ServiceNow makes generative AI accessible to more areas of its low-code development platform, putting it at the heart of the chatbots that businesses are starting to use to interact with their ServiceNow applications.
But as software vendors like ServiceNow, Salesforce, and SAP offer new ways to leverage generative AI capabilities, such as summarizing text or generating new text and images from simple prompts. There are risks CIOs need to consider before giving technology freedom. with their data.
ServiceNow just last month announced its first generative AI tool. ServiceNow Generative AI Controllers for connecting Large Language Models (LLMs) to software automation platforms, and Now Assist for Search to generate natural language using these LLMs and company-specific data. Responses to queries made in Virtual Agent.
Start Virtual Agent Assist
Our latest addition, Now Assist for Virtual Agent, builds on that foundation, making it easier for businesses to adopt generative AI more broadly in the design and execution of their business processes.
Like Salesforce with its Einstein GPT product, ServiceNow chose to embrace generative AI in a modular fashion, allowing CIOs to choose which LLM providers to integrate with.
For ServiceNow, the options are initially somewhat limited to either OpenAI, creator of GPT and other public models, or Microsoft Azure, which also uses OpenAI technology. However, the company recently partnered with Nvidia to help companies develop custom LLMs trained on their own data, which companies can use to build private models tailored to their own needs. We also worked with Hugging Face on an open access LLM.
Neil Ward Dutton, vice president of IDC analysts for AI and Intelligent Process Automation, said this public-private distinction is important.
“There is a lot of confusion between the public foundation models that OpenAI (such as GPT-4) promotes and the generative AI models (which don’t necessarily have to be public) that we believe will ultimately provide value to businesses. You can see,” he said. .
Many uses of generative AI will only be attractive to businesses if they are protected from public access and have access to specialized models trained and tuned for their industry or specifically for their own organization. Other applications that do not require company-specific data or high levels of accuracy can be built on the public model.
“Salesforce, ServiceNow, and other vendors haven’t always been able to make a clear distinction between these two approaches,” he said. “They all partner with OpenAI, Google, Anthropic, etc. to get access to their public models to avoid risk, but to help implement customer-specific models he has partnered with Nvidia, Hugging Face We are also partnering with Cohere.”
ServiceNow, which runs shared services internally on the Now Platform, recently began piloting the use of generative AI in virtual agent conversations, according to the company’s CIO Chris Bedi (pictured). This is used by the go-to-market team to access a knowledge base of policies and processes to facilitate contract renewals, he said.
The idea is that instead of the virtual agent generating links to piles of knowledge base articles that employees have to read for themselves, at various points they say, ‘Here’s some helpful bite-sized content.’ It means that We should be more productive and faster in this conversation,” he said.
Garbage in, Garbage out
Bedi also believes there are risks in giving generative AI tools access to the wrong data, but this is nothing new.
“This is the garbage inflow and garbage outflow problem that IT people face forever,” he said, but with a twist. With traditional search tools, “if you have bad data, it will surface,” he says. “But generative AI shines a brighter light because it makes information easier for humans to find and understand. I have.”
The risk of incorrect data being displayed isn’t unique to ServiceNow implementations, but IDC’s Ward Dutton told CIOs about the origin of generated AI elements included in software suppliers and the data used for training. This is why we encourage you to ask about
According to him, companies need to know whether the underlying models are public or private to the organization, what data they were pre-trained on, and how they can protect against bias in the training data. They say they want to know.
Some software vendors are beginning to add layers to their generative AI platforms to increase the confidence of the displayed information.
Over time, even that can be handled by large language models, Bedi said. “You can have models look at models,” he said. “That technology is being built very quickly.”
Doing it makes a difference, Ward Dutton said, and advised asking CIOs how, if they have a vendor layer of trust, it actually ensures data quality. “Is it done by controlling how the model is trained in the first place, or by fixing or minimizing the content issues that the model creates after the fact?” he said. .
He advised CIOs to set up lab environments where they can safely test generative AI technologies, research use cases, and consider vendor claims.
This is what ServiceNow’s Bedi used their own technology to see how it performed. His CIO looking to do the same will likely do better by following his diversity strategy. That means conducting a pilot with a mix of tenured employees and new hires.
“We thought it was important because people who joined recently ask questions that people who have been in office don’t ask because they only know through their knowledge of the tribe,” he said. Told. “And people who have had tenure find things that feel weird or look weird far more often than people who have just joined the company.”
This allowed us to start working without worrying about having the perfect content repository available to Virtual Agent.
Some ServiceNow customers already have access to Now Assist for Search and now also have access to Now Assist for Virtual Agent. For others, the company plans to make the feature generally available as part of his September 2023 Vancouver platform release.
