From the course: Introduction to AI-Native Vector Databases

Human-understandable versus machine-understandable data

From the course: Introduction to AI-Native Vector Databases

Start my 1-month free trial Buy for my team

Human-understandable versus machine-understandable data

“

In the previous video, we introduced some differences between structured and unstructured data. Now let's think about how all of these unstructured data types we've discussed previously, text, images, audio, and video can be understood by humans. Words in a text document can be read and understood one after the other. Images stored as pixels of colors stacked on top of each other are understood as objects interacting with each other. Audio files, stored as sound intensities and frequencies can be listened to and understood as music or audiobooks, and et cetera. Videos are time evolutions of lots and lots of images that can be understood when played forward in time. This is how we consume media stored in data files through our eyes, ears, and might. How do computers understand the same data? Let's examine one by one. An image can be thought of as millions of pixels stacked on top of each other, each having a unique number value contributing slightly to the overall picture. We can string these pixels out and understand these as a trail of numbers, one beside the other. We can then perform mathematical operations on these pixel values to capture visual features of what's in the image itself. These visual features are themselves numbers that can be stored in a computer, and used to understand and summarize what's in the image. A text document can be turned into pieces of words, each of which gets a numerical ID. We can then capture the meaning of words by looking at what other words it's used with. This is known as the co-occurrence of words and can be used to understand their meanings. For example, if I say three sentences, the tiger played with her cubs versus the jaguar played with her cubs versus he put the lettuce into the salad. Here due to similar words co-occurring, you can understand that a tiger and jaguar are more similar to each other than to lettuce. We can do the same for audio and video files, each of which can be represented as numbers for the computer to understand, and then these numbers can be summarized into smaller groups of numbers, having to do with the actual content, patterns and concepts found in the data. Computers understand all data as these groups of numbers, and these groups of numbers are known as vectors. Vectors capture the meaning behind the data, allowing a computer to understand data similar to how we do. For example, to a human, you would say, pass the salt, please. But to a computer, you'd have to say pass the 1.2, 1.1, 6.2, 1.3, 9.8, please. That set of numbers is what a computer knows salt as. They understand this vector representation better. Similar to when translating from one language to another, we want to make sure our data makes sense to computers as vectors, and we want to ensure that, if possible, no meaning is lost in translation when going from human understandable data representations to machine understandable data representations or vectors. So, to summarize, vector databases use vector data representations to understand data. That's why they're called vector databases. In the next video, we'll talk about what these vectors actually look like. See you there.

Contents