back to article Look out, Wiki-geeks. Now Google trains AI to write Wikipedia articles

A team within Google Brain – the web giant's crack machine-learning research lab – has taught software to generate Wikipedia-style articles by summarizing information on web pages... to varying degrees of success. As we all know, the internet is a never ending pile of articles, social media posts, memes, joy, hate, and blogs. …

  1. Teiwaz

    First law of computer programming

    'You can't get out more than you put in'

    Heard that on a Jason King/Department S episode at some point as a kid a long time before I took up IT as a career.

    I've a cousin who used to badger me a number of years ago about some way of getting a computer to write articles for his web page for him,

    A Classic example of the underpants...???...., profit scenario.

    I've not seen that many wikipedia articles which were mostly incoherent, at least as far as syntax was concerned....

    1. asdf

      Re: First law of computer programming

      Granted machine learning requires putting in a sh1t ton of data but you can get unexpected results which is why its fairly revolutionary.

  2. Pete 2 Silver badge

    Nothing a bit of editing can't fix

    > Overall, it’s still pretty readable. The text generation seems to work OK, in your humble vulture's opinion

    The "human" generated text focuses much more on the personality and biography of the guy who created the site and it tells us nothing of any use about the website itself.

    While the AI version has the feel of being written by someone learning English (the repetition, instead of short-form: "it", "the site", "they"). However, it provides much more useful and relevant information to inform the reader of what to expect if they visited the site.

    Once the AI overcomes its English as a foreign language issues it seems to me that its output will be of a higher quality than the human-generated version. I would suggest that, just as with real journalism, there is an additional AI-editor role needed, above just an AI text generation function.

    1. veti Silver badge

      Re: Nothing a bit of editing can't fix

      Reminds me of the 'AutoSummarize' feature that Word used to have. Except that, as I remember it, the Word version worked better than that. Less repetitive, and it knew what capitals were for.

      Anyone know why it got taken out from Word 2010?

  3. Pascal Monett Silver badge

    Change in article category title

    Please rename the "Artificial Intelligence" category to "Pseudo-Artificial Intelligence".

    That will make things more realistic, thank you.

    1. handleoclast

      Re: Change in article category title

      Please rename the "Artificial Intelligence" category to "Pseudo-Artificial Intelligence".

      And then put it in charge of the FCC? Already been done.

  4. Nick Kew

    It was a dark and stormy clock

    Opening line of a computer-generated story I encountered as a student in the early 1980s.

    It was surreal, but then we'd been brought up on Monty Python, the previous generation had had the Goons. This Very Silly Story may not quite have been Gogol, but was in a tradition going right back to Aristophanes.

    It scanned rather nicely, without the turgidity reported here. Which rather suggests that a more readable narrative style might be a solved problem, if it had been on these researchers' agenda.

  5. Rafael #872397

    It’s impossible to read and keep up with everything.

    Why would we want to keep up with everything? I used to read at least the headers of several online news sites, then noticed that except for op-eds most are copied from each other. Same with cats' memes.

  6. Anonymous Coward
    Anonymous Coward

    ..if such computers could condense information down into useful snippets...

    Something happened in the Middle East. Brough to you by Google and Carl's Jr.

  7. Brian of Romsey

    Punctuation ???

    Surely it would be more helpful to the world at large if these 'AI' efforts could turn their attention to punctuation. Taking the crap that people (and this 'AI') produce and formatting it for readability. You know stuff like no spaces before commas and full stops, one space after a comma, two spaces after a full stop, Oxford commas (now that the US courts see them as pretty darned useful), and capital letters at the beginning of sentences and on proper names. Once this is done all manner of help can be offered to real people - "fix up your Facebook posting for you Sir?", "Fix up your email Ms?", "Would you like to look less dumb on Twitter?". Just sayin'.

  8. Anonymous Coward
    Anonymous Coward

    Better than human

    As a long-time Wikipedia editor, I can only say that the average AI edit will probably be better than the average human edit. As long as Google doesn't allow it to learn paranoia, rudeness, stubbornness, stupidity, bias, deviousness, dishonesty and all those other loveable characteristics of so many of my fellow human Wikipedians.

    1. Wensleydale Cheese

      Re: Better than human

      "As long as Google doesn't allow it to learn paranoia, rudeness, stubbornness, stupidity, bias, deviousness, dishonesty and all those other loveable characteristics of so many of my fellow human Wikipedians."

      Can't wait until we get a couple of bots arguing with each other about the vaiidity of an article :-)

      I'll just leave this here:

      Wikipedia articles must not contain original research.

  9. sisk

    So basically Google has made a bad Wikipedian. Don't we have enough of those in the world already?

  10. Cuddles

    To be fair

    "The software-scribbled passage a bit difficult to read without clear capital letters at the start of new sentences"

    The human-written entry doesn't appear to be any better in that respect.

    1. veti Silver badge

      Re: To be fair

      My thoughts exactly. There's quite a lot of pages in Wikipedia that are pretty badly written by humans - particularly, though by no means exclusively, when the humans in question were obviously writing in a foreign (to them) language.

  11. Martin Summers Silver badge

    I wonder what amanfrommars has to say about it.

    It would be nice if El Reg admitted that it is their pet AI they unleashed on us commentards a few years ago...

  12. dddd000

    AI vs. human

    The AI version is indeed longer, but it gives some information about Wings Over Kansas. The human version gives some information about some Wichita native who created it and precious little about the purported topic. I may be interested about Wings Over Kansas, but I have absolutely no interest on this fellow.

    So as long as the AI can be trusted to dispense with the ego issue, I'll take it.

  13. allthecoolshortnamesweretaken

    "The model works by taking the top ten web pages of a given subject [...]"

    Well, there's your problem.

    Ranked top ten how and by whom? GIGO applies to "AI" as well. Possibly even more so.

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like