What This Article Covers
When you think of generative AI, what comes to mind first?
Many people probably think of AI that exchanges text with us, like ChatGPT.
Today, AI can create images, make music, generate videos, and even control robots. Yet for most people, the first thing that still comes to mind when they think of AI is “AI that can have a conversation.”
Why did generative AI begin by learning language before images?
People Communicate with AI Through Language
People use language to communicate with one another. The same is true when people communicate with AI.
In the past, you needed to know a programming language to make a computer do something. But the emergence of generative AI changed that.
Now, you can ask questions and make requests naturally without having to write complex programs.
In other words,
Human ↔ AI
The most natural way to connect the two was through language.
That is why AI first needed to understand language in order to understand people.
Language Was the Richest Source of Training Material
Language is also the most extensively accumulated forms of information in the world.
Most of the knowledge we have created has been recorded in text.
- Books
- Newspapers and news articles
- Academic papers
- Online posts
- Blogs
- Wiki pages
This enormous amount of text data provided excellent material for training AI.
Of course, in AI training, it is not only the amount of data that matters. How well the data has been digitized and how well it has been organized in terms of quality are also very important. As we saw in #07 with the example of Korean, AI performance can vary even within the same language depending on the quality of the data and the digital environment.
Images and videos, on the other hand, require much more storage and are also more difficult to interpret in terms of meaning.
For this reason, generative AI began by learning to understand language using the richest available text data, allowing language-based AI technologies to develop the fastest.
Starting with Language, AI Is Gradually Learning to Understand the World
At first, AI emerged that could understand and generate text.
Then came AI that could create images,
understand human speech,
generate videos,
and process multiple types of information at the same time.
More recently, AI has emerged that can perceive the physical world through cameras and sensors and take direct action within it.
The development of generative AI can therefore be seen not as a series of entirely new forms of AI suddenly appearing, but as a process that began with language and has continued to expand the range of information AI can understand.
PART 2, which we will explore from here, follows this very progression.
DANA NOTES in One Line
Because people communicate with AI through language, generative AI first advanced rapidly in technologies for understanding language. Since then, AI has continued to expand toward understanding images, speech, video, and even the physical world.
In the Next Article
In this article, we explored why generative AI began by learning language first.
Then how did AI move beyond language and begin to understand visual information such as images?
In the next article, “How Does Image AI Create Images?”, we will look at how AI understands and generates images.
📖 Related Articles
#07. Why Does AI Struggle with Korean?
