Knowledge Graphs and Monkey Business with Generative AI

Throughout the year I got poked and prodded and challenged in a bunch of different ways by my friends, colleagues, and neighbors in the language services industry. Their questions all boiled down to:

What the *%$@! are we going to do now that gen AI is here? 

And my answer across keynotes, interviews, conversations, and blog posts was, in essence, just four words: 

Make even better AI! 

Gen AI produces startlingly good written output in many styles and languages. Unfortunately, this energetic, little gen AI monkey is cute as hell but nowhere near housebroken yet. There's lots left for the grownups with advanced language skills to do.

Remember that gen AI is a monkey see, monkey do technology:  it mimics what humans write on the internet. Not the best place to start if you want engaging, artistic, or legally bulletproof writing. And it seems to use wind-up monkeys at that:  we start it up and it goes pretty much where it wants to. Not exactly what you need if you're in a regulated industry which requires standard terminology and specific style guides. Of course, you can feed gen AI your own brand of bananas (with “RAG systems”) and that nudges the monkeys in the right direction. Most of the time. If they're in the mood.

Come to think of it, these monkeys often don't adapt well to the jungle of technologies that you already have in your company. So you need to spend a bunch to build new monkey houses or villages – with tight security so they don't get loose and wreak havoc with the neighbors. Probably not your best bet if you don't have the budget for really secure enclosures and a long time to train the monkeys. Even so, some companies are actually trying to clone their own monkeys hoping that they'll obey better that way.  Whatever the starry-eyed engineers say, I'm predicting years of work for these clones to actually be helpful.

In the meantime, those of us just trying to get some work done have to put up with a gazillion little monkeys screeching, jumping all over, and making a huge mess throwing things and breaking stuff. And you thought it was a zoo at work before, eh? More fun than a barrel of monkeys? Just wait until they deploy gen AI to “help” you where you work…

And even so, I still hear: Aw, but they're so cute, those monkey-like little chatbots!  

You already know that it's possible to train monkeys to do complex and sophisticated things – even to reason statistically (but not to write Shakespeare). If you've ever trained monkeys, though, you also know how incredibly slow and difficult it can be. You can't simply tell them what to do – even if sometimes they do seem to understand. And you need real experts to actually get the job done. Engineers and CEOs, for example, don't have much success with training monkeys.

Gen AI is really not very different. 

The experts you need in this case are language professionals. Not more engineers, not cooks or personal trainers. Language professionals are the only ones who know enough to get text-based gen AI to behave reliably. So your chatbot doesn't foam at the mouth, cuss out your clients, or invent unplanned discounts that give away your profits.

These pros generally tame the AIs in two different ways. 

First they improve the AI's diet. No more junk food from Internet writings -- it makes the AIs too agitated and unpredictable. They re-train the AIs with high-quality professional writing for specific domains – surgically (through “fine tuning”) or via injection (with “RAG systems"). This teaches the AI to pay attention to what's important and ignore the junk.

Then they send the AIs to school. They teach them about the real world, not just about the Internet. They create glossaries and encyclopedias (called “ontologies” or “knowledge graphs”) that not only mention things in the world (like regular sentences do) but actually explain what words mean and how things work or interact. Since language professionals take the time to build up all this knowledge about the world or about your business, their AIs actually know what the heck they're talking about. 

So rather than monkey around, go find the right kind of people to train your AIs to do cool and even useful things.

If you need strategic help assessing your data Gaps & Opportunities, or technical help getting your data AI-ready, then just message me directly here on LinkedIn.

To find out more about how language professionals will drive the next wave of meaningful AI, check out more of my work from 2024 below and subscribe to my Knowledge Architecture newsletter on LinkedIn here. 

• It takes a village to build AI 
• A sixth Sense for building Knowledge-rich AI
• Knowledge Graphs for Information Sherpas
• Knowledge graphs for content creators
• Better knowledge for better AI 

• The brand-new Common Sense Advisory report that I wrote with Arle Lommel on Taming Global Content with Knowledge Graphs

• Are language professionals doomed?, with a link to my video interview with translator  Sylvie Lemieux
• Knowledge Graphs as jet fuel for AI, an audio  interview with Larry Swanson

• Knowledge Graphs, Linguists, and the last-mile problem, with a link to The next “new normal” for language professionals (a video of my keynote at MemoQFest in Budapest)
• Knowledge Graphs: There's more to translation than slinging strings, a video of my keynote at the Association for Machine Translation in the Americas meeting