By: Ace Green, Reporter, Production Manager
When discussing the greatest technologies to emerge in the coming years, it’s difficult to avoid Artificial Intelligence. It seems to be on the news all the time, with new breakthroughs constantly being headliners. It’s rarely mentioned, however, what Artificial Intelligence is, much less how it works.
In short, Artificial Intelligence is the application of self-developing programs. These programs are able to identify patterns in data and use these observations to do things they weren’t necessarily programmed to do.
There are thousands of applications for AI, but today, we’re going to focus on a simpler and more common utility; Image Recognition.
Image recognition is far too complex to be programmed by humans. There are over nineteen trillion possible 1080p images, meaning it’s virtually impossible to program an image identifier by hand or even with an algorithm. Instead, one is able to develop Artificial Intelligence to recognize what the dedicated image is.
Let’s say we’re going to create an AI called FlowerNet, which will try to tell whether an image includes a flower. Assuming we already have a framework to read images, we would begin by giving FlowerNet thousands of images along with whether the image includes a flower.
Once the AI has a pool of images it knows to be flowers, it can begin to identify patterns within the images. Perhaps FlowerNet realizes that the images with flowers usually have green pixels arranged in the shape of a stem and that they often have a colorful top to them. Perhaps it discards information that it doesn’t think is helpful. This process is called ‘Training’ an AI.
You might be wondering, though, how can you verify that FlowerNet actually knows what is a flower and what isn’t? Just because FlowerNet can recognize patterns doesn’t necessarily mean that it recognizes the correct patterns.
Well, just like with people, once an AI is trained, it must be evaluated. For FlowerNet, we will give it images it has never seen, and tell it to respond with whether the image includes a flower. You’ll then see whether the AI correctly identifies flowers or not.
If they do, then great! FlowerNet is online. If not, then we’ll have to repeat the Training process again, usually with some more data each time. Most AIs need between ten thousand and one million training cycles, so you generally have to automate the evaluation.
And there we go! FlowerNet can be released to the public, and you’re officially an AI expert.
Edited by: Seth Birdsong