Google AutoML Beats Researchers: AI Writes Superior Machine‑Learning Code

Source: cnBeta (link: https://cnbeta.com/articles/tech/661493.htm). Original article: https://thenextweb.com/artificial-intelligence/2017/10/16/googles-ai-can-create-better-machine-learning-code-than-the-researchers-who-made-it/

Google’s AutoML system recently produced a series of machine‑learning code, its efficiency even surpasses that of the researchers themselves. This clearly strikes another blow to the “human superiority” theory, as robot “students” have become masters of self‑replication.

AutoML was developed as a solution to the shortage of top AI programming talent. The team proposed a machine‑learning software that can create self‑learning code; the system runs thousands of simulations to determine which aspects of the code can be improved, and after each change continues the process until the goal is reached.

This is a perfect demonstration of the “infinite monkey theorem”, but Google did not let a monkey type Shakespeare; instead it built a machine that can self‑replicate programming, and these machines in a few hours perform better than human programmers working weeks or months.

Although it sounds a bit scary, AutoML indeed outperforms its creators in programming machine‑learning systems. In an image‑recognition task it achieved a record 82% accuracy.

Even in some complex AI tasks, its self‑created code is superior to human programmers. It can label multiple points in images with 42% accuracy, whereas human‑built software reaches only 39%.

Of course, this does not represent a “Skynet” or a creepy “digital ghost”, because we are not yet at the singularity of self‑aware machines; we have simply added another boost to AI’s technical potential.

Google announced AutoML only five months ago; given its ability to produce a better AI system than its own researchers in such a short time, the results of the next year are clearly worth looking forward to.