Hunch Engine - Eric Bonabeau
============================
1.45pm, 7th March 2006

Two main themes: Obscenity and Infinity

We're good at the first, and bad at the second.

We're great at recognising obscenity, but very bad at defining what it is. "I can't define what hardcore pornography is, but I know it when I see it" - from a famous supreme court case.  (Jacobellis v. State of Ohio, 1964)

Our brains have been wired for situations where we have to respond very quickly. We remember very few things - 4-5 items in a shopping list. 

need better ways of expressing the rules of obscenity.
for that you need exposure (pun intended?)

We're very good at making snap decisions, but very bad at exploring alternatives.

There are two axes to any decision we make: Search and Evaluation. Search is finding things, Evaluation is making sense of them. Some things are very easy for humans to evaluate (and maybe hard for a computer), while other things have few options but it is very hard to decide between them.

most interesting quadrant is simple evaluation and complex search.

Our brains have evolved to embrace complexity, resulting in a number of biases. There are lots of biases in the way we search. When something is easily available, it biases our search for alternatives.

decision heuristics:
* evaluation traps
* search traps

"Are there more English words starting with k than words that have k as their third letter?" - people assume the first because it is easy to come up with words that start with k, but the second is actually true.

It's hard for us to get rid of ideas that are already in our head.

search traps:
* suffer from availability - we use with what we have at hand
* suffer from self confidence (shark cage doesn't work vs skinny sharks) - when you are SURE you have the right answer.
* belief perserverence
    hard to let go of an idea
* socially mediated
    Everyone else thinks they found a solution - groupthink - moves you to evaluate when you should still be gathering

* dominated by short term rewards
    bias of getting it now
    I don't care if I miss the perfect solution, as long as I get something that is OK. The nut-ball paradox: I know that dinner at my friend's house will be great, but I eat all the nuts even though I know I will have a great meal later.

Classic Dalmatian optical illusion - humans are GREAT at seeing patterns, sometimes even patterns that don't exist.

Terrible at exploring alternatives + great at spotting patterns = we need exposure!

Humans are great at detecting patterns, even to the extent of seeing patterns which don't exist.

To be able to be good at recognizing things, we need to be exposed ot lots of things
    - easier to decide between a few judgements, but starting from a blank page is much harder.
    - needs options (like a menu)
      "know it when I see it" - jump starting is hard

Eric is good at criticising or recognizing something good, but not good at designing, starting from a blank sheet of paper.

computers are complementary: 
    good at exploring alternatives
    terrible at detecting patterns

Outsource the right brained tasks to a human being
    - ex: mechanical turk


Animal (dog) breeding is a useful example:
    - selective agent: man
    - evolution: heritable variation
    - variation
        * mutation: generates diversity
        * recombination: choose parents  .. see what happens next
selective breeding gets you closer to desired result.

Use evolution to find something you like

There is no feedback loop in the mechanical turk; it's mono-directional. The dog thing is a navigation trick, "navigating through the space of dogs."

Example 1: Design, billions of possible designs

    Can I be creative under tons of constraints?
    (car designer .. 2500 things to remember  .. how think out of box?)
    Told by boss to be creative, but given a huge number of constraints.
    37signls: embrace the constraints
    use a tool that enforces the constraints so bad alternatives are not considered.
    The designer doesn't have to think about the constraints any more - the tool has done it for them.
    Example: Engineer from Honda who started with 9 random designs, then used the obscenity principle - "I know it when I see it" and converged to what he wanted after 4 generations.

coinage: "heuristicrat" :-) - Bruce Sterling seemed particularly impressed. not sure that was "impressed". more like "you are on my turf".

All human beings can use the obscenity principle.

Architect: the heuristicrat. He knows all the rules that apply to what you can design from your site.
    - says no to all the ideas (basically) because he's ruled by constraints
    - too constrained by experience within his/her field
    
in a sense, software architectures express the constraints and what is developed on top of this benefits from the know it when i see it.

Example 2
    Molecular Space
    "Drugability" impossible to describe
    how to get drug developer to look at more options w/out becoming overwhelmed?
    

    Evolve the molecule designs, keeping to constraints, and get the chemist to look for useful/interesting ones.

Demo of chemical modeling tool, mobius.
using genetic alroithms and human judgement to select parents for next generation. shows the data (fitness or "drugability") but allows chemist to use intuition to select.

There is a lot of power in this approach, as the professional examples have shown. But what about other people? What about my Grandmother?

Example 3  --     consumer oriented space
    Text Visualization
        Where is meaning?


demo of tag graph .. various views of how words relate / networks

demo of text network visualizer: context navigator


Example 4
    Empower Grandma
    
        Photo filtering & search
        name & pattern design

Doesn't know anything about Photoshop, doesn't want to know.

So... provide an interface that shows the photo along with previews of a bunch of random filters. Each one lets you save, view or compare. 
    use the photos selected to move to next generation of tweaks
        (successive approximation? :-)

pattern design example
    lots of starting points - picking one and iterating (lots of
        contro

Principles:

1. Search as design
2. Not just static visual search: sensory search/design, with moving pictures, food, sounds

assisted evolution?


Note: this whole talk reminds me of Zelazny's Amber Chronicles.  Where you'd walk ("through shadow") toward areas that felt closer to where you were trying to get, and eventually you got there.  

contributors:
--------------
Simon Willison - http://simonwillison.net/
Gabe Hollombe - http://avantbard.com/ - gabe@avantbard.com
Leon Chism - leon@chism.org
Peter Kaminski - http://peterkaminski.com/
Daniel Smmith - dls@daniel.org
Nick Zadrozny - http://zadrozny.com/nick