A couple of years ago, I posted about how google was becoming a command line for the internet age.

The great power of the command line is that it can provide access to a nearly infinite amount of functionality and information in a small amount of space. The classical limitation of a command line is that the means to accessing that information is opaque. The user must know before and the proper commands.

This limitation is addressed by the Google example in two ways. First, Google uses what can loosely be termed “natural language processing” to guess at what answers the user is looking for. Then, it uses the space afforded in a web page, and the navigational ease of a point and click interface, to offer those answers to the user.

SI’ve thought about the issue from time to time, considering other implications of the paradigm. One insight is that people are actually a lot more patient with poor AI if 1)it still helps people solve a big problem 2)it’s presented in a way that allows them to quickly discard the bad results and focus on the good ones.

This came up recently because I was reading about a machine learning application with a 20% false positive rate. My initial reaction was “man, that really sucks, not even geeks would use that,” then I realized that search often gives a far lower success rate, and yet it can be useful.

In anycase, I remembered all this because I came across a new utility called AppRocket from candy labs that seems to leverage some of these ideas to provide easy access to all the crap on your computer. I’ve just installed it, so I’m not sure how well it works, but its cool to see attempts to innovate in this area.