Practical AI

Artificial Intelligence (AI) can mean a lot of things to different people. Ironically, most of us see or interact with AI on a daily basis, whether we recognize it or not! In order to get to the ‘practical’ in Practical AI, we need to define what AI is. And you can’t define AI without understanding intelligence.

For a moment, forget your preconceptions of intelligence. We often mistake ‘smart’ or ‘lucky’, or even ‘appearance’, with intelligence. We tend to limit our evaluation of ‘being right’ with ‘being intelligent’. If this is how you define intelligence, I’m going to ask you to stretch that definition a bit.

Consider for a moment, that intelligence doesn’t really have anything to do with knowing facts, but in making connections. I know plenty of people that appear to have access to millions of facts, but they can’t figure out how to order a latte (forgive me, I’m from Seattle). The myriad of choices bogs them down every time. We even forgive this behavior, convincing ourselves that this person is ‘contemplative’, one who considers their choices carefully. But the fact is they are quite often just stumped.

For the purposes of exploring Practical AI, I offer the connected PC. Connected to the internet, it has access to nearly unlimited facts (and fictions!). But alone, the PC is not intelligent. There are even programs, websites, and apps that help make data connections. Not intelligent. It takes a leap between seemingly unrelated facts, in combination, to create a new path of reasoning, to display intelligence. Machines (on their own) don’t do this.

Here’s a simple example: How many times have you seen someone by the side of the road, run out of gas? The ‘machine’ constantly monitors the level of fuel in the tank (not intelligence). There is usually a light that comes on when the vehicle reaches a critical level (not intelligence). Intelligence is displayed in your personal evaluation that you likely have X miles left to drive, AND that there is likely a gas station between here and X. AND you act, sometimes even going out of your way, to drive to that station to get more fuel. This is demonstrated intelligence. Relatively few facts, action taken to change the outcome.

Machines Don’t Make Decisions

Regardless of what any industry might say, machines don’t make decisions. They may appear to make a decision, but they are programmed. The outcome is pre-determined, based on input information. Probably the best example is chess playing computers. They appear to make decisions, based on our game play. We could even make the same moves, game after game, and the computer can take different paths. But it is merely spitting out a program, with the variations based on statistical algorithms. Not thinking, and certainly not intelligent.

But AI offers us the opportunity to leverage something computers do very well: process lots of information, quickly. Computers don’t get tired, and more importantly, they don’t get bored. And barring the occasional bug (or programming error), they don’t make mistakes.

Remember when I defined intelligence as ‘making connections’? One of the many, many holy grails of AI is in making connections from seemingly unconnected data. We’ve all experienced this, even if we thought it was creepy at the time. You’re daydreaming of your next vacation at work. You Google Greece, or Spain, or South America. You send an email to a spouse or friend, asking about their experience when they visited the islands. You may even check out fares on Travelocity. Then you notice a ‘side-bar’ ad at Amazon, advertising ‘cruise wear’, perhaps even a Caribbean cruise ad in an unrelated Google search. Coincidence? Even we aren’t that na?ve. Data has been mined. Connections have been made. No advertising guy made the decision to show you that ad. A programmed algorithm showed you something based on a combinations of your recent actions. This is rudimentary AI in action.

Perhaps this alludes to the ultimate value for AI in society. It’s not in self-driving cars, or computer-controlled production. Not even in ever more efficient machine maintenance (although all of these offer great opportunity). The ultimate value for AI may just be in commerce, and in connecting sellers to buyers, and to move buyers to buy. So what does this have to do with Practical AI?

Going back to the definition of intelligence: Making connections. Have you ever watched an episode of NCIS? Mark Harmon calls IT guy. “I need to know everyone who purchased a specific shovel, within a 20 mile radius, in the past 5 years.” McGee: “Got it, Boss. There have been over 200 purchases.” Harmon: “Give me males between 25 and 32, with a criminal history.” McGee: “There are 18. Texting the list and their addresses now.”. Science fiction? Perhaps. Possible? With AI, it is completely plausible.

There is no way that McGee was accessing databases, making queries, pulling the data, compiling, and making sense of it. Not in any real-time way. Such a search may be possible today, but it could take weeks or months. Certainly not at the pace required for a 1 hour drama. But what if the connections were already established? What if the queries had already been developed? What if the data was ready to show, and McGee only had to filter it?

This isn’t commerce, to be sure. But it illustrates what computers do best. Mundane, repetitive tasks. Over and over again. MAKING CONNECTIONS. But what connections to make? How can a computer ask the right questions? How would it know what connections to make? The fact is, it doesn’t. Sure, we can develop systems to learn the ways we search, but at the basic level, the computer can just slug it out. Which connections to make? All of them. Put as many queries together as is possible.

Remember the chess computer example? It isn’t make the ‘right’ moves, it’s making ALL the moves. If your first move is Pawn to King4, it doesn’t need to make any Pawn to King3, or to Q4, or to Knight3, or further calculations. It looks forward (a certain number of moves) along the P-K4 path. As you counter, the computer switches paths, based on statistical likelihoods. But make no mistake, it is making ALL the moves, a certain number of moves ahead.

Similarly, one of the promises of Practical AI is to make ALL the connections among data along certain paths. Based on how we work, how we search, and other parameters, it can appear to make these connections in real-time, providing answers to questions not only that we’ve answered, but to those we probably should be asking. And if we should be asking, the system will likely suggest it to us anyway. Not real intelligence, but artificial. At some point, we won’t really be able to tell the difference.

Have you ever gotten a call from a robot? Some of these are quite sophisticated. They sound like people, and have a myriad of response routines. The only real way to tell the difference is in small hesitations, when your response doesn’t match expected norms. And when they loop (ask the same or similar questions). They also don’t respond well to interjections in the middle of sentences (“How much does that – ARE YOU A ROBOT?”). Many of them give a forced laugh and say, “Of course not!”.

Like it or not, Practical AI is coming to your workplace, your profession, your desktop. You can fight it, and you can deny it. But there is a 20 something that doesn’t really want your job, he wants to create something to make your job obsolete. The only way to fight this is to embrace it. Be the growing professional, who leverages AI, instead of the resister (who always ends up getting cut from the team). Whether it’s Alexa telling you it’s time to buy more milk, or your design software telling you a better steel to use, or your finance software showing you better ways to flow cash, get in front of it. Explore the possibilities.

If you are curious about how AI can affect your professional future, do a search on Genpact Cora. Genpact Cora is a robotic automation and machine-learning platform, which can help you make the right connections, and give you ways to get better results.

Bill Kluck is a Lean Strategist, Analyst, and Innovator with Genpact, LLC.