Google's quest to build the ultimate machine-learning system has produced some lessons of its own.
The project, code-named "Seti" in a nod to the search for life in outer space, is being used on huge data sets in an attempt to solve what Google calls "hard prediction problems." Google revealed the name of the project in a blog post Tuesday that also acknowledged the limits of engineering.
Machine learning is a favorite topic of Google co-founders Sergey Brin and Larry Page, and is useful for improving translation algorithms and semantic understanding, according to Google. But, of course, it's an enormously complex notion, one that can intimidate even Google's computer scientists that could make use of such a system.
In designing the system, Google's Simon Tong said the company realized if it was going to get its researchers to use a machine-learning system, it needed to be simple, even at the expense of accuracy.
"It is perhaps less academically interesting to design an algorithm that is slightly worse in accuracy, but that has greater ease of use and system reliability," Tong wrote. "However, in our experience, it is very valuable in practice."
And while many may think machine-learning systems are a perfect solution to complex problems, sometimes they cause more trouble than they are worth. Google's Peter Norvig, director of research at the company, was one of those skeptics as recently as 2008.
"We saw very early on that, despite its many significant benefits, machine learning typically adds complexity, opacity, and unpredictability to a system. In reality, simpler techniques are sometimes good enough for the task at hand," Tong wrote.