Daniel Rocha <danieldi...@gmail.com> wrote:
> There isn't,though, any meaningful advance in AI for decades. What happens
> is the massive feed of data for very repetitive tasks in a very narrow
> field of expertise.
Your information is out of date. It is true there was little progress for
many decades, and for a while in the 1990s and early 2000s, progress was
mainly from massive databases. That is how Google made progress in machine
translation, for example. However, in the last 5 or 10 years, there have
been breakthroughs in neural network technology. It has been improved with
multi-level networks, where one network produces output for another, which
passes output to another, and so on. This has resulted in things like AI
programs that recognize a photo of a cat taken from just about any angle,
and programs that can even synthesize a generalized image of a cat. These
programs work much better than previous ones. Previous ones would look at
something like a meaningless pattern of stripes or dots and indicate that
was a picture of a giraffe (or some other off the wall choice).
Neural networks go way back in AI research, but in previous implementations
they were only one layer deep. For a long time, the neural network approach
was more or less abandoned. Until someone thought of the multi-level
The results of the new approach are dramatic. They include things like the
program that beat the world's best Go player, and a dramatic improvement in
the quality of Google translate. Do a search for:
google translate improved with neural networks
. . . and you will find many articles about this. Such as: "The Great AI
The article quotes a translation of this sentence in Spanish by Borges:
Uno no es lo que es por lo que escribe, sino por lo que ha leído.
The old Google translate system rendered this:
One is not what is for what he writes, but for what he has read.
The new one:
You are not what you write, but what you have read.