AI Chatbots – Are They Really Safe on Corporate Websites?

• Customer service chatbots are typically ‘rule-based’ rather than AI-powered
• But this is changing as some companies integrate ChatGPT.
• This is dangerous as they are still susceptible to prejudices and hallucinations

Chatbots are becoming more and more ubiquitous in 21st century life. When you visit an e-commerce website, don’t be surprised if a friendly chatbot pops up in the bottom right corner and you hear a “pudding” sound asking if there’s anything you can do to help.

Over the last few years they have become more and more popular as customer support agents due to their obvious advantages. They are available 24/7 for him, ready to answer simple queries at any time, giving human agents time to deal with more complex queries. It also encourages consumer interaction and, if it provides a positive experience, so does his SEO of the site.

The rise of ChatGPT has caused the term AI to be thrown around wildly when discussing chatbots, but the vast majority of customer service representatives don’t actually rely on the technology. Instead, they are “rule-based” chatbots that simply operate on “if/then” decision tree logic when given various prompts. Therefore, the conversations they can hold are strictly limited to purpose-based code.

Chatbots: the new generation

Real AI-powered chatbots behave differently. It utilizes advanced natural language processing (NLP) techniques and machine learning algorithms to understand and generate human-like responses. These “generative AI” chatbots have been trained on vast amounts of data and are able to analyze and interpret the context and intent behind user queries, allowing them to provide more nuanced and personalized responses. increase. While more advanced than rule-based chatbots, they require more time and financial investment to set up.

Since the release of OpenAI’s ChatGPT in late November last year, there has been a significant increase in interest in this type of chatbot. The so-called “gold rush” of generative AI adoption that followed the advent of ChatGPT has led many companies to consider integrating generative AI into their chatbots to make them more advanced. I was.

A Gartner report this year found that just 8% of customers used a chatbot during their most recent customer service interaction, and only a quarter of those said they would use a chatbot again. Such a change is justified by what has become clear. This is mainly due to the low resolution rate for many features, such as only 17 percent for billing disputes, 19 percent for product information, and 25 percent for complaints.

Meta, Canva and Shopify are some of the companies that have started to implement large-scale language models with things like ChatGPT on their service bots, but the decision is not without risk. ChatGPT is generative AI. That is, in theory, create a new sentence in response to every prompt. This kind of unpredictability is often undesirable for brands as it creates a certain element of risk.

“Hallucination” in this context refers to AI confidently stating false information as fact. This is not a problem when creating static content that can be proofread before publishing anywhere, i.e. when there is a knowledgeable human between the AI ​​and the consumer. However, this is not the case if the bot is having a real-time conversation with the customer.

A famous example of this is when Bard, an AI chatbot developed by Google’s parent company Alphabet, incorrectly answered a question about the James Webb Space Telescope in an ad. Although there was no chatting with potential customers or anything, the error wiped out $100 billion in Alphabet’s stock.

However, the technology has evolved in a short period of time since AI chatbots became mainstream. As such, some companies believe he can safely use his GPT-4, a more advanced version of ChatGPT, on his website. Intercom, which provides a customer service platform, is one of them.

“We had an early look at GPT-4 and were immediately impressed with its increased protection against hallucinations and more advanced natural language capabilities,” said Fergal Reid, senior director of machine learning at Intercom. he told econsultancy.com. “We felt we had pushed the limits of what this technology could be used for in front of our customers.”

Digital experience software provider Yext sources data for AI-powered chatbots from CMSs that store information about a customer’s brand when answering questions, helping prevent hallucinations.

But apparently powerful chatbots with extensions can pose their own problems. Michael Rendelman, customer service research specialist at Gartner, says these deployments will “confuse customers about what chatbots can and can’t do.”

He added, “It’s the service and support leader’s responsibility to direct customers to a chatbot when the issue is appropriate, or to another channel when another channel is more appropriate.” added.

A gateway to frustration and data vulnerability?

In fact, they can lead to: increased Instead of reducing customer dissatisfaction, site visitors fresh from conversations with ChatGPT or another bot that has similarly made a significant financial investment are bolstered with a certain level of expectations for chatbot functionality in general. will be

If the bots on your site do not meet the same level of competence, anger and frustration can arise, and the bots are ill-equipped to deal with them. A 2021 study found that “chatbot personification negatively impacts customer satisfaction, overall company ratings, and subsequent purchase intent” when customers are already angry.

In fact, a very real customer service chatbot opens the door to fraud. Malicious individuals can exploit these to create convincing counterfeit products that may unknowingly engage customers in fraudulent activities and make them vulnerable to fraud.

AI angel investor Jeff Renault told Forbes: There’s a lot of anticipation and excitement, but misinformation and scams can quickly take you down the wrong path. There are people building standards, but the technology is spreading too fast to keep up. “

Another issue is bias. AI bots can pick up biases from the data they are trained on, especially if they come from the internet. One notorious example of brand damage caused by biased chatbots is with Microsoft.

In 2016, the tech giant launched a generative chatbot named Tay on Twitter and other social media platforms. However, within hours of its release, Tay began posting inflammatory and offensive remarks that he had learned from his interactions with deliberately provocative users. Inappropriate and offensive behavior forced Microsoft to suspend the service and apologize.

Recently, the National Eating Disorders Association (NEDA) announced it would lay off some of the human staff working its hotline and replace them with an AI chatbot called Tessa. However, just days before Tessa’s official launch, test-her users began reporting that it offered advice on harmful diets, including encouraging calorie counting.

As with any digital system, deploying a customer service chatbot can introduce vulnerabilities that open the door to hackers. But generative AI can pose even more dangers, as interactions often involve the submission of personal customer data.

Vulnerabilities include lack of encryption during customer interactions and communication with backend databases, inadequate employee training leading to backdoors and unintentional exposure of personal data, chatbots, websites, and databases used including potential weaknesses in the hosting platform used. These can be exploited to steal data or spread malware, so businesses must remain vigilant and be prepared to patch them once they are identified.

Companies that want to leverage cutting-edge chatbots now have to invest significant amounts of time and money, but this will shrink as the technology becomes more commonplace.

Exactly what this means for human agents is still unclear, but it could be a game-changer for customer service. Companies must adapt and redefine employee skills and roles to ensure a harmonious integration of technology and humanity.