The best computer for a dog is one that smells.
People are weird, but we should make the best computer we can for them.
Danny Hillis
he's a dude.
Danny Hillis likes to say how in the 80's he was heckled at a conference,
the guy joked about putting computers in doorknobs,
then when he returned a decade later, the hotel had RFID in every door knob.
people listen to him now
There's this nagging idea that there could be a universal data format.
Not one perfect format, but one format that's self-documenting.
One that's available to be consumed in any unplanned ways.
That's been the point, for a while,
the smell-computer for humans.
Doug Lenat
he's my favourite guy.
He tried this. Cyc was a quite-ridiculous (but also beautiful) attempt to explain knowledge to a computer, in a systematic way. Lenat got a research grant and stuck a ton of grad-students in a room for several years. It was a pretty nuts thing to do. The project went nowhere.
Worse, there were several suicides among AI researchers in the 90's.
it was bad.
the notorious example is their
14 different senses
for the word "in":
  • - ... 'in like a pencil in a mug'
  • - ... 'in like a city in a country'
  • - ... 'in like a subject in a paper'
  • - dot dot dot
  • If Cyc had succeeded, the world would be much different than it is now.
    Imagine adding your monthly bills, and your family tree, to the same database, and the computer just understands them both.
    A computer would be a thing you could throw things at in a carefree way.
    Computers would just be one seamless database.
    Think about this next time you're doing your taxes. We do not live in this world.
    Cyc failed, but the internet (arguably) did not.
    In some ways we're better at doing projects like this now, post-wikipedia.
    In some ways we're not.

    2004
    it was in this context that Danny Hillis wrote this paper called "Aristotle" .
    He's big into turning the internet into a Montessori school.
    I love him a lot.
    This was when google went public. Wikipedia was becoming a real thing.
    Hillis raised 15m (and then 42m) to make a less-pedantic, crowdfunded cyc.
    It would have an open API.
    It would have an office in San Francisco.
    Everything in one database.
    People started whispering. Hillis hired a big chunk of MIT.
    I first heard about freebase from this talk:
    Within a decade after giving this talk, Giannandrea would become one of the most powerful people on the planet.
    I had been heavily involved in wikipedia, and was watching it start to break: -
    - if Brad Pitt marries Angelina Jolie, do we have to write it on both pages?
    - if a person in Category:Blonde_cheerleader dies her hair, what then?
    - Do we need both Earthquakes in Canada and Earthquakes in Ontario ?
    or maybe it's
    Category: Formerly blonde cheerleader ?
    it's amazing how much my life changed watching that youtube video.
    My favourite part is in the questions, when the guy tries to think of an example that freebase couldn't handle, but he can't really think of one. I laugh every time at this. It's sort of an amazing moment

    Ok, briefly: (promise)
    Anything is a Topic (like Arnold Schwarzenegger)
    Topics belong to Types (like politician, body-builder)
    Types have Properties (like political-party, weight benched)
    that's all!
    Schwarzenegger is a person,
    and person has the property 'spouse'.
    Subject
    Verb
    Object
    Arnold
    spouse
    Maria Shriver
    Maria Shriver
    birthplace
    Chicago
    . . .
    Wait, but when, and where did they get married?
    To add more detail, it could break apart:
    Arnold
    marriage
    [their marriage]
    [their marriage]
    date
    April 26
    [their marriage]
    place
    Massachusetts
    This process of breaking apart a topic became a big part of Freebase. It was clever.
    You could make things as basic, or as detailed as you'd like.
    Sometimes people went too-detailed, and the data became too complex to query.
    Other times things were too-simple, and had to be refactored later.
    There was a lot of this sort of discussion, it was fun to do.
    Two people would make schema at different levels of detail, and that was fine.
    (on wikipedia this is not fine)
    We got to think about -
    or whether
    a horse and a race-car share useful properties
    or whether
    a book is an idea,
    or ideas come from books
    There was furious disagreement. It was a very fun thing to do. It also felt important.
    Tom Morris demonstrated his changes appearing in a google search, just hours later. Facebook, Bing, and everybody started using it. There were plenty of additional rumours.
    That winter, we bought my grandmother a new laptop.
    she opened her browser, and typed in 'swine flu'
    Bing displayed metadata from the infections disease ontology that I wrote.
    stupidly, i said 'i made that!'
    and my grandma was like, good job sweetie.

    This freebase structure was really quite good.
    Endless demos showed these these jazzy queries that jumped around subject matter in a fun way:
    useless shit
    - but they sure did illustrate what we were building.
    It was all one dataset! A post-modern non-stop wiki for a star-trek society.
    - like a James Burke BBC show, but real.
    something to teach children for free.
    a real, internet-like footing for knowledge.
    we actually do need this!
    we were actually building it!
    .... and it was working!
    It did not fail.
    - This is the important point I want to get across -
    Freebase got killed because it succeeded.
    I'm certain of this - a startup built today, that copied Freebase exactly,
    - today this would get bought and killed, in exactly the same way.
    It simply worked. It would work again. Nothing is different.
    2019:
    ◔_◔
    Douglas Lenat is much older now - (still a babe).
    I had lingering hope somebody famous would take the freebase RDF dump, and make it editable. This is now a very unlikely thing.
    Firefox recently discontinued their RDF support.
    The semantic-web conferences are winding down.
    Wikidata people are running-down their grant-money.
    Most of the freebase employees are still at google, which is sad because this seems to mute their twitter accounts, and their personal projects.
    Danny Hillis is kicking-around some ideas about starting a new freebase using distributed hash tables.
    (if it became a thing, I would quit my job)
    At this point, getting excited about a 'bitcoin wikipedia' feels a little like this jewish proverb:
    "if you are plowing your field and you hear of the second messiah, finish plowing your field"
    Barak Michener re-wrote the graphd database in Go, and open-sourced it.
    It even supports the freebase query-language. It's beautiful.
    He works at Lyft, and is one of the world's best people.
    There are some neural-net startups that have resuscitated Cyc data.
    Maybe Freebase will find a way into a new startup-cycle, in the future.
    When humans are sent to mars,
    they will have a 20 minute communication-delay.
    We're gonna need to build a local internet for them, somehow.
    I hope it's good.