What Is Artificial Intelligence?
How chatbots talk: predicting the next word
Summary
Artificial intelligence is the general name for computer programs that do jobs which need human intelligence, such as learning, language, seeing or making decisions. Today's chatbots are "language models" trained by reading enormous amounts of text, and at heart they do just one thing: predict which word is likely to come next in a text. Done on a large enough scale, this simple-looking job brings out surprising abilities; but because the models produce what is likely rather than what is true, they can also say wrong things with confidence.
In depth
The term "artificial intelligence" was born with the Dartmouth meeting of 1956. Early approaches tried to imitate intelligence with hand-written rules; the 1966 chat program ELIZA is a famous example. Running into the limits of writing rules, the field gradually turned to machine learning, in which the computer learns by itself from examples. After 2012, with big data and powerful processors, deep neural networks made breakthroughs in recognising images and speech. The transformer architecture, introduced in 2017, opened the way to large language models. These models split text into small pieces and work out the probability of the next piece by looking at the earlier ones. Claude Shannon tried the same idea in 1948 by counting words; the tiny model in the interactive uses Shannon's method.
Rate What Is Artificial Intelligence?
What did you think of this content? Click the stars to leave your rating.
Leave a Comment and Rating
Sign in to comment and rate. Membership is free.