Google Instant Previews evoked some discussion in the SEO world when it was launched officially. However, as far as I know, this discourse was limited primarily to web design and landing pages. Even so, some writers tried to link Google's Panda Update to Instant Previews, which was presented at about the same time. The ground for this linkage was user experience, which both updates had focused on.
Now, one may argue that Instant Previews is simply a search aid to enhance user experience. However, as studies in search behavior confirmed over and over again, most users ignore such tools probably due to short term memory overload. This was probably the motivation behind Google's minimalist design.
Moreover, Google had avoided the inclusion of a previews tool in its homepage, although other search engines had suggested previews years before. Thus, there must be another reason for Google to add a column that occupies the same width as the search results column for Instant Previews.
Although similar tools had preceded Instant Previews, Google was the first to combine visual previews with relevant textual snippets (AKA "call-outs") on the fly. In addition, when officially announced, Google advanced its benefits in determining web-page layout before clicking on a search result. Although the layout in itself may help to identify web-pages with thin content or the web-page medium, when combined with text call-outs it may benefit the users by positioning their keywords in the page layout.
At the same time, Instant Previews may be used for Google's quality control. In other words, if the user avoids entering a web-page after looking at the page preview it probably means that the users keywords appear in marginal parts of the page layout. Of course, the user may avoid a web-page because he is intimidated by some formats like forums or comments sections. At the same time these formats may represent an unrefined data that may have less value than refined information for the average user.
Extracting the main sections from web-pages may be helpful for web-indexing where text is mixed with ads, navigation, and comments sections. (Moreover, websites owners may use different layouts in the same domain.) Therefore, getting an implicit feedback from users may help Google to improve their indexing algorithms to locate keywords that are placed in strategic positions. Moreover, this may represent a gradual progress from bag-of-words to semantic indexing algorithms.
The Semantic Web
Although the semantic web vision poses some serious technical issues, it seems that it had made a gradual progress, at least in the web-standards issue, namely HTML5. However, there are still some unresolved problems like how to deal with writers' semantic-web self-annotations in terms of valid phrasing and reliable tags utilization.
Google has stated it wont parse HTML5 markups until they will be widely used . Google has further stated that it would support only the tags that benefit its indexing system. Still, as a leading web-search company it is probably experimenting with HTML5 at some level. (Incidentally, Google parses RDFa semantic web annotations which is now a part of HTML5). Now, Google's Instant Previews may serve as a reliability test specifically for HTML5 "structural" tags. Thus, I can think of some current applications for Instant Previews: limited testing, evaluating semi-supervised algorithms, and evaluating unsupervised algorithms.
First, Google may use Instant Previews to evaluate which HTML5 tags may predict the structure of a web-page and their frequency in common practice, as well as to possibly evaluate algorithms for recognition of unreliable annotations. Second, Google may use HTML5 as a markup language for semi-supervised learning algorithms. In this case, Instant Previews may serve as a reliability test for its automatic annotation algorithms. Third, as some of Googles employees had declared, they may prefer to compliantly ignore any markup language and interpret web-pages as-is with unsupervised learning algorithms. In this case, again, Instant Previews may be utilized to evaluate the reliability of these algorithms or even incorporate them into the machine learning algorithm. Of course Google may test more than one of these strategies simultaneously.
Search Tool or Test Tool?
As Google states, it constantly tests its users response to possible updates in the search engine algorithms and the user interface. Now, there is at least one disclosed example where Google used a search tool to test user experience. When Google discussed its new automatic spelling correction feature, they exposed that it has included an escape hatch for both users and Google" let users search their original query and in turn provide Google with feedback to help evaluate its algorithm. Actually, the idea of implicit feedback is the heart of many new personalization algorithms. Although, in this case, extra search aids are not necessary.