Following a few earlier papers on the same subject, in 1964 Eugene Garfield set out his vision of a unified Science Citation Index as an integral part of the World Brain, a descendent of Vanevar Bush's Memex. Almost 40 years later, how far have we come?
Currently, the closest resource available is ISI's Science Citation Index, sold as part of ISI's Web of Science, which is part of ISI's Web of Knowledge. Searching with Google for 'science citation index' leads you to a closed door, with a link to the product website.
This website contains no obvious information about how to access the Web of Knowledge, but advertises that "ISI Web of Knowledge brings promising research to light". Clicking 'See how' opens a Flash animation, that leads to a page containing, amongst other links for downloading PDFs, a button for requesting a trial - this leads to an online form where you can send off your information and wait.
While you wait, if you're lucky enough to be affiliated to a UK institution, you can access the Web of Science through a portal at MIMAS. To use this you need a username and password for the ATHENS authentication service. You need to fill in a copyright acknowledgement form and take it to your library, who give you a username and password. You use this to self-register with ATHENS, where you choose a personal username and password. You can then use this to log into the Web of Science at MIMAS.
Once inside the Web of Science, you choose which index to search, and whether to perform a general search or search for a particular reference. Choosing a general search opens a search form, the default response of which (if you press enter) is to clear your entries and return to the start page. From a successful search, selecting an entry title gives you the abstract, but no link to the full text.
There are also the nice features for which the citation index was designed: 'show cited references', which shows a list of all papers cited by this paper, and 'related records', which uses a similarity algorithm based on finding other records that cite the same references. There's also a count of the number of times each paper has been cited.
That's it. You can mark references for export and send them to Endnote, which works nicely, but there's still no link to the fulltext. The URL in Endnote leads to ISI eSource, which is the same as the Web of Science, only you have to pay $30 to be able to view the records for 10 citations.
All said, I don't think this is quite what Garfield had in mind. I think he would have liked CiteSeer though.