[Sigia-l] 20questions--a little bit OT, but not entirely

[Boniface Lau]

> From: Samantha Bailey
>  
> I was thinking "trophy" and the app wasn't able to get it in 20
> questions (although it did get it in 28 questions)--in the grouping
> of items categorized with trophies, it included disco balls and that
> struck me as a very kind of "Women, Fire & Dangerous Things" kind of
> classification.

With 20q being a neural network, such grouping reflects how a human
brain comes up with creative ideas - by drawing together distantly
related information or associations.


> But it would appear that the categories they have developed are
> working pretty well, as the app is quite impressively at "guessing"
> the object.

The impressive performance was due to 20q's learning. Its neural
network learned from the millions of times that the game was played.

If you like, you can play as a member. When you win, you can update
20q's knowledgebase with answers of your latest play. You will then
know what new things (categorization), if any, 20q has learned from
that play.

BTW, playing as a member allows you to see what is in 20q's "mind"
(potential guesses). Thus, you get to watch the list of potential
guesses revised as a play progresses.


Boniface