Archive for May, 2009


Following from yesterday’s post about Google’s inference engine, someone directed me to a similar project called Wolfram|Alpha which is going live next week.

This is an incredibly interesting video (if a little drawn out):


No Comments

Google’s Inference Engine: The Google IQ

Something that’s been on my mind for a bit, brought to the fore today because Google is launching a new product based on it:

Inference Engine

One of the wonderful concepts of artificial intelligence is the inference engine.  It’s a kind of computer algorithm that allows computers to reason things.  Even in a very simple form, it can achieve (what we perceive as) quite complex reasoning.

If it knows:

  • Barrack Obama is the US president
  • The US president lives in the white house

then it can infer that

=> Barrack Obama lives in the white house.

    So the system can know a third “fact” by knowing the first two. It’s kinda smart.

    If you add more facts, the system gets smarter, learning just like a child. And the more facts you add, the smarter it gets, just like a child. This is the basis for most modern forms of artificial intelligence. So what stops people making a super-intelligent inference engine? Someone has to teach it the facts. And, considering the number of facts required to make a system usefully intelligent (in a general sense) – it’s not worth the effort.

    Enter the Internet, and Google

    Google trawls the internet every day, which contains billions and billions of facts – waiting to be reaped. Up until recently, it just stored the content of the pages, so if you searched for words, it showed you the pages that contained those words (or had links containing your words pointing to them). But, as of late, it has started using some kind of inference engine (I noticed this lately when searching for a company director, it gave me the location of his company). And today, Google announced that they do in fact use an inference engine, and will be using it to help searching in their future products.

    So, here we have an incredibly large source of information, containing most of mankind’s knowledge, and an inference engine that is reading it, and gradually getting smarter – based on its knowledge. In time that inference engine will be very smart indeed. Which would be fine, if it weren’t for its source of knowledge: the internet is not fair, just, true, or moral.

    This system is currently in its infancy. Do we want it to grow up having learnt everything it knows from the internet?


    No Comments

    An Ode to Scrubs

    As the third best show on TV (the 2nd is Lost, and the number one slot goes to House) comes to a close, I feel it is time for me to express my deep sense of regret in poetic form.scrubs

    Oh Scrubs,
    We’ll miss you.

    After eight glorious years the funniest show on TV has come to an end.


    No Comments

    getmail FutureWarning with python2.3

    Some python (I assume) update on an old Centos box made getmail bork:

    /usr/lib/python2.3/ FutureWarning: %u/%o/%x/%X of negative int will
    return a signed string in Python 2.4 and up
    return ("<%s at 0x%x: %r>"

    It gets more than a little annoying when you get emailed those few lines several hundred times per day. However, according to the getmail FAQ, it’s not their fault – it’s the nasty evil lazy python developers’ fault.

    Changing the top of the /usr/bin/getmail file to ignore the warning made it go away:

    #!/usr/bin/python -Wignore::FutureWarning


    No Comments