Semantics and Conversational Understanding in Generative AI

Elizabeth Tobey starts by asking: You know what they say: Context is key. Regarding AI, precision of language could never be more important.

Picture this: you are heading back to work after a coffee break, but during the elevator journey up to your coworking spot of the day, the car shudders and stops. You find yourself somewhere halfway up a downtown office building, stuck. The internet still works, though, so you send a quick message to your colleagues reading, “GUYS, I’M STUCK IN THE WEWORK LIFT.”

And your Slack channel’s AI helpfully chimes in that “guys” is a gendered term and, to be inclusive, you should use a term like “everyone” or “folks” or “team.” Correct and yet it completely misses the point of your call for help.

Context here is disregarded completely.

The elevator scenario detailed above is not a fabrication: it happened to someone in March of 2022. In that instance, the “inappropriate language” auto response from a Slack chatbot did no harm, but it certainly did not help create a seamless user experience, and it missed the important context of the author’s message.

Chatbots like these are trained to ping specific keywords and phrases and draw on a bank of pre-formed answers to solicit a reply. As we raise our expectations of AI and chatbots, however, this “old way” of doing things cannot keep pace with the complexity of language. We need to use generative AI, and the large language models it is trained on, so that chatbots can comprehend and appropriately react to the messages being sent. Natural, conversational language requires AI to find multiple keywords within a sentence or paragraph and connect them together to parse context.

While gender-inclusive language is, undoubtedly, an incredibly important part of a strong and collaborative workplace, and the auto response in the scenario above was a worthy addition to a company’s main communication channel, it missed the important context of a high priority message: that the author needed emergency assistance.

In CX, these kinds of red flag emergencies happen all the time, and brands need their chatbots to understand the context of a consumer’s question immediately. In high-stress situations (power outages, delayed flights, flat tires, urgent package arrivals) consumers are likely to phrase their needs, wants, and high-priority issues in many ways. AI must understand what is being communicated and respond with a clear and correct solution the first time – every time.

This is where the capabilities of generative AI, partnering with models crafted from hundreds of voice and text customer experience interactions (CXi), can make what is an important (but sometimes unhelpful or irrelevant) AI-powered chatbot an effective, trusted customer experience tool. By drawing on the most complete library of successful customer interactions, our AI tools can derive meaning from multiple keywords. This helps AI rapidly gain understanding of the exact nuance and context of any specific scenario, eliminating the pitfalls of previous generations of chatbots. While “traditional” chatbots can give us a response that may be directionally correct for our needs, those responses are not specifically tailored to our individual scenario. Partnering this trove of information with generative AI means we can now take the wealth of unstructured data and create unique, novel utterances, written specifically to completely answer the customer’s needs, not just react to one keyword.

To circle back to our friend stuck in an elevator: the situation would have progressed differently with generative AI involved. Once the request for help was received, generative AI might have pinged maintenance and alerted the employee’s teammates to the situation, prioritizing the most urgent need within the entire phrase first. Then, later, it could have sent a reminder to the employee, suggesting better, inclusive language. By matching and weighting multiple keywords in order to understand the full context of the message, the AI machine powering this employee’s experience could have been significantly leveled up – solving both the short-term, red flag issue as well as integrating the important, but less urgent, information the company weaves within its systems and culture.

To achieve this level of contextual understanding, it’s imperative to build your next conversational chatbots using not only the power of generative AI to create human-like responses, but also to train that AI on CX-specific interactions. Then, by integrating the technology across your entire customer journey, it is possible to facilitate a higher level of understanding and caliber of outcome no matter when, where, or how the conversation begins.

Take the next step to engage customers like never before with brilliantly human self-service.

This blog post has been re-published by kind permission of NICE – View the Original Article

For more information about NICE - visit the NICE Website

Call Centre Helper is not responsible for the content of these guest blog posts. The opinions expressed in this article are those of the author, and do not necessarily reflect those of Call Centre Helper.