Diary of a Blind American

Ball Vision

Entry #59.0 July 9th, 2026

4 Balls image rendered by Google's Gemini
4 Balls image rendered by Google's Gemini

How can a computer see? How does it know what a ball looks like? Let's answer these questions as best as I can.

If you want to know how traditional film cameras work and how those photos are produced I am the wrong one to ask. But let's talk about how a computer displays a photo. Your phone is a computer let's establish that first. Second, a photo being displayed on a computer is a grid of very small dots called pixels. A rectangle grid of very small square pixels. So small, the square pixels can look like round dots. Remember the displays of the 80s where things were a lot more blocky! Those pixels were a lot bigger back then. Each pixel has a color value. Depending on the format that the picture is in determines what values are present. Different formats can display different values. For example a PNG, or portable network graphics Image can have transparent pixels. This is largely what the graphical user interface of your phone is made up of.

Now that we've established that, let's talk about how a computer can see what's in a photo. It does not have eyeballs. But it can read the values stored in a grid of pixels. You can train a computer that when it sees a specific pattern of pixels that more than likely this object is a ball. Once you've trained a computer to identify a ball, then teaching it to generate an image of a ball is possible. (making that ball look realistic is the difficult part.) From there you can extrapolate out how all of this stuff works. You teach it the color red. Then all of a sudden it can produce a red ball. Teach it the patterns that are present on a basketball and all of a sudden it knows how to identify and draw a basketball. Then of course you have to teach it that Americans are determined to feel different... so a ball does not necessarily mean it's a ball shaped object So you have to specifically teach it what a football (American version) is. Because Americans have to be different. lol No I am not anti-American. I just notice things like this.

Written by B. On July 9th, 2026.

Notice I did not use the word artificial intelligence once in this. you can now see why machine learning is a much better term. We just figured out how to scale this technology using something called a data center and really fast processors and memory.