Aloha, RFID

Aloha, RFID

Hawaiian farms launch a food traceability program based on RFID technology.

By David Blanchard

March 1, 2009

A statewide food traceability initiative based on radio frequency identification (RFID) technology has been launched throughout the Hawaiian islands. The Hawaii Produce Traceability initiative, a three-year pilot program, will provide product visibility down to the farm or field level. The program is being coordinated by Hawaii's Department of Agriculture and the Hawaii Farm Bureau. A number of Hawaiian growers, distributors and retailers are participating in the program, which is centered on fresh produce, including asparagus, eggplants, pineapples and tomatoes.

Phase 1 of the program is based on an RFID solution developed by Lowry Computer Products that uses RFID labels and inlays from UPM Raflatac, hardware from Motorola's Symbol Technologies group and system software from Globe Ranger. This solution is designed to provide real-time supply chain data of when boxed produce is planted and harvested, what pesticides are used, and when and where RFID-tagged boxes are scanned. The data will be automatically uploaded into a database, for use by program participants. More information is available atwww.hawaiifoodsafetycenter.org.

Some of the goals of the program include:

• optimizing harvest productivity;
• strengthening food processing controls;
• increasing cold chain visibility;
• reducing produce dwell time on shipping and receiving docks;
• accelerating transportation times between supply chain partners;
• improving inventory turns.

This solution should make it possible to enable product recalls within one hour. Those growers who might be affected by a recall would be able to localize the recall to a specific field, greatly minimizing their losses.

"This project provides the backbone for future and more preventive closed-loop sensor technologies, which are capable of measuring and reporting biocontaminants and temperature variations via the RFID system as produce moves through the supply chain," explains John Ryan, Quality Assurance Division, State of Hawaii Department of Agriculture. "The RFID system will provide managers with improved real-time control over potential food safety problems and help to prevent widespread human and economic impact."

Future phases of the program could include the deployment of RFID-enabled cellphones and the implementation of temperature tracking to reduce the threat of food spoilage. The full implementation could involve as many of 5,000 Hawaiian farms.

See Also

• Green is the New Black
• Green Manufacturing: By The Numbers

6 timesaving tips for Word 2007

Unleash the power of Word 2007 with these quick tips

Logan Kugler

February 27, 2009 (Computerworld) Microsoft's Word has always been an impressively powerful piece of software, with dozens, if not hundreds, of features most people never knew existed. The new "Ribbon" interface in Word 2007 attempts to make some of those features easier to find and use, but there are still a lot of really handy features that you may not even know to look for.

Here are six quick things you can do with Word 2007 that will save you oodles of time. Once you give them a try, you'll never want to do without them.

(Editor's note: Some of these tips work in earlier versions of Word. We've noted where that's the case, along with the variations in the steps you'll need.)

1. Highlight all appearances of a word or phrase in a document.

Traditionally, to find a word or phrase in a Word document, you would hit Ctrl-F (or select Edit --> Find from the pull-down menu), enter your search term and click Find Next to go to the next instance of the word in the document. From there, you'd continue to click Find Next to search for the next usage, and the next one after that, and rinse and repeat.

In Word 2007, you can run a single search that highlights every instance in the document, making it easy to scan through and find the exact instance of the word or phrase you're looking for (or to see if you've used the phrase far too often).

Choosing Highlight All in the Find dialog

Start a search as you normally would by hitting Ctrl-F and entering your search term. But instead of clicking Find, click on the Reading Highlight button at the bottom of the search window and select Highlight All. Every instance of your search term will be highlighted in whatever highlight color you have selected on your Home tab (yellow by default).

Voila! Your search term is highlighted throughout the document.

To remove the highlighting, simply hit Ctrl-F again to bring up the search window, click the Reading Highlight button and select Clear Highlighting. The highlighting will disappear.

2. Search for a word and all its variations.

Another powerful feature in Word 2007's search function is the ability to search for variations on a word, like plurals or verb tenses. (Earlier versions of Word can do this as well.) Click Ctrl-F to open the search form as usual, enter your search word and click the More button at the bottom. Check the box next to "Find all word forms."

