AI: Separating Artificial From Intelligent

It’s been a typical day for you as chief executive. Too many meetings, an impending budget crisis, and an analysis of last week’s production outage. And then a board member walks in; “Hey, I’ve been hearing about this Artificial Intelligence thing. We should buy one and jump over our competition. Let’s get on it”.

Say what?

Today, we suffer a never-ending stream of pseudo-tech predictions depicting AI as a cure-all or, conversely, as the first domino falling towards a Siliconocracy. Still other ‘industry experts’ dismiss AI as just the latest clever parlor trick. Yet despite these sensationalists, AI continues to make steady advancements into a widening spectrum of industries. So how can we sift through the rubble to find the real AI gems?

Firstly, you can feel very safe in categorically disregarding the Utopians, Dystopians, and Agnostics. History shows, they unerringly overestimate the short-term impacts of emerging technologies while underestimating the long-term effects of progressive innovation.

Out next logical step is to set aside the seemingly natural urge to immediately demo an AI solution. Instead our efforts will be better spent if we can identify, scope, and analyze the one problem which will best leverage AI and return the highest return on our technological investment. This, of course, is much easier said than done.

“AI requires learning and learning requires data … and lots of it.”

We can now turn to finding a solution to the target problem, which inevitably begins with the obligatory question of buy versus build. In only the rarest cases, where you possess AI gurus complete with beanbag chairs and hipster beards, should you attempt to develop your first AI solution in-house. Instead, look at vetting one of the many AI-based solutions already on the market tailored to your target use case. The good news is there are hundreds of them – the bad news is there are hundreds of them.

Artificial Intelligence is cracking the code on a whole family of non-procedural problems, such as image recognition, autonomous systems, and anomaly detection. Our research team recently completed detailed studies on how AI-based user behavior analytics is changing cyber security and smart operational analytics is accurately predicting impending system failures. These are but two examples of legitimate AI game changers.

These advancements, however, are often drowned out by a deluge of disingenuous marketing. An all too common tactic used by AI hucksters is to position old-school savvy as new-wave smart. Savvy systems can only apply pre-programmed rules, logic, and inferences to existing knowledge. Nonetheless, savvy systems can solve an amazing array of complex problems like controlling a nuclear power plant, landing an airliner, or publishing the zillionth cat video.

Yet, the best of today’s savvy systems are being eclipsed by smart systems. Smart systems ingest copious amounts of information, discover patterns in the data, adapt to these new patterns, and apply this evolving knowledge to their dynamic environment. While savvy systems are adept, smart systems adapt. So, how can you spot savvy dressed up as smart without completing a degree in Computational Learning Theory?

There are two cold truths in AI. AI requires learning and learning requires data … and lots of it. If we probe these two axioms, we can quickly determine whether the product was designed in a vendor’s Engineering or Marketing department. Let’s hone Occum’s Razor a bit.

A smart system’s learning is not programmed into the software by smart coders. Instead, it emerges naturally from training in which it adapts to new data patterns. There are several tried and true open-source platforms on which most AI applications are built, so, when we ask a legit product vendor about how their product was developed and trained, we are not asking for them to divulge trade secrets. Instead, we should see a gleam in their eyes like asking a grandmother for pictures of her grandkids:

Smart Question Savvy Answer Smart Answer
Learning What makes your solution smart? The expertise of our engineers, coders, programmers and developers. We trained a convulsion neural network to identify predictive failure indicators.
How does your product learn? It uses a deep knowledge base and rules engine. We initially trained it on reference data in our labs but the product will adapt to actual behavior it sees in your data center.
What platforms were used to develop your AI? Our own proprietary

AI platform. Can’t share much – secret sauce.

We used SKIL and DL4J to develop our model and Oryx for the large-scale production stream processing.

As stated, learning requires data. But not just any data. AI training requires vast amounts of relevant preprocessed data – what data wizards call ‘wrangled’ data. If learning is like an engine, then big wrangled data is its high-octane fuel. We can discern wannabes from the real players by probing their training data:

Smart Question Savvy Answer Smart Answer
Data Can we train the product with our own data? No need. Our deep learning has been pre-programmed with all the rules you’ll need. Indeed. We recommend a two-week training period after installation before going live.
What data sources did you train with? Sorry, that’s proprietary. I could tell you, but then I’d have to kill you. Glad you asked. We dumped all call detail records since 1994 plus all seismic data and …
How much data did you train with? Seriously, I am

going to kill you.

We ingested three weeks of telemetry from 50,000 pipeline temperature and pressure sensors – over 43 TB in total.

AI is very real and very powerful. But so is marketing. Therefore, we must separate hyperbole from reality, discerning savvy from smart. A quick salvo of well-crafted questions can provide an effective litmus test. From here the actual implementation, testing, and deployment of the system is no slam dunk, but most AI solutions splice nicely into today’s most common software development lifecycles.

By ignoring the sensationalists, developing a high value use case, identifying a suite of candidate AI solutions, and filtering out the pretenders, you are well on your way to moving your enterprise from savvy to smart.

Mark Campbell
Mark Campbell is the Chief Innovation Officer at Trace3 where his teams review over 1,000 tech start-ups each year. Based out of sunny Denver, Colorado, Mark is a researcher and industry watcher who leverages his 25 years of real world IT experience to help enterprises adopt emerging technologies to tackle their toughest technical and business problems. Mark holds telecom patents, writes frequent articles for tech publications and works with the world’s largest IT venture capital firms.

PARTNER CENTER