Chat bots and ChatGPT - ready for deployment?

Simple answer - no, but let us explain why.

ChatGPT and alike have really set people’s minds racing as to the possibilities of Artificial Intelligence and how that may help businesses provide a better and faster experience to their customers.

We have offered AI assisted chat bots for years - explain your query, suggest a known knowledge base article, and hopefully answer the customer’s query.

Large Language Models (LLMs) like ChatGPT open a new frontier of not just replaying set articles to the customer but actually trying to answer their specific queries. Chat bots go from answering a question such as “Am I covered to drive in France” with a list of countries you may be covered in, to a simple definitive confirmation whether France is included or excluded.

This certainty though in the responses can lead to unintended consequences. ChatGPT and alike have a well known problem of both hallucinations and prompt injection.

Hallucinations are where these LLMs will state fiction as fact. A customer may ask about coverage for an expensive operation, and the bot could potentially confirm coverage when that’s not the case. That simply isn’t acceptable - even if it occurs 0.1% of the time.

The following article is a recent example of this where a company was sued because the chatbot gave incorrect information to a bereaved customer: https://www.cbc.ca/news/canada/british-columbia/air-canada-chatbot-lawsuit-1.7116416

Prompt injection is another danger with LLMs. Specially crafted messages or badly implemented chat bots can lead bots to swear, talk discouragingly about your own business, or even agree to sell you a car for a dollar.

There are countless examples of these instances happening, take a look a the following articles that highlight this.

Swearing chatbots: DPD error caused chatbot to swear at customer - BBC News

Bargain $1 Car: Driver uses ChatGPT hack to get dealer to agree to sell new car for $1 in ‘legally binding deal’ in blow for AI rollout | The Sun

The examples we have highlighted have no doubt caused a lot of embarrassment and reputation damage for the companies involved, but luckily so far no individuals have been caused any real harm.

But could the use of unmonitored chat bots lead to the death of an individual?

Gartner have sent out a chilling warning that GenAI will lead to the death of an individual by 2027, you can read the full article here: Gartner Warns that GenAI “Will Directly Lead to the Death of a Customer” by 2027 - CX Today

Given these dangers, we certainly aren’t comfortable putting an LLM out there unattended. QContact gives your staff the power and productivity boost of the LLM to quickly answer customer queries mails and messages, but always with the human checking the content before sending.

It’s not to say that LLMs in the future won’t be able to fully automate some level of task in the future, but right now, they’re just not ready for prime time unattended.

#AI #GenAI #AItools #customerservice

You can find out more about QContact's AI tools here:

https://www.qcontact.com/platform-overview-communication-ai/

Where you can also book a demo to see how we can help you meet your specific business needs.

https://www.qcontact.com/book-a-demo/