Humans power technology. Simple truth. Machines have become such an integral part of our lives that we subconsciously accept them as beings, which is probably why we get so annoyed when they act like idiots. While we liberally abuse Windows, Google Image Search, voice recognition software and countless other software and services, we forget that old gem-GIGO (garbage in, garbage out). Write bad code, get bad software. Muck up a search query, get irrelevant search results.
“Help me help you,” says the salesman to his customer, says the doctor to his patient, says the PC to its user. Funny how that last bit never occurs to us. Your computer won’t do your taxes of its own volition, and the Web won’t give you the right answers unless you ask the right questions. Technology won’t serve us better till we help it do so.
The Old-fashioned Way
The best way to get help, of course, is to ask for it. One of the most challenging tasks for a computer is labelling images so they’ll turn up in relevant search queries. A program can’t know that DSC00023.jpg is a picture of a pigeon in a banyan tree, and will never be able to tell-unless it’s truly intelligent, and we all now know enough not to hold our breath for AI. Humans, on the other hand, can immediately come up with a whole bunch of labels for any image. But the show must go on, so somewhere in the world, someone is tagging these images so you can see it when you type in “pigeon tree” in an image search. It’s still an incredible task for a small group of people, but what if all of us got involved?
The concept is simple. Sign up at Google Image Labeler (http://images.google.com/imagelabeler/) or the ESP Game (http://www.espgame.org/) and play the image labelling game. You get paired with a random partner from a random corner of the world, and you’re both shown the same image. Both of you spew keywords to describe the image, and when your keywords match, you both get points. You get your few minutes of entertainment (not to mention the pride of earning points for it), Google’s Image Search algorithm gets away with being just a bit of code, and a satisfied user somewhere gets the search result he wanted. Everyone’s happy.
Then again, not many people are always willing to help…
Getting Sneaky, Part I
Image labelling isn’t the only way you can do your bit-in fact, there’s another thing you’re helping with already. When you register for any community site (forums, blogs and more), you’re asked to “prove that you’re human” by typing the word you see in a verification image. Instead of the computer using complex AI-like algorithms to detect spam-bot behaviour, it just asks you to use your human intelligence, stumping the bots. It’s called CAPTCHA, and now, Carnegie Mellon’s Luis von Ahn is taking this one step ahead with reCAPTCHA.
Right now, the Internet Archive (www.archive.org), among other groups, is digitising classic books to create an online repository of books that were written before the computers came. The physical books are being scanned and given to OCR (Optical Character Recognition) programs-all very well, but there are some snags. OCR software isn’t perfect, so a lot of words come out misread (luckily, most programs tell you that there may have been an error). The only way to get the right word is to have it read by a human, and with the volumes of text we’re talking about, a small group isn’t feasible. So von Ahn decided to take advantage of the fact that people solve 60 million CAPTCHAs a day-if each CAPTCHA was replaced with a word that the OCR software was having a problem with, then people would not only be proving their humanity to the site, they’d also be helping digitise the books without errors! Every unrecognised word is paired with another word that the program knows the answer for; if you get one right, it assumes that you got the other one right too. It doesn’t stop there, though-now that you know how it works, you’ll likely try to see if you can get away with typing gibberish for the second word (you know you want to). To counter this, the program accepts the answer for the unknown word only after a certain number of people have given the same answer.
The best part is that nobody has to go around soliciting help from Web users-they’re doing what they would do anyway, but their activities have an effect they don’t fathom.
Getting Sneaky, Part II
You only use 10 percent of your brain, they say-how does the thought of leasing the remaining 90 strike you? While that’s not exactly accurate, Microsoft Research’s Desney Tan and Pradeep Shenoy from the University of Washington do plan on using your mind to help computers detect and recognise faces better. Face detection algorithms are extremely complex, and depend on camera quality, the angle that the face is photographed at, the background and what-have-you. A picture with five faces in it may well get rejected as junk. We, on the other hand, can pick a face out of a crowd, at night, and with our eyes half-closed. Incidentally, the human brain isolates faces subconsciously-you could be sitting in a crowded bus, reading the newspaper, but your brain is processing all the faces that enter your peripheral vision.
And that’s where Subconscious Computing comes in. Tan and Shenoy’s system records subjects’ EEG (Electro-encephalograph, or brain-waves) patterns while they are viewing images-the responses are starkly different when they view images with and without human faces. With a single subject, the computer was able to identify images with faces with 72.5 per cent accuracy; for eight subjects, it’s 98 per cent. Imagine this system hooked up to a security guard watching a surveillance video-using his EEG waveform, the system can determine which frames have faces in them, so if the feed needs to be analysed, hours of useless footage can be safely discarded.
Sure, it’s spooky-we don’t want machines to have access to our subconscious minds (hell, we barely know what’s going on in there), but the concept is in the nascent stage, so we’ve got plenty of time to sort out the ethical issues: one of the ideas is to have images flash on screens in your peripheral vision while you go about your daily business-so you aren’t interrupted, and the system gets the output it wants. Of course, this can get pretty intrusive.
