Human Curation in Knowledge Management

There is a problem in relying on document reading systems and other automated processes to ingest an organization's knowledge for storage and search. The problem is quality.

For shorthand, let me use Artificial Intelligence (AI) as the term for automatically reading in an organization's materials in a way that makes them available for search by employees and stakeholders.

Ingestion Effectiveness

Getting data from its current situation into a searchable form takes many different paths, depending on the source material. While there are many PDF files, MS Office documents, and other electronic formats where knowledge is kept, there are still a LOT of 3-ring binders, forwarded emails, and posters-on-walls with critical information.

Using the electronic format files is straightforward for getting text into knowledge systems.  Intent and meaning are often lost, though.

Getting non-electronic stored knowledge into the system ranges from difficult to almost impossible using purely AI processes. At best, someone must scan documents in a binder into an image file, which can then be read using OCR systems. At worst, someone must transcribe what is written on a whiteboard into an electronic document. At KLONE, we see all ends and everything in between.

 Dumping all of the information into an AI process is not easy, but it can be done.

Translation to Knowledge

The bigger problem comes when surfacing the information as knowledge to people who need answers.

Just because something is ingestible into a knowledge repository doesn't make it useful or usable to people.

Simple things like when one system refers to clients while another calls them customers stymies simple searches. That kind of vocabulary difference is solvable with technology, but it does require a little bit of work (it also requires ongoing work to make sure new source vocabulary is mapped to existing materials too!). 

Differences in vocabulary internally between people or departments make simple search tools difficult to use. Different groups or departments sometimes (often?) use their own vocabulary that doesn't translate well to other parts of the same company. Also, organizations usually create their own vocabulary which hinders using widely available libraries and dictionaries.

Nightmare Examples!

Most organizations seem to have a lot of information stored in emails. Almost certainly, you can imagine from experience asking someone how to do something and getting an email in response that has been forwarded and replied to many, many times.

Those emails are critical to making processes work. There are almost always modifications and changes to the original instructions embedded in all the forwards above the original email instructions. No AI process can consume the email, merge the disjointed changes, and then re-summarize the process knowledge into a readable format. 

Maybe AI may have that capability in the future, but for now, it is not realistic. It is possible now to custom create AI processes for each email example. Still, you will invest orders of magnitude more time in development than simply having a person edit the material.

Human Curation Adds

Having people work with material is critical. Despite the power of AI for cracking and consuming documents (through Natural Language Processing or various text analytics), human changes and clarifications are usually needed with existing material.

Often, existing material has more text than what is necessary. By that, I mean there is usually more text and information than is necessary to answer a question. Surfacing long text sections to people usually leads to people ignoring the answer and eventually ignoring the entire knowledge system. The most effective knowledge system looks to a subject matter expert to change the material, which directly aids people.

That type of editing and focusing is not possible with AI systems except is custom developed AI code. The cost of custom development to the level needed is beyond the reach of all but the largest organizations.

Scalable

Of course, having people do everything to get useful knowledge into a central system is not cost or time-effective. Subject matter experts cannot spend all their working time to re-craft answers. KLONE helps our customers by creating a framework using AI to extract materials from different formats. KLONE services then presents this extracted material to subject matter experts for final curation by subject and questions.

Combining both AI and human curation the way KLONE does optimize the best parts of both to overcome each's weaknesses.

Benefits

The obvious benefit of having subject matter experts finish content with their curation is the focus and correctness of the questions and answers stored in the system. The other major benefit is the standardization of vocabulary. 

There are other reasons, but these are two critical success factors in a knowledge system.