Robert_Charlton - 3:07 am on Dec 7, 2011 (gmt 0)
I should have said that I believe that one of the reasons those training wheels are there is for mobile devices. Google is throwing a huge number of considerations into that one search box.
My observation about the quality of results with auto-complete, at least in this particular query, had to do with how I think Google classifies data into a virtual model of searcher behavior. The auto-complete structure, I'm guessing, provides evidence of an underlying structure that Google uses to more efficiently organize queries and to correlate them with its index, and with user engagement and the rest of the algorithm. It's likely that Google is using the training wheels to guide us to statistically-determined "most satisfactory" results in the shortest amount of time.
But in the case of AI, even the training wheels need training. In longtail searches, I've observed that when you go outside of the auto-complete suggestions, Google falters. Here, the auto-complete may be not so much a feature as simply a reflection of a limitation. AI needs lots of data. When an area of the index is infrequently searched, the longer it will take to train the AI engine to determine where the problems are and to apply algorithmic fixes. This search is one such problem area.