Choosing "Find all word forms" in the Find dialog

When you run the search, Word will locate not only the word you entered, but all its variations. For example, say you want to find every time you describe the sport of running in your document. Enter "run" and Word will locate not just every instance of the word "run," but also "ran," "running" and "runs."

Note that Microsoft's terminology here is a little misleading. It doesn't actually find all forms of all words. For example, if you search on "music" and select "Find all word forms," it won't find "musical" or "musician." But if you enter a verb, Word will find all its tenses; if you enter the singular form of a noun, it will find the plural as well; and if you enter an adjective like "good," it will also find "better" and "best."

You can even use this trick to find and replace one term with another -- "run" with "walk," "ran" with "walked," "running" with "walking" and so on. (Rather than choosing Replace All, it's better to review Word's suggestions individually to be sure the correct form is used each time.)

Combined with the new Reading Highlight feature in Word 2007, this makes for a powerful editing tool. For instance, most writers try to avoid using the verb "to be" too often in their work, preferring more active and descriptive verbs (e.g., "runs" rather than "is moving quickly").

Enter "be" as the search term, select "Find all word forms" and click Reading Highlight --> Highlight All. Every use of "is," "was," "has been," "are" and so on will be highlighted, making it easy to pay special attention to them while editing.

The various forms of "to be" are highlighted.

3. Work in two parts of a document at the same time.

Word's Split View allows you to view and work in two parts of the same document in one screen. This makes cutting and pasting chunks of text from the start to the end of a long document a breeze; adding references or endnotes is also a piece of cake.

You can even keep your outline or table of contents visible in one pane while you work away in the other. Each pane in the Split View can be scrolled independently, so you can move around as much as you need to.

To activate it in Word 2007, simply go up to the View tab, and in the Window area select Split. (In earlier versions of Word, select Window --> Split.) Your cursor will jump to the middle of the page on top of a horizontal "split line," and moving your mouse will move the split line up and down. Find a place where you want to split the page and click your mouse to place the line.

Your document will now be viewed through two independent panes, and you can work in either one. If you don't like where you put the split line, just grab it with your mouse and you can move it around as much as you want.

Look at two separate parts of your document in the same screen with Word's Split View.

Once you've entered Split View, changing back to the default is as simple as repeating the process. Just go up to the View tab and instead of Split, this time you'll see Remove Split. (In earlier versions of Word, select Window --> Remove Split.) Click that and presto -- your screen is back to normal.

4. Make changes in the Print Preview screen.

You're all ready to print, but first you decide to have a look at the Print Preview screen (select Office button --> Print --> Print Preview or hit Ctrl-Alt-I) to see how the document will appear on paper. That's when you spot the glaring error you missed. Normally, you'd have to exit out of the Print Preview screen and return to the document, then find the error all over again.

But with Word 2007, you can easily change the behavior of the Print Preview screen so you can edit your document without exiting Print Preview. In the Preview area in the Print Preview ribbon, uncheck the box next to Magnifier. As soon as that's done, clicking on the document will place your cursor just as it would in the normal editing screen. You can fix a typo, add a word or even rewrite whole paragraphs if you choose.

Uncheck the Magnifier option to edit your document in the Print Preview screen.

Then click Print and send it directly to the printer. You can save your changes here in Print Preview or back in normal view.

5. Give yourself some extra room between paragraphs.

If you're someone who likes to make notes between paragraphs while you edit or revise a document, or if you want to add new text to a single-spaced outline, give yourself a little extra room to work in by automatically adding a line of space between your paragraphs.

Select the whole document by hitting Ctrl-A, then hit Ctrl-0 and a new empty line will appear after every paragraph. Select All again and hit Ctrl-0 to make the extra lines go away. (This tip works in earlier versions of Word as well, except hitting Ctrl-0 a second time doesn't remove the extra spaces.)

6. Add random placeholder text.

Sometimes you need to add random text to a document as part of a demonstration or to get a feel for a new layout. The designer's standby is lorem ipsum, a block of Latin derived from Cicero's De Finibus Bonorum et Malorum (On the Ends of Good and Evil).

