NVIDIA Reads Between the Lines

NVIDIA wants it’s chatbots to get to know you really, really well.

Photo by Sundry Photography via iStock

Sign up to uncover the latest in emerging technology.

NVIDIA wants its AI chatbots to be a little more understanding.

The company wants to patent a method for “determining intents and responses” with machine learning in conversational AI. Here’s how it works: the first machine learning model takes a passage of text, such as a query from a chatbot user, and boils down the first initial intent or meaning of the text. (FYI, an intent could be just about anything, including asking it to book a reservation, schedule an event or create a shopping list.)

That data then goes through another machine learning model to break down what it calls “sub-intents,” or additional meaning or context. This process is rinsed and repeated “in order to determine a final intent associated with the text,” creating a branching “tree-like structure” to associate intents with one another.

Deciphering intent behind text can be a difficult task for a machine learning model. If you train a model on a set of specific intents, as you would in an “intent-slot-based model,” you need a ton of training. For example, if you use this kind of model for a customer-service bot, it may only be able to answer a very narrow set of questions based on the number of intents it’s trained on.

Alternatively, you could use what’s called a “zero-shot model,” or one that essentially categorizes the intent of text based on a smaller number of pre-trained categories. While this model requires less training to answer more questions, the result is often lower accuracy.

NVIDIA’s system, however, circumvents both these issues by passing the data through machine learning models multiple times, until a final intent is uncovered.

Photo via the U.S. Patent and Trademark Office

NVIDIA is without a doubt a heavyweight in the AI space. Much of its grip, however, is in hardware: The company controls 80% of the market for AI chips and GPUs, and is miles ahead of competitors like Intel and AMD.

But NVIDIA isn’t new to the software or chatbot space, either. The company has long been working on conversational AI tools with a focus for industries like financial servicescustomer service and telecommunications.

In April, the company also announced new software called NeMo Guardrails, which claims to address the issue of large language models hallucinating incorrect answers. The company also has sought several patents for chatbot-related tech, including one that can utilize both open and closed domains to answer questions more accurately.

OpenAI, Google or Microsoft are often the first names that come to mind when you think of an AI chatbot. While these companies lead in gaining consumer attention, NVIDIA has targeted the enterprise market, Romeo Alvarez, director and research analyst at William O’Neil. Since NVIDIA is already the “gold standard” when it comes to hardware, the fact that it’s able to offer a neatly wrapped package for a variety of different enterprises to embed AI into their operations gives it a leg up, Alvarez noted.

Making that AI more intuitive, as this patent aims to do, will only draw in more business, Alvarez added.

“When it comes to more enterprise-oriented technology, where they don’t have to compete with OpenAI, they do have the advantage because they have the complete stack of software and hardware,” said Alvarez. “It’s a combination of the fact that they are working on both that gives them the advantage.”

Have any comments, tips or suggestions? Drop us a line! Email at admin@patentdrop.xyz or shoot us a DM on Twitter @patentdrop. If you want to get Patent Drop in your inbox, click here to subscribe.