Topic: Guess the character Pages that link to <a href="https://ozoneasylum.com/backlink?for=29814" title="Pages that link to Topic: Guess the character" rel="nofollow" >Topic: Guess the character\

 
Author Thread
argo navis
Bipolar (III) Inmate

From:
Insane since: Jul 2007

IP logged posted posted 12-27-2007 11:54 Edit Quote

Pretty nifty - web apps reading my mind and guessing my "secret ID"

Can you beat it?
http://www.guessthecharacter.com

French version (truely scary, it guessed Shakira and Jesus in a row) :
http://www.akinator.com

Suho1004
Maniac (V) Mad Librarian

From: Seoul, Korea
Insane since: Apr 2002

IP logged posted posted 12-28-2007 02:33 Edit Quote

Hmm... I tried the "guess the character" one and it failed to get both Aragorn (from the LOTR films) and King Kong.


___________________________
Suho: www.liminality.org | Cell 270 | Sig Rotator | the Fellowship of Sup

argo navis
Bipolar (III) Inmate

From:
Insane since: Jul 2007

IP logged posted posted 12-28-2007 03:14 Edit Quote

Righto, it didn't guess my second attempt..

BUT I'd love someone who speaks french help me fool proof the second : I have not been able to beat it a single time.
Or I'd love to find one as accurate in english : basically, these are "expert systems".

As you can see, at the end, it tries to learn from it's mistakes.

The french one certainly has an impressive knowledge base already (it knows
thousands of celebrities, including Anakin, Darth Vader, Alf, Jesus, all possible models, actors, actresses, from various countries, fictional characters, etc..).

Systems like this learn over time, they learn an expertise in one specific field : stars guessing for those two.
Here is one that works well in english, provided you have seen a whale recently (it's in Java) http://www.aiinc.ca/demos/whale.html

My problem here is that most systems of this kind are devoted to financial decisions, or medical decisions - they can make a diagnosis or a market analysis.

But finding one that is "homebrewd", able to hit broader audiences, and properly done is hard - akinator, the french star guesser, is by far and large one of the best I've seen,
and I am not easilly impressed.
............
The aim of my game here is creating a concrete impression of what those can achieve : some of you may remember that, at some point,
I was working on one myself. It got me a wonderful grade, but would be SO useful here and "alive" instead of locked in my vault.

Suho1004
Maniac (V) Mad Librarian

From: Seoul, Korea
Insane since: Apr 2002

IP logged posted posted 12-28-2007 06:36 Edit Quote

Yeah, unfortunately my French is not good enough for the task. I can tell you that I wasn't too impressed with the English one, though. I also have to wonder about relying on user input. I saw a number of questions in there that were grammatically incorrect, etc., and it would be a simple matter to introduce a misleading question. I wonder if there is a means of weeding out incorrect information...

Skaarjj
Maniac (V) Mad Scientist

From: :morF
Insane since: May 2000

IP logged posted posted 12-28-2007 16:42 Edit Quote

Also, for a real glimpse of terrible English, go to the first site, and look for the "Which Transformer Are You?" one.

quote:
"Are you build by schokwave?"




Justice 4 Pat Richard

argo navis
Bipolar (III) Inmate

From:
Insane since: Jul 2007

IP logged posted posted 12-28-2007 17:19 Edit Quote

To adress both the questions of language and semantics :

1) A spellcheck can and should be run on any input. Probably is in the french version.
2) The expert system works like this, when made properly :
---------------------------------------------------------------------------
"The star to guess is Anakin sywalker if he fullfils these conditions -
- he is a man
- he is a fictional character
- he has brown hair
- he is a fighter
- he has been in space
- he has a role in Star Wars
- he has kids
- he helps people

That's just an example, but if a stupid fact slips into the above answers, called "facts base", it invalidates the whole - the algorithm used
are called "backwards and forward chaining of facts". "If he has blonde hair, he cannot be in the star wars franchise", for example, one fact
validates or invalidates the whole chain.

Now, I'd love to spend the day poking at the mistakes they made when implementing the english version : ESPECIALLY since it would
make for one french victory, for a change

But I'd rather focus on the benefits of such an application when it works properly.

Skaarjj, if you still live in Australia, have you seen a whale recently? Why not find out which species it belongs to ,)

Suho1004
Maniac (V) Mad Librarian

From: Seoul, Korea
Insane since: Apr 2002

IP logged posted posted 12-29-2007 04:57 Edit Quote

One fact to add to the Anakin Skywalker facts base: he is portrayed by a man who has the acting talent of a popsicle stick.

And yeah, spellchecking should be a given. It's the grammatical errors that might be harder to catch. The problem with systems that rely on open human input is that a lot of humans are retarded. But from what you're saying, the expert system would be implemented to avoid some of this retardation. Or at least that's the way I interpreted the above.

As for French victories, I think this one is pretty clear cut. I would be willing to concede French victories in other areas as well, such as cuisine.

argo navis
Bipolar (III) Inmate

From:
Insane since: Jul 2007

IP logged posted posted 12-29-2007 13:45 Edit Quote

Exactly, "retardation control inside" Here, the idea was more, and I have layed out a working "plan" for it a while ago,
to adapt such a system to the sink cleanup.

It would take two AI concepts implemented at once : fuzzy logics, and an expert system.
Eg. a bot to archive the stuff that is lying down the sink, and dormant (and sadly useless).

Basically, to create such a bot, we'd have to identify criterias on which it could work :
-----------------------------------------------------------------------------------------------
1) the words inside a thread, many occurences of "chritianity" for example, would mean philo.
2) the posters, each one of us being usually drawn more to this or that area of interest.
3) full blown expressions, like "intelligent design".

It could learn easilly from those three fields - "teach me a word specific to Philo and not to Photoshop",
"teach me an expression specific to general chit-chat and not to Philo",
"show me a poster who seldom posts in DHTML and often posts in Photoshop", etc.

Fuzzy logics is required because there is no way to do pre-emptive retardation control : it would have to tolerate spelling mistakes
to a degree to take in account all possible variations of a poorly spelled expression.

Then we could leave cuisine to the french, and claim for a victory in the field of classy expert systems that DO speak decent english.



Post Reply
 
Your User Name:
Your Password:
Login Options:
 
Your Text:
Loading...
Options:


« BackwardsOnwards »

Show Forum Drop Down Menu