The lorem ipsum text has no meaning for modern readers, which makes it perfect for not being distracted by what the text says, so you can instead focus on what it looks like.

Place your cursor where you want the random text to start and type:

=lorem(x,y)

where x is the number of paragraphs you want to insert and y is the number of sentences each paragraph should contain. When you hit Enter, Word will insert the desired chunk of text. For example, =lorem(1,4)generates this:

Lorem ipsum dolor sit amet, consectetuer adipiscing elit. Maecenas porttitor congue massa. Fusce posuere, magna sed pulvinar ultricies, purus lectus malesuada libero, sit amet commodo magna eros quis urna. Nunc viverra imperdiet enim.

Prefer your dummy text in English? Word 2007 will also insert English text, drawn from a Word 2007 tutorial, by replacing lorem with rand, like this:

=rand(1,4)

(Interestingly, in earlier versions of Word, the =rand function inserts repetitions of the classic typesetter's dummy sentence, "The quick brown fox jumps over the lazy dog." The tutorial text that Word 2007 uses is longer and more varied.)

Have your say

What are your best tips for Word 2007?

Both =lorem and =rand are handy ways of generating natural-looking text at blazing-fast speed.

I hope these tips save you some time next time you use Word 2007. Some of the features mentioned here were available (in some form or other) in earlier incarnations of the program, but if you didn't go digging through layers and layers of menus, you'd have never known it. New or old, though, all these functions can be useful to anyone who works with Word.

And there are probably further functions yet to be discovered -- I'm still looking for that Write Article button I just know has to be hidden in there somewhere.

Logan Kugler is a Los Angeles-based freelance writer who has written about technology and business for more than 30 national magazines, including MacLife, PC Magazine, Success and Advertising. Age.

New photo programs from Apple and Google include revolutionary face-spotting technology.

Friday, February 27, 2009

Face Recognition: Clever or Creepy?

New photo programs from Apple and Google include revolutionary face-spotting technology.

By Simson Garfinkel and Beth Rosenberg

Snap: Apple’s iPhoto automatically identifies faces within a collection of images. Credit: Simson Garfinkel and Beth Rosenberg

MULTIMEDIA Detecting Faces in Apple iPhoto '09

We have more than 25,000 digital photographs stored on our computer hard drives--most of them of people. Until now, our sole means of tracking down a familiar face was to search manually: by date, EXIF data, "tags," or the brute force of our own memory. Now computers can do the searching, thanks to the nifty face-recognition feature that Apple and Google have put into the latest versions of their photo-management systems.

Face recognition was one of those brilliant but technically iffy and ethically tricky counterterrorism technologies deployed as a result of the September 11 attacks. The idea was to automatically screen out terrorists as they walked through security checkpoints--only it didn't work out that way: at a test in Tampa, for example, airport employees were correctly identified just 53 percent of the time. Civil-liberties groups also raised concerns about false positives--people being mistakenly identified as terrorists, and possibly arrested, just because of their looks. And so, without a demonstratable benefit, face recognition largely dropped off the public's radar.

That's the public's radar, mind you. Many countries, including the United States, quietly revised their requirements for passport photos to make them friendlier to face-recognition software. The National Institute of Standards and Technology, which had been testing the technology since 1994, conducted large-scale face-recognition tests in 2002 and 2006. Oregon and some other states began using face recognition to detect when one person tries to obtain a license under different names. And all the time, the technology kept getting better. Much better.

In order to have a functioning face-recognition system, a computer must first be able to detect the face--that is, given a photograph, it must be able to find the faces in it. Technically, this is easier and more reliable than identifying a particular person. This technology was pretty much perfected just after September 11, 2001. The result: face-detection systems began appearing in digital cameras and camcorders a few years ago. These algorithms generally work by searching for objects that look like eyes, a nose, and perhaps something that's kind of round. They identify boxes where faces are likely to be, and then tell the autofocus system what part of the photo needs to be in focus. After all, everybody hates it when Grandma's eyes are blurry, right?

