本文テキスト
06. Artificial Intelligence
AI, or artificial intelligence, is widely used these days.
When we search for information by keywords on the internet, AI summarizes the information for us.
When we want to read information written in a foreign language, we can use an AI system to translate it into a language we understand.
We can also ask AI systems to draw pictures or create music.
The idea of AI actually has a long history.
A practical AI system was developed in 1989 and used in post offices in the United States to read handwritten ZIP codes on postcards and envelopes.
This system was quite primitive compared to modern AI systems, but the basic idea is the same.
AI systems are designed to do a lot of simple calculations on input data, like photos of handwritten numbers or keywords we use to find information.
AI systems contain parameters that affect the results of these calculations.
They make use of a huge number of parameters in order to do complex tasks such as reading handwritten numbers or summarizing information in response to keywords.
AI systems cannot do anything useful right after they are created.
They need to be trained.
When engineers train an AI system, they give it some input data.
Then the AI system does lots of calculations on the data with its parameters and produces a result.
At first, the result is usually unsatisfactory.
The AI system then adjusts its parameters and does the calculations again.
This process is repeated again and again until a satisfactory result is produced.
AI systems are trained in this way before they are used for practical purposes.
Modern AI systems understand human languages.
This means that they can handle meaning, but how?
AI systems work by doing a lot of calculations, so they convert meaning into numbers that they can calculate.
Scientists have found out that a unit of meaning can be expressed as a vector.
The dimensions of the vectors used in practical AI systems are often very high, sometimes over 1,000.
Similar units of meaning are converted to similar vectors.
The vector for the meaning of "king," for example, is similar to the vector for the meaning of "queen."
Either of these vectors is quite different from the vector for "chair" or "apple."
In order to understand human languages, AI systems also need to handle the context.
Words can mean different things in different contexts.
For example, when we are talking about how to use a computer, the word mouse usually means a computer device.
On the other hand, when we are discussing different kinds of animals, the same word mouse means a small mammal.
AI systems can understand this difference by analyzing all the vectors that correspond to the words in the context.
Thus, AI systems understand human words and their contexts with the help of vectors.
When AI systems answer our questions in a human language, they do calculations on these vectors to choose words that flow naturally.
It is amazing that AI systems can do a lot of useful things by doing calculations on numbers.
Each calculation is simple, but it is possible to do complex tasks by doing numerous calculations.
This is in a sense like our brain.
Our brain contains a lot of neurons.
Each neuron only does a simple thing, but numerous neurons work together so that we can do complex things.