Why ChatGPT will not replace Customer Service Bots anytime soon!

I've recently seen a lot of articles on how ChatGPT will revolutionize Customer Service Bots. Fact is that transformer models, which ChatGPT is just one of many, have had a role in Customer Service Bots since late 2018. As such, for current use cases of transformers, ChatGPT is but a linear progression. The big misconception, however, is that ChatGPT will replace current approaches to building Customer Service Bots. I have my doubt that it will do that on the short term.

Customer Service Bots are representatives of the business. To be a good representative, it needs knowledge of that business, it needs knowledge of the processes, of the topics that users will ask... and it needs knowledge of the users themselves. If I ask ChatGPT, "I want to cancel my credit card", then the answer is going to be quite general. If I ask my bank's customer service bot, at a minimum I will get an answer of how to do that - contextual to the bank I am actually cancelling my credit card with. A really good customer service bot however, would know what credit cards I had, the process involved and how to navigate that process. At the end, my goal of cancelling the credit card might even be accomplished in the conversation with the bot.

How does it know the user, the topics, the processes? Well, it doesn't. The designers of the conversation research this, and ultimately cleverly design conversations specifically for that company. Just as the understanding of the user is a human designers' understanding of the user that is captured in clever design of the dialogue.

The big change for Customer Service Bots are the shift from using clever design - that imbues the understanding of the designer into the dialogue - to using AI models to create an internal representation of the users context, goals, beliefs and values. This is far beyond intent categorization. And using AI models to generate dialogue based on an AI understanding of the business, processes, topics and the users. This move from designed to generated interactions requires this deep understanding of both the business and user to work.

How does this relate to ChatGPT? It does because ChatGPT and other transformers does not seek to model the user, it seeks just to predict what topical knowledge you want and how you want it. It's a vastly impressive prediction model that tries to predict exactly what you want to hear. Many times in customer service however, you need to tell the customer things they might not want to hear, but that they need to hear.

What ChatGPT does show - is that it's within the realm of possibility to actually go from cleverly designed scripted dialogue to autogeneration. But to do that, more advanced and sophisticated approaches are necessary - approaches that take into account a lot more than what a user wants to hear. The model above shows the deep knowledge that needs to be mapped.

Models like ChatGPT will have an impact on Customer Service Bots, just like other transformers before it. But it will not replace them. The greatest thing it brings however is proof that building bots that deeply understands the users and are able to help them without scripted responses is well within the realm of possibility!