*

by Charles Miller

As promised last week, I will offer this week an overview of the different categories of AI Chatbots. Yes, there are different categories of AI Chatbots. Further; it should come as no surprise that there is some disagreement as to how many categories there are. Some technical articles I have read list a half-dozen while other papers manage to identify a full dozen distinctly different categories. In this column I am going to go with a lower number to avoid the risk of every reader's eyes glazing over before reading to the end.

Generative AI Chatbots are programmed to generate responses using predefined models. They are capable of dynamic, context-sensitive, and human-like conversations which includes being able to adapt style to user inputs. Examples include ChatGPT and other GPT-based web sites.

Menu and/or Button-Based Chatbots generally do not accept questions typed by the user but rather guide users through predefined menus or buttons used to select options. This type of chatbot limits conversational flexibility in favor of using a simple-to-use experience where only a limited number of responses are available.

Rule-Based Chatbots are similar to but more advanced than the above. They operate on limited and predefined rules and decision trees. This type of chatbot is suitable for simple, repetitive tasks like basic customer support but do not handle complex queries well.

Voice Chatbots are those that interact with the user through spoken language using speech-to-text and text-to-speech technologies. Examples include Amazon's Alexa, Apple's Siri, Microsoft's Cortina as well as most times you phone customer support.

Keyword Recognition-Based Chatbots rely on specific keywords in user input to trigger relevant responses. This class of chatbot is useful in environments with specialized terminology or controlled vocabularies. Many medical-related chatbots fall into this category because doctors usually know the exact medical terms they want to search for while laypeople not familiar with Latin or medical terms would have difficulty.

Hybrid Chatbots combine rule-based logic with AI capabilities, meaning they are programmed to use pre-defined rules for responding to straightforward queries, but can use AI for more complex or ambiguous questions.

Before I get to (my) final AI Chatbot category I need to recount how completely gobsmacked I was two years ago to listen to one of my tech podcasts and hear one of the "experts" respond to the question of whether or not AI was able to learn from its interactions with users and improve itself. He condescendingly responded "No! You can't have every idiot with a modem making changes to the knowledge base!" That led to a long conversation about how AI Chatbots are not designed for or able to learn from their mistakes. Just like all other software, AI Chatbots are programmed to perform in the way their engineer-creators designed them and though they seem quite interactive, what they do is more accurately described as mimicry. Moreover, all of them are subject to the biases of the programmer-gatekeepers who curate the databases.

AI-Powered Chatbots are different, and are what most people incorrectly assume all AI Chatbots are. In reality almost all of the previously-described categories of chatbots are really just skilled regurgitators that never learn anything new until new information is added by the human programmers who curate the databases and algorithms. AI-Powered Chatbots attempt to understand user intent, context, and phrasing variations to improve responses over time by learning from interactions, and make certain changes to their databases autonomously. In other words, they can learn.

Auto-GPT (maybe that is another category?) can write and run its own software, effectively functioning autonomously. This is a category of AI Chatbot still in its infancy and for now mostly found in academic laboratories and the new-product development departments of many software companies. Sooner or later, perhaps even by the time I finish writing this, you might see a new web site offering, Auto-GPT Chatbot, so keep your eyes open for that one day soon.

**************

Charles Miller is a freelance computer consultant with decades of IT experience and a Texan with a lifetime love for Mexico. The opinions expressed are his own. He may be contacted at 415-101-8528 or email FAQ8 (at) SMAguru.com.

Discover Lokkal:
Watch the two-minute video below.
Then, just below that, scroll down SMA's Community Wall.
Mission

Visit SMA's Social Network

Contact / Contactar