As AI technologies are taking over, individuals and companies are trying to figure out what exactly AI is and how this technology can be incorporated into their day-to-day operations.
Let’s focus in on Generative AI. As the same suggests, Generative AI can generate new, original content such as text, images, audio, music or even video all with a simple* user prompt. By using a source input such as text or an image, the generative AI has learned how to interpret that request by studying hundreds of thousands of pieces of data.
Simple right?
Okay, let’s break it down. Let’s say you’ve never seen a horse before. At the beginning, you have no idea what a horse is. But then you are shown a handful of pictures of horses (maybe forced to watch a few episodes of Gunsmoke) and then asked to draw a horse. With your new-found knowledge of horses, you should now be able to replicate an original image of a horse.
Depending on your skills, it may not be the best image ever, but it probably generally looks like a horse. But don’t feel bad that it isn’t an exceptional piece of art, that is exactly how early forms of generated AI used to work, as well.
As machine learning algorithms, particularly deep learning neural networks, have become more sophisticated, so have the quality of content AI is able to produce. Modern algorithms, like Generative Adversarial Networks (GANs) are capable of analyzing and reproducing intricate details leading to more realistic images.
In essence, they have been taught what a horse is.
These GANs are not just using basic images to teach AI what something is, they are going much deeper by explaining every detail of the content. As large-scale datasets became more available, AI models are able to be trained on a much larger range of data.
Imagine you are that artist that could only draw a vague representation of a horse based on being shown several images. Now imagine you have been trained in art by the world’s greatest artist. You have studied anatomy of horses, the history of every breed of horse, read every book every written-on horses, watched every episode of Gunsmoke, developed a deep understanding of color, lighting, shading and whatever the heck a forelock is.
You should now be able to create an original work of art that actually looks like a horse (or a person with a horse’s head if that’s what your asked to do, assuming you have also been taught everything there is to know about human anatomy, clothing types, race, religion, ethnic variations of features, and whatever they heck a popliteal space is.
This is what Generative AI is. As technology advances and the data sets used to teach AI grow, the output of the AI can become more sophisticated. AI is able to take what it has learned and not just deliver regurgitated responses but can use its knowledge to create original content that other wise didn’t exist.
Is AI just compilations of images across the Internet?
Not exactly. While AI is trained with images, it doesn’t just search the Internet and morph different images together. AI simply learns from those images.
As with the horse example, AI has learned from these images, videos, audio files, and any other data that it is given.
AI can not only create images of horses, but also images of cats, dogs, even people. But it goes even beyond that.
Because AI can be taught almost anything, advancements in generative AI have been seen in writing, programming, audio and video. Just in the past year, AI has advanced in many of these areas to be almost unrecognizable from the real thing.
This, of course, is what scares people and companies from using AI. An advanced technology that can learn and “think”? A machine that can do computations faster than even the most brilliant scientist? A computer that can take over the world with just a prompt? It all may be more real than you think . . .
Comments