Overview

There's a lot of buzz surrounding artificial intelligence (AI) and machine learning. Everyone wants to ride the wave of excitement and be labeled a next-gen company using top-notch technology.

However, there are many nuances to understanding what AI truly entails and what it can ultimately deliver.

We will unravel the evolution of Conversational AI throughout the ages, leading up to the present-day Goliath of Level 3 Conversational AI (L3CAI).

As we strive to embrace advanced AI technologies, it's vital to understand the foundational levels of Conversational AI, namely Levels 1 and 2. These form crucial stepping stones toward more advanced AI capabilities.

What is Conversational AI

This is a field of applied science that merges various technologies to solve real-world problems. Ultimately, we focus on the end goal: great conversations. The objective of this technology is to understand what the end user is saying, process it, and then take action. The sophistication of this technology improves incrementally with each part of this process:

  1. Understanding Input
  2. Intelligent Processing
  3. Taking an Action
  4. Generating Response

We can use these four characteristics to trace the evolution of AI over the years, noting significant improvements.

Level 1 AI: Scripted Assistants

The most basic form of Conversational AI, Level 1, is essentially a Scripted Assistant. These assistants operate based on predefined scripts and follow a linear, rule-based trajectory. They cannot handle deviations from the script or understand the user’s context beyond the current exchange.

For example, imagine an online pizza-ordering bot. If a user says "I'd like to order a pizza," the bot might ask, "What size would you like?" However, if the user asks "What are your working hours?" the bot may fail to respond appropriately. It might continue with its script, neglecting the user's query as it falls outside its scripted responses.

This generation of AI can only respond to direct commands, such a as button clicks, and follows a predefined response tree. The actions are limited to retrieving the next response in the sequence. There is no generation of responses, as they are always pre-defined.

Level 2 AI: Retrieval-based Assistants