Google Squared

» see google.com/squared as it lived, on the wayback machine

~ the obituary ~

Type "roller coasters" into Google Squared and you did not get ten blue links. You got a table. One row per roller coaster, columns for height and top speed, with a thumbnail in the first cell, assembled on the spot from pages all over the web that had never agreed to be a database. You could delete a row that did not belong or add a column it had not thought of, then export the whole thing to a spreadsheet.

The timing was not an accident. Google announced Squared on 12 May 2009, three days before Wolfram Alpha opened to the public with a much louder claim about computing answers rather than finding documents. Squared reached Google Labs on 3 June. The two products were aimed at the same anxiety, which was that a list of links looks primitive next to a direct answer, and Google could not afford to be the one still shipping the list.

It worked more often than it had any right to and it was wrong a lot. The extraction had no way to tell an authoritative page from a bad one, so a square would confidently report a wrong height for a building or a wrong birth year for a person, sourced from somewhere plausible. Google's own framing kept the escape hatch open: it was a Labs product, it was learning, your corrections helped. It stayed in Labs for its entire life and was never allowed near the main results page.

Google Labs was closed in July 2011 as part of the tidying that followed Larry Page returning as chief executive, and Squared was on the list. It went off on 5 September 2011. Saved squares were deleted, with an export offered to anyone who noticed in time, and Google said the technology would keep improving search from the inside.

~ the case file ~

~ the wolfram problem ~

Wolfram Alpha and Google Squared were answering the same question from opposite ends and both answers were unsatisfying. Wolfram curated: real data, checked by people, covering whatever the team had gotten to and returning nothing at all outside that. Squared scraped: total coverage of everything the web mentioned, with an accuracy that depended entirely on what the web happened to say.

The resolution came from a third direction. Google bought Metaweb in 2010, which brought Freebase and a structured graph of entities built by hand, and that became the Knowledge Graph in 2012. So the answer was neither pure curation nor pure extraction. It was a curated spine with extraction hung off it, which is roughly what every large model does with retrieval now.

~ the quiet reuse ~

The panel that appears on the right of a Google results page when you search for a person or a film is the descendant of this work. Same instinct: stop making the user open five pages to assemble a fact table in their head. It arrived with a different data foundation underneath and without the editable grid, which was the interesting part of Squared and the part nobody kept.

Squared's peak overlaps my working years exactly, and it was still a demo rather than a tool. Google Labs in that period was a page you checked to see what the company was thinking about, more like reading a research group's notice board than using software. I could not honestly tell you a single task I ever finished with Squared, which is true of almost everyone who tried it, and is why it was closed without anyone complaining.

@}-->--

              )
             (_)
            .-'-.
            |   |
            |   |
            |   |
            |   |
          __|   |__   .-.
       .-'  |   |  `-:   :
      :     `---'     :-'
       `-._       _.-'
           '""""""

« previous  ·  random  ·  next »