Our expectations of AI are a little unfair, mostly because of what we’re fed by movies, shows, and other forms of science fiction. Those AI are all seamless, carry more or less human intelligence, and perform tasks as if you told an actual human being to do them. The real thing might not be as cool, but it’s quite a miracle in itself. There is little doubt that AI will become supremely competent, it might even develop sentience, and as such pose a threat to our existence, but not quite yet.
Now, before you decry this article as a scare piece warning of the dangers of AI and the impending extinction of humanity at the metallic heel of our robot overlords, allow me to comfort you with the assurance that AI is a good thing in its current state, and something you should understand and harness, rather than fear.
You, like many others out there, may be wondering how AI works. The long answer requires a few years of post-grad study, which we aren’t sure you have time for right now, so we’ll give you the abridged version. There are a lot of facets to the whole AI phenomenon, so we can look at a few in turn.
Before anything else, we have to define what we’re talking about. The shortest way to describe it might be a program that allows machines to perform tasks as a person would. Now, this is terribly loose, but for the average consumer, it’ll probably do. So, your computer’s spreadsheet isn’t exactly AI, as it’s just an automated calculator, but something like AlphaGo, which defeated a human Go champion, is.
If your smartphone is of a fairly recent generation, you’ve probably played with its own AI features. Siri, Google’s Assistant, and Microsoft’s Cortana are all AI assistants that, when prompted, listen to your speech, and perform a task. Pretty easy, right? Well, not really.
There are a few things that need to happen before your AI assistant sets your alarm or something, and it doesn’t all happen on your device. If you’ve used your phone’s AI a lot, you’ll notice that it doesn’t work when it’s not connected to the Internet. For the assistant to know what you want it to do, it has to distill the speech into the key ideas behind it, so it knows what to do. For humans, as long as it’s heard clearly, and in a language we understand, it’s more or less automatic. You might not realize this, but that requires a lot of specialized hardware inside your brain. Your phone, while incredibly adept at performing calculations and other math-related processes, isn’t quite designed to handle human speech, so it requires more compute power than it has right on the device.
Because of the wonders of the Internet, there’s a clever work-around. Your phone records the voice command, and sends via your telco or internet provider, to the cloud, where more vastly more compute power is available to do a few things, over a few layers of processes.
Firstly, it needs to figure out exactly what you said. If you’ve ever had to transcribe a voice recording, you’ll know that’s not the simplest thing to do. Once that’s done, another layer takes the words, and makes sense of what each word means, and more importantly, the relationship of the words to each other. That’s quite high-level stuff, and for someone who doesn’t have a degree in linguistics or AI, it can appear to work like magic. It’s hard to believe otherwise, to be honest.Without coaching, it knows that “I need to fly to Berlin next week,” means it should show you available airline tickets to Berlin from the start of next week, just by looking at the words in the sentence.
Once it knows what you want, it will then hand it over to another layer that does an appropriate web search, or system trigger, or whatever else it has authority to do, and gives you feedback on the same, which itself is a layer that takes information, and relays it in a fashion that sounds convincingly human-like, as opposed to something that’s more like a traffic bulletin. Tell your actual human assistant to do that, and he or she has years of experience making calls, searching for flights, and reasoning which options fit your parameters. AI does it without a physical body. Let that sink in for a minute.
Amazon’s Alexa is one of the more impressive examples of this technology. Smooth, convincing, and consistent, it is able to understand, process, and decide on actions based on user voice inputs. It’s not quite the AI future we know in fiction, but if you don’t actually feel like you’re talking to an AI assistant as you tell it to order groceries, you have some very high standards.
AI as a whole generally works in that fashion. Take Google Photos, for example. If you haven’t tried, head over to your Photos app, and search for “cat.” It should go on to find all your pictures of cats, all without tagging, or any input from you, the user. While it seems like magic, it’s all about layers of extremely specific programs, each doing their thing, and passing the request off to the next layer.
There is, for example, one layer that only looks at the edges of the elements in the photograph. Once the elements are identified, another layer looks at color, and hands it off to the next layer, which then looks for, say, eyes. There are layers on layers of little programs that look at the different parts of the photo, each grading it with a score of how close it is to what it knows a cat to be. Does it look like it has fur? Might be a cat. This photo has wheels. Probably not a cat. And so on. Think of it as a series of sieves, and as it goes through, images that don’t have elements that are aligned with “cat” get filtered out.
In a nutshell, this is how AI works. There are other much more complicated elements to it, such as deep learning, which is as disconcerting as it is exciting, as it more or less parallels how our own brains work. This concept has been around for a long time, but it has always been very computationally intensive, and has really only recently been a feasible approach to making true AI by teaching it to learn. This is where the future of AI lies, and something we have to watch with equal parts rigor and care. It seems inevitable that one day, AI could really be the smartest thing on Earth.
Also published in GADGETS MAGAZINE September 2017 Issue
Words by Ren Alcantara