To find a page all the user should have to remember is something about the page itself (maybe how it looked, what's in it, how big it was, when it was saved, what else was saved at around the same time---or just before---or just after, what other pages contain stuff that it might also contain, and so on). In other words, the user must remember the value of at least one of the page's attributes. Without that the user can't find the page.
If the user fails to find a page but then stumbles on it later, the system needs a new attribute to capture the type of information the user was able to remember during the unsuccessful search---assuming that that information is capturable with today's technology: for example, it isn't presently reasonable for the user to ask the system to create a page attribute like "I was smiling when I last looked at this page."
The system should also adapt its notions of which attributes are most important to the user based on which attributes are most often used in successful searches. Consequently, after each search, those attributes that were most discriminatory during the search should have their weights increased.
The system should have multiple search modes to allow for at least two distinct search problems: searches where the user already knows the page is on disk somewhere and searches for pages that the user doesn't know much about at all.
The system should let the user search in at least three ways:
by category,
Search by category is similar to the search strategy we use today when we drill down a hierarchy of directories except that the user can get to a particular page by multiple paths (a page can belong to multiple categories, and categories can belong to multiple other categories). So even if the user remembers only one page category that the page belongs to, the user can probably still find the page.
Search by attribute is similar to the search strategy we use today when we do an explicit search for a page using a finder tool (grep, find, or glimpse on Unix, the Finder on the Mac, and Find File on Windows), except that arbitrary attributes can be used instead of the tiny few allowed today (size, date, name, location, and type). Attribute search seems to be mainly textual but perhaps it could be made more graphical. Further, it is iterative so that the results of previous searches can be used as the basis of current searches. Thus search becomes dynamic.
Search by example has no cognate on today's desktop, however a few search engines provide a "find similar pages button" that does a roughly similar job. This kind of search lets the user hint at a category or attribute by choosing a specific page that belongs to that category or has that attribute and subtracting or adding other pages to get intersections or unions, which the user can search dynamically.
One possible search by example mechanism might be to use the page icons themselves as searchers. Clicking on an icon narrows the focus to icons of other pages that are like that particular icon. Clicking on another icon narrows the focus still further. Each selection shrinks the set of icons that are currently displayed, until the set is small enough to go through individually. This process requires no explicit booleans, no typing, no endless navigation, no arbitrary and fixed hierarchy, and no memorization of names. It's not clear, however, how to dynamically choose the attributes that are to be used to measure similarity at each stage of this kind of search.
Of course, all three types of searches can be mixed. (Note the similarity of category search and attribute search to the two main ways of accessing information in books: indices and tables of contents. Electronic searches are far superior, however, since they are not as limited by space nor as they as onerous for the user.)
