Generative AI in KM initiatives at KMWorld 2023

Every enterprise strives to harness its employee’s expertise to yield bigger and better positive outcomes, from optimizing business processes to powering innovation. Generative AI (GenAI) may hold the key to that potential nirvana; however, without the proper best practices and strategies to call upon when needed, it may prove more disastrous than blissful.

Jason McCullagh, director, marketplace alliances, Upland BA Insight Software, led KMWorld 2023’s session, “Generative AI in KM Initiatives,” to consider the ways GenAI and KM initiatives can find success as a formidable unit rather than warring concepts, walking attendees through  trends, technologies, and practical considerations of a united GenAI and KM strategy. 

McCullagh began by providing some context, stating that, “you have data in many, many sources all around your organization. This is not new; the problem is that the amount of data you have increases everyday”

This proliferation of data incites system-hopping, where users are on an endless search for information, confronted with access security, validity of data, and more. Once you throw GenAI into the mix, they generate text that gets saved into various other sources, ultimately leading to even more information and more complexity.

He explained that it has endless applications for KM, automating content generation, enhancing information retrieval, facilitating decisions, enabling personalized knowledge delivery, and more.

McCullagh then offered a use case, where a legal firm implemented GenAI to improve its processes. These included:

• Firm-wide billing improvement
• Optimizing global client retainer work
• Auto-generating draft advice for lawyer validation
• Developing and powering new client services lines
• Extracted entities such as key dates, which were then built into a timeline

“These types of technologies are all very new, so you must avoid the mile-pick view that it improves productivity,” warned McCullagh. “Gartner suggested that a large part of decisions made through generative AI were wrong…it can’t be automatically trusted to improve productivity.”

There are other concerns that enterprises must contend with while investing in AI, including data privacy, IP and copyright, misinformation, national security, and human bias. Outside of these concerns, there are various challenges, such as costs, change management, lack of standards, and little understanding.

Furthermore, McCullagh stressed the need to differentiate enterprise search and GenAI. He explained that enterprise search is based on existing information and rules, where GenAI creates something entirely new.

While some may say enterprise search is dead, McCullagh argued that it is, in fact, imperative. The ability to use RAG (retrieval augmentation generation), which works between search indexes and an LLM system, entirely depends on the integration between GenAI and enterprise search. Supported by a search index that consists of enterprise content, users can be assured that the content that a GenAI surfaces is derived from a trustworthy, accurate source.

“Generative AI can only impact or learn from the data it’s connected to,” McCullagh emphasized.

The connection between the AI and the information is fundamentally critical. If you haven’t centralized all your data, you might as well stop chasing AI, according to McCullagh. If you only have “garbage” to put in, you’ll only get “garbage” out.

McCullagh offered this strategy for successfully incorporating GenAI into KM:

• Define clear business objectives
• Ensure data preparation and quality assurance
• Research and select the right AI model
• Factor in all costs
• Deliver user training to adoption
• Outline acceptable usage policies
• Track rigorously and intervene quickly

“The race is on at the moment to either add a LLM to a search engine or add a search engine to an LLM,” predicted McCullagh. “I suspect over time that this will become a single point of access for all of your information. If that fails, you’ll end up going to the internet.”

In the end, McCullagh concluded that people must “understand what generative AI is, and what it isn’t. It’s only as good as its data; what we have today will be outdated tomorrow.”

KMWorld returned to the J.W. Marriott in Washington D.C. on November 6-9, with pre-conference workshops held on November 6.

KMWorld 2023 is a part of a unique program of five co-located conferences, which also includes Enterprise Search & Discovery, Enterprise AI World, Taxonomy Boot Camp, and Text Analytics Forum.