Not everything needs to interfere with Life As Usual…
The idea is based on the fact that a single human step can power two 60-watt light-bulbs for a whole second. What if you multiplied this to tens of thousands of people, taking tens of thousands of steps so many times a day? Graham and Jusczyk have the idea of a railway station with movable tiles on the floor-people walk, the tiles get depressed and power a dynamo, which generates electricity. “Crowd Farming,” as they call it, will use the movement of crowds-especially at rush hour-to generate power, which can then be used for lighting signboards, they say. Get them a large enough crowd (and we have so many of those in India, don’t we?), and maybe we can power whole sections of the railway station! The idea also extends to rock concerts-as fans jump in frenzy, they could be powering the artists’ amplifiers.
Their working prototype is a stool in a train station in Torino, Italy; people sit on it, causing four LEDs to light up. Silly, but it’s the idea that counts. The trouble is that when the crowds stop, so will the power; even assuming that the energy is being stored in batteries, there’s only so much they can store with today’s technology.
Speaking of batteries, it might be that we won’t need them anymore-not for our personal gadgets, anyway.
| How You Can Help |
| We’ve mentioned the ESP Game and Google Image Labeler earlier in this article; there are two other games in Luis von Ahn’s “Human Computing” initiative that you can play to help computers make the Web a better place: Phetch (www.peekaboom.org/phetch) Phetch is like an online treasure hunt, only with images. You find yourself in a group with other users, one of whom is a describer; the rest are seekers. The describer tells the seekers what kind of image he or she wants to see-“a monkey eating a banana,” for instance-and the seekers then use their favourite search engines to hunt down the image. The first seeker to get the right image becomes the next describer. How it helps: When you play Phetch, you’re helping the visually impaired use the web better. When each round is over, the describer’s demand is associated with the image that won-much like a caption. So when a visually impaired person is using a text-to-speech converter to read the site that’s hosting the image, the caption can be read out to them-“a monkey eating a banana”, instead of “monkey.jpg”. Peekaboom (www.peekaboom.org) Like the others, this game pairs you with a random partner online, and you take your turns as Peek and Boom. As Boom, you’re shown an image and a keyword that represents an object in the image (a family photo, for example, with “boy” as the keyword). You then reveal that part of the image to your partner, Peek, and he or she will start guessing what the keyword is. When Peek gets it right, the round is over, and you switch roles for the next round. How it helps: Peekaboom is supposed to be used to train computer vision algorithms. Right now, computers recognise objects using basic geometry, but this isn’t working too well. If you show a computer lots of images of an object, taken at different sizes and angles, it’s possible that they may recognise the object better. This is exactly what you’re doing when you isolate objects for your partner to guess. In addition, you’re also helping add more labels to the image, hence indirectly contributing to better image searches. Verbosity (www.peekaboom.org/verbosity) Think of Verbosity as the online version of word- and phrase-association games like Taboo. You and your partner alternately play the roles of guesser and narrator. The narrator is given a word (“water”, for example), which he or she has to explain to the guesser using clues (“quenches thirst,” for example). When the guesser gets it right, roles are reversed. How it helps: Common sense. We have it, computers don’t. By looking at the clues that you give your partner, a computer gets a database of “common sense facts” that we take for granted-like “water quenches thirst.” This should cut dow on the time it iakes learning machines to learn the most basic things in the world. |
I Vant Yer Blud!
When your body needs energy, it uses the glucose in your blood, which, in turn, comes from the food you eat. Back in 2003, Panasonic’s Nanotechnology Research Laboratory developed a prototype device that synthesised the glucose in human blood to generate power, the same way your body does. If it were to synthesise all the glucose, it could theoretically generate 100 watts of power, but that would mean you’d have to be eating constantly to compensate for the loss to your body. To keep the balance, the device won’t be allowed to consume too much sugar, thus putting a cap on how much power it’ll be able to generate, so don’t expect this to power your cell phones and PMPs. So you can’t go about killing yourself for louder music just yet.
At the other end of the world, Zhong Lin Wang of the Georgia Institute of technology is working on a “nanogenerator” that will use the flow of human blood to generate electricity. It’s based on the principle of piezoelectricity-the property of some materials that can convert mechanical energy to electrical, and vice versa. The material in this case is a nanowire gird made of Zinc Oxide; the prototype is stimulated by ultrasonic waves, but according to the professor, they can also be made to react to blood flow and muscle contractions.
The nanowire array might end the age of recharging batteries for personal gadgets
At this stage, the device only generates nanoamperes of current-enough only to power, well, nano-devices. Still, there is hope-it’s been proved that by increasing the size of the nanowire array, the generator can pump out more current-enough to even power personal electronics.
Think about it-a cell phone that will die only when you do.
All About You
Perhaps some day Subconscious Computing will advance to a point where data can be fed to your brain and processed without your knowing it. If you don’t want to wait that long, you can start contributing to making machines a wee bit smarter right now! Read the box below, go online and get started!