Large language models still need a help-desk human

Large language models (LLMs) are boosting help-desk capabilities and agent-assist offerings, but don’t fret, human IT worker.

AI advances may assist with IT help desk tasks like password resets, provisioning, and diagnostics, but some IT pros believe human-led support will remain, especially because people have two traits not easily found in rule-based models: empathy and judgment.

“You call into a healthcare service, and you’re worried. [If] you have a diagnosis of something that might be serious, you don’t want an automated response,” said Bern Elliot, VP and distinguished analyst at the market-intelligence firm Gartner.

What is a large language model? Trained on large amounts of text-based data, large language models can summarize data, make predictions, and generate responses—capabilities some help desks are utilizing.

• SolarWinds’s service desk added an AI-powered virtual agent in May, which aims to help end users with their troubleshooting-related prompts like, “I’m having issues with my network,” or “Our Wi-Fi isn’t connecting,” Cullen Childress, group VP of product at SolarWinds, said.
• Boost.ai recently announced a pairing with large language models, as well as generative capabilities “to suggest messaging content to AI trainers.”
• Nextthink offers IT support capability through an LLM, thanks to an LLM trained on device data. Users can write prompts and questions like, “Retrieve all devices having Outlook crashes,” and “Which users in London have audio quality issues with Teams today?”

Large language models can assist IT help-desk teams dealing with lots of data. The systems can automate the sending of frequent emails and summarize details after the conclusion of a ticket—“wave 1” of LLM support in customer service, according to Gartner.

The dream of every service desk manager is to “shift left,” said Mario Tarallo, product management director at Nextthink.

“If we introduce automated insights, these allow, for instance, a service desk analyst at level 1 to come up with a potential solution for an IT ticket without the need to bother level 2 or level 3,” Tarallo told IT Brew.

But before giving AI the help desk and its own coffee mug, let’s remember past customer service nightmares that might have ended with someone hitting “0” repeatedly and yelling “HUMAN!!” into a phone.

To build confidence with a virtual assistant, it often helps to have someone real nearby.

“When we first launch with our employees, we want to let them know, at any time, we’re happy to connect you with a human. We’re not trying to keep you away from that conversation, if you feel more comfortable with it. And that really builds trust,” said Bill Schwaab, VP of sales, at Boost.ai.

While LLMs can be used for sentiment analysis to help users to better understand the intent of a piece of content or a particular response, a human who can empathize will still play a critical IT role according to Elliot.

Maybe an employee wants a new application unapproved by the company. Or maybe that same employee gets frustrated being locked out of a legitimate one.

“How they’re treated at the help desk is taken as a reflection of how the company feels about them,” said Elliot.

And no employee wants to feel like they’re just some cog in the machine learning.