Simplr Inc., a chatbot-powered customer experience technology provider, said today it’s integrating generative artificial intelligence that’s guided by “comprehensive safeguards” into its platform.

The new offering, called Cognitive Paths, is designed to prevent AI-powered chatbots from making false or misleading statements. Cognitive Paths was created through an integration with OpenAI LP’s large language model GPT-4, the generative AI behind the famed chatbot ChatGPT.

ChatGPT took the internet by storm in November 2022 thanks to its ability to respond to questions and prompts in a humanlike, conversational way, and its capabilities have rapidly been integrated into numerous industries. One of its most obvious use cases is the customer service industry, where it can potentially automate the way companies handle thousands of customer queries.

However, companies need to be cautious when using generative AI to engage with their customers. One of the dangers of AI is its tendency to “hallucinate” and engage in off-brand topics when puzzled by a customer query.

AI hallucinations occur when generative AI doesn’t know how to answer a particular question or query. Instead of admitting it doesn’t know, generative AI will oftentimes simply fabricate a response instead. So brands need a way to prevent this, or risk serious damage to their reputations.

This is where Simplr says Cognitive Paths can make a difference. The company says its integration of GPT-4 is governed by a strict set of AI training parameters that guide chatbots to the right customer resolution, without any risk of hallucination. It does this by reducing the volume of information available to the chatbot, so it cannot tap into false knowledge.

Cognitive Paths works by generating a unique, proprietary dataset for each customer. This serves as an extensive, curated knowledge base that’s built on highly targeted and customized datasets. The LLM will then be instructed to only pull information from specific datasets, based on the nature of the customer interaction. By reducing the available information the chatbot can dig into, Simplr says it all but eliminates the risk of AI hallucination.

In an interview with VentureBeat, Simplr Chief Executive Eng Tan said the guiding philosophy behind Cognitive Paths is “database segregation.” He explained that AI hallucinations happen because generative AI models are exposed to too much information. Out-of-the-box LLMs cannot autonomously discern how accurate and authentic a knowledge base is, resulting in false or nonsensical responses.

Besides providing safeguards, Simplr said Cognitive Paths will enable customer service chatbots to automate resolutions to far more complex issues. Tan explained that Simplr has focused on understanding the full range of multi-turn inquiries that chatbots cannot adequately handle. These interactions therefore have to be handed off to human agents, but Simplr says Cognitive Paths will be able to handle many of them.

“Other chatbot providers have always thought ‘What can be deflected?’,” Tan said. “Meanwhile, Simplr has always focused on ‘What can be resolved?’ ‘What replicates the very best human agents?’ That is what gives us a distinct advantage in the customer service generative AI race.”

Simplr said Cognitive Paths will be able to handle “Level 2” complex inquiries such as upsell and cross-sell opportunities as well as technical support scenarios. As a result, it can automate many kinds of questions and inquiries that remain outside the purview of other chatbots.

Amit Sood, chief technology officer of Simplr, said Cognitive Paths literally and figuratively “changes the conversation” around AI chatbots. “Cognitive Paths dramatically expands the types of customer interactions which can be effectively resolved through automation,” he said. “At the same time, customer service leaders can rest easy knowing their brand, CX, and reputation will only improve through quicker, more impactful customer resolutions.”

Image: Simplr