What is Artificial Intelligence?
Whether it’s called AI, Business Intelligence, or Deep Learning, we’re well and truly living in an age where Artificial Intelligence is everywhere around us, and widely used in every day applications. But what is AI? And what makes it different from regular old computer programs we’ve been using for decades? Let’s take a look.
Means vs Ends
Although many experts in the AI space will use different definitions for what Artificial Intelligence is, they all have a couple basic concepts in common. The biggest distinction between traditional computer applications and AI systems are how users give instructions to the machine.
With a traditional computer application, like creating a new Word document, the application has a list of steps it follows to create that document. It opens a new window, it sets aside some memory for the new document, it names it Untitled, and so on. Each of these steps usually is composed of many smaller steps, and might branch depending on your selections, but fundamentally, there’s a procedure for creating a new Word document, and the computer follows that procedure.
AI systems operate differently. With an AI system, we describe an end situation we want, and then the AI systems decides how to use the simple tasks to make that happen. The system has a degree of decision making. It decides which simple tasks use as opposed to following a rigid instruction set. . For instance, when we save a file to a cloud storage drive, like OneDrive or GoogleDrive, the cloud service doesn’t actually just follow a single set of instructions to move your document onto a specific server somewhere off in California. It knows you want the document to be accessible in the future, and it searches all the best possibilities, makes a decision about the optimal location, then puts the document there, without the user having to instruct it do so.. It makes a thousand little calculations about where it should store it, where it will be the most accessible,, how it will make sure it’s backed up, and how to minimize storage and bandwidth costs for the service provider. All without a human user providing the instructions.
If you’re not clear on the distinction, it might be easier to explain this in terms of the quinessential anthromorphization of AI: a robot. Suppose we have a robot in our office who files real, physical paper documents for us. We have two robots: Ludd and HAL, who respectively use traditional and AI methodology for filing.
If we give a document to Ludd to file, he takes the document, walks six feet north of us, turns to face the filing cabinet, opens the top drawer, and puts the file in, and closes the drawer. Then Ludd waits for us to ask to file another document. Very simple, very straight-forward.
When we give HAL a document, HAL doesn’t know how to file it per se, but HAL does know we want this document to be in the filing cabinet. HAL knows a few things, like how to check if there’s a filing cabinet in front of it, how to open a drawer, how to put a document in an open drawer, and how to move a short distance. HAL also has the ability to tell if the space directly in front of it is blocked or not. HAL has a rough map of the office in its memory.
HAL starts by making a plan for how to file the document. It looks at its mental map and works backwards. It knows it needs to have this document in a filing cabinet. It knows to put a document in a filing cabinet, it needs to be standing next to the filing cabinet, and the drawer has to be open. It also knows it can open a drawer if it is closed. It knows roughly where it is, and roughly where it believes the filing cabinet is. So HAL constructs a plan to, in inverse order:
4. Put the document in the filing cabinet 3. Open the filing cabinet drawer 2. Turn to face the filing cabinet 1. Walk six feet north.
So, it seems like HAL did a lot more work to accomplish the same thing here. Why would we ever want HAL instead of Ludd? The answer lies in the fact that HAL responds to changes in the environment while Ludd can’t..
For instance, imagine we moved to a different desk. Now HAL needs to move further than six feet to get to the cabinet. Fortunately, since HAL recalculates its route every time it moves, he can find its way to the cabinet from anywhere in the office without being told. HAL can also handle obstacles. For instance, HAL might know how to unlock the cabinet if it tries to open it and find it can’t. HAL can add steps to its plan that involve moving to where the key to the cabinet it, getting the key, returning to the cabinet, unlocking the cabinet, then returning the key to its place, and then continuing storing our document.
Of course, HAL isn’t perfect. . HAL might not be able to negotiate stairs, so if we moved to a different floor, HAL might not know how to get to us, or how to get to the cabinet. There are limits to machine learning as of today.
In these two circumstances, neither robot can handle every situation thrown at it, but HAL can adjust to small changes in its environment much more easily. HAL uses vastly more computing power than Ludd does, but whereas Ludd could be thwarted by a box placed in its path, HAL will be able to cope with a variety of changes without anybody needing to tell it what to do.
So goes the analogy with the other ways AI is making its way into our work. More advanced AI systems don’t just have a fixed list of things they know how to do but can learn new things on their own by experimenting with applying their simple tasks in a variety of ways. Other systems are designed not to take any outward action, either physical or virtual, but rather to just classify data into different buckets; that’s how YouTube or Facebook decide what to recommend to you. Still other systems try to tease out relationships in your browsing history to decide what advertisements might be most likely to entice you.
At the end of the day, Artificial Intelligence as a concept isn’t about a machine that can talk to us and potentially take over the world. In fact, AI would rather just sell you stuff. AI is a little more boring, a little more saleman then Terminator. It is about changing the way we interact with machines and putting the cognitive burden on the machine. It’s a fundamental change in how weinteract with computers., We’re right in the middle of the transition. It’s an exciting time, and now you should have an easier time spotting AI in your everyday life…and watch out for your wallet