Tuesday, July 30, 2013

predictive search

While I understand why technophiles would think this is an exciting possibility, I doubt if it will catch on in the near future. Basically, it would have to be really really good to not be annoying. No matter how good the program is at snooping and how expert the algorithm is at piecing stuff together, it's still going to misread the signs on occasion. And that is what will kill its appeal for most people.

We already have a lot of programs that try to guess what we want. Think of autocorrect or all those annoying functions in Microsoft Word. Both have spawned an entire subculture of complaints. The problem is that even if the computerized guesswork is mostly good, the user is only going to remember the times it fucks up. Despite all my bitching, autocorrect usually guesses right, probably 95% of the words I type. But I tap out a lot on my phone, so that 5% failure rate comes up every second or third sentence. Although the technology is highly successful with a 95% accuracy rate, the user feels like it is failing all the time. Even if the screw up rate was significantly reduced, to only 1% or a half percent, it will happen too often for the user to feel like it is a success.

A predictive search app like the ones described in the NYT article have to be on all the time. After all, the idea is that it will run constantly in the background and remind you of stuff that you haven't even thought of asking about yet. But always on means that any small chance of error will quickly pile up. Unless it is virtually perfect, the user will suffer through constant reminders of its shortcomings. Which is why I don't think it will fly.