Have you ever came across a solution to a problem that is so simple, that you should have thought of it long ago, and so elegant that it solves what had previously been viewed as an intractable issue? I have. I recently had the article Humans can; computers can't mailed to me which shows one solution to an old problem; namely, how to fix all of those optical scanning errors that crop up when computers are trying to read text, handwriting, etc. (See page 4 of Digitization and its discontents for a few examples) Traditionally even the best recognition software has had issues reliably converting scanned optical files into searchable digital files, and the idea of sitting some poor person down to enter documents by hand just seems cruel and effectively impossible for many organizations, especially given the volume of documents often being considered. Unless, of course, you can pull the old Tom Sawyer whitewashing the fence maneuver and convince other people to do it for you.

Enter the evil genius' at Carnegie Mellon. Drawing files that couldn't be read from the book scanning project at the Internet Archive, they have created several games that allow bored internet users to correctly enter the unscannable words--generally an easy task for a human, but a very difficult one for a computer. (See www.recaptcha.net) (or a similar concept and my personal favorite at www.gwap.com) The corrected answers are then used to assist in the completion of the finished digital text--generally without the game player ever knowing that they are helping.

I can only wonder how many other simple solutions are out there for some of the problems facing the digitalization of records.

Hats off to Carnegie Mellon,

Jeremy

Anyway, I canlt not do Friday fun, so here are a couple of serious games, (I guess?) that you might like to try out.

First up, is the Play the News game, in which you are presented with information relating to a story or situation that is in the news, and you choose one of several actions that people in the story might take. As well as comparing your predictions to other peoples, you also get scored according to how well your predictions turn out (that is, whether the course of action you selected is the one that actually happens).

To what extent is this approach likely to engage you in a news story, and potentially learn a little more about it, compared to something like the New York Times news quiz (which is also available as a social network (Facebook) application)?

The second game - or rather, set of games - that I'd like to mention appear on the beautifully named gwap.com site. That is, Games with a Purpose.

Several years ago, a game appeared on the web called the ESP game, that required two players who didn't know each other, and who just happened to be online at the same time, to try and find matching words that described a particular picture. The intention was to help index images so that they could be found by a search engines, and the approach represents a form of "human assisted computing".

Anyway, Luis von Ahn, the creator of the ESP game, has just released several more games on the GWAP website:

• Tag a Tune: both players hear a tune and have to describe it to each other; they then have to decide whether they are listening to the same tune. The purpose? Help a search engine learn more ways of finding songs (for example, whether they are happy or sad).
• Verbosity: players take it in turns to describe a secret word to each other - one person describes the word, the other has to guess it.
• Squigl: both players seem the same image, and are presented with a word; they each trace round the object described by the word as it appears in the picture.
• Matchin: two players are shown the same image; each player picks the image they like the most.

Now who was it working for who again?

The Machine is Us/ing Us, M. Wesch

Ah yes, we all work for the machine... ;-)

There are currently five games at the site. They tend to be coordination games. For example, the ESP game show you and another player the same image, and asks each of you to tag it with words. When you've agreed on a tag, you get points and move on to the next image.

For more on GWAP, you can see posts by Sarah Perez and by Nicholas Carr, and a 3-page paper (pdf) by GWAPmeister Luis von Ahn of Carnegie Mellon University. Then there's a wider-ranging piece about the relationship between games and work by games consultant Margaret Robertson, just posted on the BBC site.

Here are a couple of my favorite quotes about games and their purpose.

• We do not stop playing because we grow old. We grow old because we stop playing. Variations on this are credited to many different people. I'm inclined to give the credit to Anon, one of my favorite writers.
• When playing a game the goal is to win, but it is the goal that is important, not the winning. Reiner Knizia, my favorite game designer.

Looking at the second of these, it occurs to me that all five current GWAP are cooperative, rather than competitive. Hmm...

Los juegos disponibles en Gwap son muy sencillos, algunos casi infantiles y son en total 5

En ESP Game ves una imagen y tienes que decir que lo que ves o te sugiere.

En Matchin te muestran dos imágenes y tienes que decir cuál es más atractiva.

En Tag a Tune tienes que describir una canción.

En Verbosity, tienes que explicar o adivinar una palabra secreta que es dada a un jugador.

Por último, en Squigl tienes que trazar una línea alrededor del objeto que te es indicado.

La idea es transferir la inteligencia humana a los ordenadores, ya que todas las operaciones que se realizan en los juegos de Gwap no pueden ser desarrolladas por ordenadores. Estos juegos han sido desarrollados por Luis von Ahn, un pionero en la computación humana. Estos juegos permitirán mejorar los sistemas de búsqueda de imágenes, quizás música en un futuro, y otras cosas más. Desde luego un proyecto tan ingenioso como interesante.

