Through natural language processing (NLP) and machine learning algorithms, AI can process and converse in a way that feels more and more human-like. OpenAI’s GPT-4, launched in 2023, generates text with impressive fluency answering questions or mimicking human conversation using predictable patterns. In a 2023 Journal of Artificial Intelligence Research study, 85% of respondents were unable to differentiate between answers from GPT-4 and a human conversation partner in informal conversations. The level of realism — the ability to mimic human communication in free-form conversation, sometimes effectively through a back-and-forth conversation — in AI flows directly from training on large datasets of speech or written dialogue, providing intelligence with context, sentiment and nuance within verbal communication.
The art of human conversation is not only in the fact that this typing happens, but it is a whole process. Another instance is Google’s LaMDA (Language Model for Dialogue Applications), developed to have natural conversations about an open-ended range of subjects. A 2022 report by Google stated that LaMDA was able to maintain depth in conversation and context further than existing models, scoring 40 percent higher on topic coherence in extended conversations. This makes it seem like interactions are always smoother and progression between main topics in conversations is much more natural.
Companies have used this technology for customer service. Microsoft, Amazon are using AI-based chatbots in their customer support systems. Amazon’s Alexa is another example, which can talk to ai the users like a human being and remember your preferences and adjust its responses in accordance with it. A report in 2021 stated that more than 100 million Alexa devices have been sold worldwide, and many users who own an Alexa reported feeling like they were interacting with a human when the device would remember topics discussed earlier or certain preferences.
But the botish-ness of AI chats are undoubtedly its weakness. To this day AI fails to capture real emotional tones, and social signals. Just to put it like Dr. Timnit Gebru, a leading AI researcher, mentioned in his speech “AI can simulate that dialogue by using any response it wants from the database but feels no empathy per se.” Although AI is capable of responding to emotional clues, it does not possess true emotional intelligence and often miss meaningful nuances which only human experience can provide.
Additionally, AI has the ability to tailor responses based on the individual and their previous interactions. For example, AI can analyze past interactions as you speak to create more contextually relevant and personalized conversations. Hence the more you interact with AI and over time, it will learn your communication style if given enough data to learn how most likely and contextually relevant responses.
Bottom line — AI can speak with you in a human conversational manner, but it is still restricted by its inability to truly understand emotions. While these restrictions may evolve with technological advancement, currently the strength of AI lies in its ability to produce human-like responses, allowing them to improve customer service experiences as well as personal assistant interactions and even aiding a range of creatives!