Why We Struggle Understanding AI ?
--
In just the past 12 months, artificial intelligence (AI) seems to have gone from being a fringe emerging technology into one of the most popular business buzzwords of the year. Everybody, it seems, is trying to develop a new AI-powered solution. Or, if they’re not doing it themselves, they are partnering with someone who is. But that doesn’t mean AI is easy. In fact, AI is hard. Very hard.
What is AI, really?
One reason why AI-powered solutions are so difficult to pull off is because there are so many different definitions and interpretations of what the term “artificial intelligence” really means. Even the mainstream business media struggles with the term, despite writing article after article about AI. How many times have you clicked over to an article, expecting to read a compelling story about the latest AI innovation, but the article gets bogged down into a point-by-point analysis of how “deep learning” differs from “machine learning,” or how “machine learning” is different from “neural networks.” The distinctions may seem subtle or nuanced to the casual reader, but for the people actually building the AI systems, they are very different. You can immediately see why this leads to so much confusion at the outset of an AI project.
Machines learn differently from humans
Another reason why AI is so hard has to do with the way machines and humans learn. To assume that machines learn the same way that humans do is to make a fundamental mistake. If you show a three-year-old child a picture of a dog and a cat, you can pretty much be certain that they will never mistake the two animals, no matter where they see them or under what conditions see them. But that’s not true with machines. Have you seriously ever tried to train a machine to recognize anything? Even if you show a machine 999 images of a dog, there’s no guarantee that the machine will be able to recognize the 1,000th image as a dog. And, yet, in the same amount of time that a machine is trying to figure out whether the image of your pet poodle is actually a dog or a cat, it can solve mathematical and computational equations that would take a single human thousands of years to solve. Funny how AI works, right?
AI comes with a lot of baggage
Let’s be honest, Hollywood’s dystopian views of AI don’t help, either. Doesn’t every new Netflix movie seem to be based on a premise that intelligent machines will eventually take over from humans and enslave the entire human race? When the typical executive in Corporate America hears the word “AI,” they’re probably thinking of some kind of Terminator robot powered by Skynet. When you try to tell them that, no, no, no, your AI solution is really just a very clever algorithm written in computer code that can do mundane things like recognize a photo of your aunt Annie on Facebook, it’s always fascinating to see what the reaction is going to be.
Garbage in, garbage out
And, finally, it’s important to note that machines learn using data. Data is what powers algorithms. Data is the oil that fuels the AI machine. So what happens when the data used to train machines is flawed? So many data scientists and others building next generation AI solutions actually spend a good amount of time scrubbing the data, cleaning it up, and putting into a format that your computer can actually use.
So, yes, AI is hard. As long as you keep that in mind at the start of any new project, you’ll understand why we’re still a long way off from any of those futuristic super-intelligence scenarios that Hollywood keeps trying to foist on us.
Hey! I’m Tomer, an entrepreneur and maker. You might know me from Mevee, Crane, and Shots, among other products I’ve launched! This article is a part in a larger series I’m writing mostly based on my experiences, and is largely made of my and my team’s opinions.
I hope this helps you to avoid making the same mistakes I did, and remember to keep shipping!