Beyond Chatbots: Can AI Handle Real Conversations Like Humans?

AI will certainly remain an astonishing conversational machine; it will aid us, provide us answers, and even entertain us.

There remain notable discrepancies between how AI ‘talks’ and how humans communicate with one another.

Source : Getty

By Vinayak Aggarwal

We’ve all used those digital aides that are chatbots made for automating customer service, providing information about store hours, and even ordering a pizza. They are excellent at accomplishing a few tasks by rote reciting a script and factually responding to queries, but they can get a little too, let’s say, machine-like. While helpful, these interactions are more robotic than functional and certainly miss the chance of being in-depth like a truly human exchange. 

Now, think about a discussion with a friend over coffee – the small details, the shared meaning, perhaps a hint of sarcasm, and the easy flow of conversation. So the question remains: can AI ever replicate that kind of experience fully? 

For quite some time, it seemed like AI was restricted to basic bots with question-and-answer functionality. Their operation was based on rigid protocols and their ability to recognise certain words. If we deviated from their rigid framework and asked something even remotely phrased differently, the most common response would have to be, “Sorry, I didn’t get that.” 

These AIs did not talk because they did not have the skill to track context from earlier portions of the conversation. The same problem — remembering complex features of human language, like turn-taking — astounds these primitive AI systems.

The Game Changer: Large Language Models (LLMs)

The arrival of large language models (LLMs) has drastically changed the field of conversational AI. New models utilise colossal text and code databases available over the internet for training. The integrated training empowers the models with the capability to recognise language patterns, including grammar, contextual logic, and various writing and speech styles.

This advancement enabled AI to accomplish tasks that were previously considered beyond its reach. These models could carry the context of a conversation, including previous turns, and refer back to them appropriately. They gained some level of grasp of context, being able to identify phrases that are used idiomatically, produce poems and code, and sometimes even use humour or shift styles. However, understanding nuances such as sarcasm is still a significant challenge. 

They also began generating text that made sense, stayed on topic, and had a logical structure, making the conversation a lot less artificial. They could also elaborate on various topics using the information in their training data and answer difficult questions in a normally unreachable manner. This progress gave birth to AI that can draft emails, write articles like this one, perform concept construction, translate texts, and hold much more dynamic and human-like conversations.

The Cracks in the AI Conversation: What’s Missing?

However, despite these advancements, there remain notable discrepancies between how AI ‘talks’ and how humans communicate with one another. One important difference is the absence of real comprehension or sentience. 

Despite all the sophistication, LLMs are advanced pattern recognition devices. They are merely words strung together and separated by spaces, waiting to be transformed into the most likely combination of sentences, based on the input they receive and the enormous information they’ve already processed. Their understanding of truth, ideas, and feelings is not even remotely like a human’s. 

They have no personal convictions, purposes, or backgrounds. If an AI says that it properly “understands” or “feels” something, it isn’t really feeling those things; it is merely repeating statements based on the assumptions it has made after encountering such phrases.

The Grounding Problem: Can AI Understand Like Humans?

One key barrier is referred to as the grounding problem. Human language is a byproduct of social interaction, sensory input, and a plethora of spatial experiences. We intuitively know the feeling associated with ‘cold’, the colour ‘red’, and the social meaning a frown embodies. 

On the other hand, AI models are designed to learn language through text alone, fully devoid of a social and physical context. Such a restriction severely hinders their capacity to understand the logic behind common sense, as well as the underlying assumptions people make in social interactions.

AI and Empathy: Can Machines Feel?

In addition, the ability to have genuine empathy and interpret social interactions is another major problem. AI tends to recognise patterns in text that correspond with different emotions and express themselves in a perceived empathetic way, but AI does not feel empathy. Authentic conversations among people are grounded in emotion and the ability to read subtle cues like tone of voice and expression (which are usually not present in text). Building real rapport establishes a connection that goes beyond the AI. AI can imitate these aspects, but people connect on a deeper level.

Shortly, AI will certainly remain an astonishing conversational machine; it will aid us, provide us answers, and even entertain us. However, speaking to a real person is still unique and profoundly personal. Their accent and the way they speak are something far more beautiful than art. As a deeply personal experience, the search for innovative AI technology will relentlessly keep developing; however, we mustn't be confused by an intelligent duplicate and the authentic touch of two humans interacting.

(The author is the Founder & CEO, BiteSpeed)

Disclaimer: The opinions, beliefs, and views expressed by the various authors and forum participants on this website are personal and do not reflect the opinions, beliefs, and views of ABP Network Pvt. Ltd.