This element was created to provide a single-line text-input control to allow any characters for entering a query string. The query string was supposed to be sent to the server identified by the base URL and return a list of pages matching this query.
No, really, don't use it. This element is deprecated. It is not implemented anymore.
On November 1992, indexes as links rather than documents started by Dan Connolly who is pushing the idea that indexes are more links than documents. In this thread, different type of solutions are proposed. The question of forms for making queries is mentioned in reference to Dynatext browser: "The browser displays toggle buttons, text fields etc. The user fills in the fields, clicks OK, and the query results come up in the table of contents window."
A thread about isindex in November 1992, Kevin Hoadley questioned the need for an isindex element and proposed to drop it. He proposed to have instead an input element (idea supported by Steve Putz). Tim Berners-Lee explains the purpose of isindex resulting in aggregated search results. Kevin replies that he doesn't like the boolean nature of isindex and would prefer a system where everything is searchable and proposes to extend the current WWW Framework with a specific httpd configuration and defined that some URIs mapping create search queries.