights from each engagement, improving their performance over time.

The limitations of rule-based chatbots, such as their poor handling of ambiguity, inability to learn, and the challenges of scaling complex scripts, were significant factors that fueled the research and adoption of NLP and ML techniques. Businesses and users demanded more robust, flexible, and genuinely intelligent conversational agents, leading to the development of the AI-powered chatbots we see today.

Today's customer service chatbots employ a surprising hybrid approach. They still utilise a graph-like structure with interconnected nodes based on specific conditions for their responses and actions, yet they no longer demand that users make selections. Instead, these chatbots can discern user intent through Natural Language Processing (NLP), allowing them to internally keep the pre-programmed rule-based node model to proceed effectively with user needs.

However, this progress also means that if users encounter a basic, scripted bot mislabeled as "AI" and it fails to understand a slightly complex query, they might become disillusioned with "AI" in general, not realising they were interacting with a much simpler form of technology.

Wrapping Up

Some advanced chatbots, though powered by sophisticated tools and AI capabilities, still rely on a 'human-in-the-loop' approach to handle complex or unstructured queries. While these bots imitate intelligent conversation, they often falter when user’ needs fall outside of their predefined workflows. As a result, users can become trapped in endless conversational loops, where the bot neither escalates the issue nor resolves it satisfactorily, leading to frustration and an incomplete experience.

However, that’s the reality we currently inhabit, and it’s essential to comprehend the reality of products claiming to offer tomorrow’s AI today, as we anticipate genuinely innovative advancements and the implementation of this novel technology.

Thanks for reading! We’d love to know your thoughts on the topic. Consider subscribing to our newsletter to be the first one to know when we publish a new article.

If you’d like to support our work further, consider buying us a coffee.