+ info en RWW

Faz pouco tempo que conheci o Gwap. O site oferece uma série de jogos que têm como objetivo aprimorar a inteligência artificial dos computadores de alguma maneira.

Por exemplo... no jogo ESP são mostradas imagens para você e para seu "parceiro de jogo" (escolhido aleatoriamente). O objetivo é escrever palavras que descrevem o que aparece nas imagens. Caso os dois escrevam a mesma palavra a dupla ganha pontos. Este joguinho hoje é usado pelo Google para aprimorar as buscas no Google Images. São humanos ensinando computadores, por meio do entretenimento.

Vale a pena visitar o site... os jogos são bem viciantes.

O responsável pelo projeto, Luis von Ahn, já é conhecido por outras coisas interessantes e bem difundidas na web, como o CAPTCHA - e o reCAPTCHA - e o Google Image Labeler que mencionei.

Propaganda dos caras:

A human could quickly look at a photograph and immediately identify all the elements in it - a man, a woman, a park bench, a tree, a child on a bicycle, grass, a pathway and a cloudy sky. But who could be bothered teaching a machine these elements that even a child can easily identify? How long would it take to type in all those words and point out each of the elements in just one photo? Then even if you did teach the machine everything about one photograph, would the computer be able to identify the same elements in another photo taken from a different angle?

What about words? A computer may have data that labels milk and cola as liquid foods, but it wouldn't know that milk goes with cereal, but cola doesn't (usually). The human ability to recognize and recall experiences allows a person to make appropriate decisions. But how can these skills be imparted on a computer? The amount of data entry required would be astronomical. The cost of the required data entry operators would be prohibitive and the staff burnout would be disheartening.

That's where Games With A Purpose (GWAP) comes in. If there's no money for the staff required to teach the machine, you need volunteers. But volunteers come and go and usually represent a miniscule proportion of society. This was the challenge facing Carnegie Mellon University. How can we get ordinary Internet users from around the world to teach the machine? And how can we be sure they are teaching the machine accurately? Who will quality control that amount of teaching?

I will. You will. We will. And we'll do it because it's great fun, motivational and promotes thinking skills. Teaching the machine is a by-product as far as the players are concerned. I've been playing it for two days now and have amassed over 100,000 points.  What for? Who knows, but it's really enjoyable. Five different games to choose from and you can come back whenever you like to carry on your score.

Quality control? You play in real-time with another unidentified, uncontactable person and the machine is watching. In the game Verbosity (my favorite), one person (the narrator) receives an input word and using pre-supplied template sentences must accurately describe that word sufficiently well so that the second player can guess it correctly. Only English words can be used. Multiple players agreeing on terminologies and contexts ensure accuracy. The game is quick at only four minutes and each player takes turns being the narrator.

In Tag a Tune, each player listens to music and using words, describes the music they are listening to. Fast, slow, baroque, violins, techno, opera, rock. Based on the descriptions they see from their unseen partner, they each must decide whether they are listening to the same or a different piece of music. All the while, the machine is learning that this MP3 file has these attributes - provided to it and cross-checked by multiple human game players!

Of course, you won't think you're actually "working for Carnegie Mellon University" because you're enjoying it so much.  And one of the first things you'll discover is that most of the people you are chosen to play with must be complete idiots!  It could never be you who is the idiot. GWAP is tackling a complex problem in a very innovative and effective way. So instead of firing up that next game of Solitaire, why not try a Game With a Purpose?

The Idea

Artificial intelligence has come a long way, but humans are still far better at computers at simple, everyday tasks. We can quickly pick out the key points in a photo, we know what words mean and how they are related, we can identify various elements in a piece of music, etc. All of these things are still very difficult for computers. So why not funnel some of the gazillion hours we waste on solitaire into something useful? Luis has already launched a couple websites that let people play games while solving these problems. Perhaps you've noticed the link to Google Image Labeler on Google Image Search? That idea came from his ESP game (which is now on GWAP).

The Motivation

What researchers need to help them develop better algorithms for computers to do these tasks is data. The more data the better. Statistical machine translation has improved quite a bit over the past few years, in large part due to an increased amount of data. This is the reason why languages that are spoken by few people (even those spoken by as few as several million) still don't have machine translation tools: there is just not enough data. More data means more food for these algorithms which means better results. And if results don't improve, then we have learned something else.

The Solution

Multiple billions of hours are spent each year on computer games. If even a small fraction of that time were spent performing some task that computers aren't yet able to do, we could increase the size of the data sets available to researchers enormously. Luis puts this all a lot better than I can, and fortunately, you can watch him on YouTube (below).

So, check it out already.