“ChatGPT does not contribute to CX for the time being”

Understanding customer demand is the starting point for effective service delivery, whether delivered by humans or automated systems. The question that concerns many: what is the significance of ChatGPT3 for customer contact?

“A lot has been said and written about ChatGPT3. The solution at the moment, however much the output can sometimes appeal to the imagination, is primarily a website where you can interact with a writing artificial assistant,” says Sander Hesselink from Telecats. “That assistant is getting better and better by adding new data. But in all the messages about ChatGPT3, a few essential features are underexposed.”

Also read: Using ChatGPT for customer contact: promising or hype?

Parrot

In the first place, according to Hesselink, the dataset with which ChatGPT3 has been working so far is always one year old. “That means that new insights are not included as a source for the output.”

Second, ChatGPT3 is what scientists call a "stochastic parrot': a system that 'parrots' words and characters back based on what it has previously seen in texts. Hesselink: “ChatGPT3 makes no connection with meaning. The text that rolls out on the basis of an assignment is created on the basis of probability. The system itself does not understand.”

Thirdly, more and more content from AI applications such as ChatGPT3 is making its way back to the internet. Hesselink: “That is not an enrichment, but an expansion of sources that may be unreliable. Because certain texts or data occur more often, the chance that other AI systems will mistakenly see them as correct increases. AI solutions are trained in this way with increasingly uncontrolled sources, which we know can contain many untruths.”

Customer contact applications

What can we – in customer contact – do with solutions such as ChatGPT3? Hesselink presents a list of possible applications for, as he says, “the periphery of customer contact”. “Think of a command like 'compile ten different messages telling the customer that a package will be delivered later' – something you could also use for chatbot dialogues. Or formulate a letter of apology, write a text for a marketing message or make instructions for installing a device.”

The advantage of this type of assignment is that the output will in any case come out with one specific tone of voice, which you can also describe in advance in your assignment. On the other hand, says Hesselink, most companies probably already have these kinds of texts available.

"AI assist"

As far as Hesselink is concerned, a second area of application is supporting customer contact employees in their work. “ChatGPT can help you on your way to articulate a topic or to rewrite texts. But I fear that if we start using robots for this, which produce new language for you like a calculator, employees will skip the step to understand certain material and really master it. That learning process – call it wrestling with difficult material – actually leads to personal growth and development. With that 'language calculator' we put this at stake.”

Personal self-service?

Another obvious solution that ChatGPT could provide for contact centers is in the area of personal self-service. Hesselink does not think it is likely that this will soon become a reality: “For services that go beyond answering general questions, you often need customer data. ChatGPT3 so far only works with open data from sources that are freely available on the internet. It seems to me that ChatGPT3 is not intended to be provided with an instruction that also contains all kinds of customer data, in order to better serve the customer. Not only is it virtually impossible to provide ChatGPT3 with frameworks so that a GDRP-compliant answer comes out, rules such as the GDPR also apply to the use of customer data. We do not know where exactly the information you provide to ChatGPT3 is stored. I would not be surprised if privacy breaches are already being committed because people provide personal data to ChatGPT3.”

Ready to see how Snowflake works?

The conclusion? When it comes to better serving customers through the use of data – as is intended with really good chatbots, for example – ChatGPT3 is not going to help with that now. “If we want to use a tool like ChatGPT for targeted customer contact, we will first have to solve a number of problems with regard to reliability and data protection.”

Many different uses of ChatGPT3 have already been developed and applied. Ziptone highlights a few.
And what do Seth Godin and Sander Duivestein say about ChatGPT?

DoNotPay, a company that automates administrative tasks, including disputing parking fines and filing for airline compensation, this week launched a video released of a chatbot that negotiates the price of Internet services on behalf of a customer. The negotiator bot is built on the AI technology that powers ChatGPT. Complaining about poor Internet service, the bot parries the points a Comcast agent made in an online chat, successfully negotiating a discount worth $120 a year.

Koko, a mental health mobile app, tested AI-generated responses without notifying recipients first. Koko founder Rob Morris revealed on Twitter that 4000 people had received counseling using ChatGPT3. According to Morris, messages to clients originating from ChatGPT3 were rated significantly higher than messages written by humans. Response times dropped by 50% to well under a minute. Although ChatGPT3's output was monitored by contributors, user responses were ultimately mostly negative.

Other uses include connecting ChatGPT3 to Whatsappp or generating emails en Twitter messages. ChatGPT3 from OpenAI is not the only application under development. BlenderBot3 of Meta is also freely accessible, and Google is also working on a large language model called LaMDA.

Seth Godin will say too (Internet entrepreneur, speaker and author) is extremely critical of this ChatGPT. “Of course this changes things, just like the camera, the typewriter and the internet changed things. It means making high volumes of mediocre material is easier than ever before.” Godin warns that attention and trust are not scalable. “Technology starts by making old work easier, but then requires new work to be better.”

Sander Duivestein (affiliated with the Exploration Institute for New Technology (VINT), part of Sogeti, among others) is going a step further and wants to ban ChatGPT because of the risks. “The great danger is that the machine fabrications are taken for truth, because the results seem so lifelike.” A machine that is better than humans at spreading misinformation should not be admired, but abhorred and even forbidden, says Duivestein.