Are you in compliance?

Don't miss out! Sign up today for our weekly newsletters and stay abreast of important GRC-related information and news.

If AI can’t bake a cake, can it really do compliance?

Bill Coffin | April 3, 2017

I have written repeatedly about the advent of artificial intelligence in the field of compliance and how revolutionary this technology promises to be. Indeed, the expectations are high for most who follow this topic, since early indicators point to how an artificial intelligence-driven compliance system can radically improve efficiency within a compliance department by removing humans from the roles of manually processing data.

However, it’s still early days for AI in general and, every once in a while, we get to see what the growing pains for a new technology look like, and the results can be pretty hilarious.

Janelle Shane, a neural network research scientist, recently posted some results from a project of hers in which she is using an open-source neural network and a large database of cooking recipes to see what kinds of meals the AI will come up with. This is something Shane has been working on for a while, running the same data through the neural network and helping to calibrate the results—essentially teaching the AI how to properly interpret the information Shane is giving it. If you’ve ever wrestled with having to teach a voice-to-text translator how to interpret the things you’re saying, it’s a similar process, only far more difficult.

Shane has spent a lot of time and energy teaching her AI how to cook by feeding it thousands of recipes, and here are some of the titles of recipes it has come up with. You are not ready for them. No one is:

Cream Cheese Soup

Cream of Sour Cream Cheese Soup

Chocolate Cake (Chocolate Cake)

Chocolate Chocolate Chocolate Cake

Chocolate Chocolate Chocolate Chocolate Cake

Chocolate Chicken Chicken Cake

Chocolate Chips with Chocolate Chips

Now, that’s when Shane has the creativity variable of the AI turned way down. Here’s what we get when the AI can get a little more free with its thinking:

Beef Soup With Swamp Peef And Cheese

Chocolate Chops & Chocolate Chips

Crimm Grunk Garlic Cleas

Beasy Mist

Export Bean Spoons In Pie-Shell, Top If Spoon and Whip The Mustard

Chocolate Pickle Sauce

Whole Chicken Cookies

Salmon Beef Style Chicken Bottom

Completely Meat Circle

Completely Meat Chocolate Pie

Cabbage Pot Cookies

Artichoke Gelatin Dogs

Crockpot Cold Water

The internet caught wind of these culinary masterworks and promptly went bonkers over it, as people amused themselves by asking themselves exactly what Beasy Mist is and what it’s supposed to taste like, or how one is expected to go about putting together Crockpot Cold Water. I mean, it looks so simple, there has to be some kind of catch to it, right?

For those fearing that AI is making a rapid ascent to replace anything and everything humans are good at, stories like this are a humorous comfort. We get to see how absurd the results are and are reminded of the limits of this technology. And yes, it is funny. I cannot read these lists aloud and not break out in giggles (try them yourself while keeping a straight face, I dare you). And no, the idea of having our own vocations replaced by machines is no laughing matter.

But here’s the thing: For however basic AI may be in its present form, it’s already proving to be an extremely useful tool. I am reminded of an app one company developed that monitors internal corporate communications, using AI-driven predictive analytics to flag when people are using certain words or phrases that might violate compliance protocols, whether it’s speaking amongst colleagues, or with clients. I am reminded of another example recently cited to me of how rudimentary AI systems are making huge strides in helping financial services institutions screen huge numbers of transactions for basic indicators of potential money laundering; once flagged, a transaction is then routed to a human for a more thorough qualitative analysis.

Right now, that’s the big divide with AI as a compliance tool—it’s great on quantitative work, but struggling on the qualitative side. And that’s good news for humans, because the qualitative side is where we excel. The fears regarding AI almost always come down to how well it can master the qualitative side of things, though, and that’s certainly the case in compliance.

However swift AI will mature as a tool we can use, it is still just that: a tool. How we use it, and what value we derive from it is up to us. Moreover, there is a stern warning to all baked in here: If the tools we are using to achieve best practices in compliance could possibly remove us from the roles we currently play, then it is up to us to figure out how to stay ahead of that curve by specializing in tasks that cannot be so easily automated. For some, no doubt, the smart play will be to leave compliance altogether. But the smarter play will be to do what every leading compliance officer is doing already: perpetually reinventing their skillset so that they can pursue best practices, whatever those practices might be. After all, we are already seeing the limits of compliance efforts that involve lots of people trying to behave like machines; they grow tired and make mistakes that cannot afford to be made. The question with automation and AI isn’t how to run a great compliance department with less people. It’s how to use those people to the greatest effect. Smaller usually isn’t the answer here, but smarter always is.

Now if you’ll excuse me, I need to tuck into a plate of Salmon Beef Style Chicken Bottom, whatever the heck that is. Wish me luck.