So face recognition starts with face detection. The face is then rotated so that the eyes are level and scaled to a uniform size. Next, one of three different technical approaches kicks in. Each of these approaches is, of course, covered by its own set of patents and bundled into various vendor offerings. One approach transforms the face into a mathematical template that can be stored and searched; a second uses the entire face as a template and performs image matching. And a third approach attempts to create a 3-D model based on the face, and then performs some kind of geometric matching. Based on our experience with the software, we believe that Apple's system is using a landmarks approach, while the Google system is doing some kind of image matching. But we could be wrong. Neither company has publicized which algorithms it is using.

We tested the face recognition in Apple's iPhoto '09 by applying it to two different databases of 17,000 and 10,000 photos, stored on our own hard drives. Google's Picasa only works with previously uploaded Web albums; we tested it on roughly 500 photos there. The verdict: both of these systems mostly work, are extremely cool, and are also kind of creepy.

Look-alike: Once a person has been identified, iPhoto will try to find other images containing his or her likeness. Credit: Simson Garfinkel and Beth Rosenberg

iPhoto '09 is certainly the friendlier of the two. The first time you run iPhoto, it searches out all the faces in your photo library; this took about four hours on a dual-core iMac. Next, you click on a photo of someone you know; click "Name" and fill in the text box beneath your subject's face. iPhoto will run through your photo library looking for other photos of the same person. (The recognition seems to be based on features inside a recognition box that is bounded by the left and right temples, eyebrows, and chin.)

Overall, iPhoto does a surprisingly good job finding a bunch of photos of the person you've selected and "named." But in the process, it finds photos of other people as well. So your next task is to tell iPhoto which photos it got right and which are wrong. iPhoto uses this information to update its mathematical models. It then looks back through your photo library for other photos of the same person. If it can't find any, you can manually point one out to give iPhoto another starting point; it will then seek out more. You can also click on a photo and ask iPhoto to try to figure out who is in the picture; if you confirm iPhoto's guess, the model gets better still.

We were astonished at how good iPhoto was at finding photos of our kids. Amazingly, iPhoto could even distinguish between our identical twins. (The trick is that one of them has a face that's a bit thinner and taller than the other's.) We were disappointed, however, that it found many more photos of one twin than the other, although we photograph both in equal numbers--and often in the same shot. A study of their photographs revealed something that we hadn't noticed, but iPhoto had: one twin always looks directly at the camera, but the other tends to tilt his head away, and iPhoto's face recognition doesn't work if the program only sees one eye. We also have lots of photos of kids in face paint. iPhoto found practically none of those, except for when the paint was confined to the middle of the child's forehead--which is outside its recognition box.

It's tempting to read a lot into iPhoto's recognition system. Searching for photos of Beth produced lots of photos of Simson's ex-girlfriends. It's tempting to say that iPhoto knows what Simson likes, but this could also be a bias in our test corpus: pick random photos out of Simson's library, and you're sure to find a bunch of his ex-girlfriends.

iPhoto was also surprisingly good at finding photos of our cats, especially the ones with white or orange fur. Unfortunately, it failed to find the tabbies--presumably, facial features are harder to distinguish when the eyes are the same color as the cheeks. And iPhoto does a startling job at finding and recognizing faces in shadows and other low-contrast situations. That's because iPhoto cranks up the contrast between face and background, presumably to make it easier to get out the features.

Since we installed iPhoto '09, our family has spent hours sitting around the computer, searching for photos of the kids, and teaching the computer what each of us looks like. We found a lot of old photos we had forgotten. We laugh at the mismatches. We try to understand the algorithms. This is one of the most entertaining programs that Apple has ever created.

Google's Picasa technology is far creepier. Instead of starting with a photo of someone you know and searching for all the similar matches, Google takes every photo that you've uploaded to Picasa, searches them all for faces, then "clusters"these faces into groups of, supposedly, the same people. You then go through each group and tell Google who a person is--including his or her full name, nickname, and e-mail address.

In fact, Google's clustering isn't all that great. It frequently puts different people in the same cluster, and it will make lots of different clusters for the same person. And unlike iPhoto, which could easily match photos of our 12-year-old daughter with her photos as a toddler, Google thought that the children were different people. But Google's user interface is pretty easy to employ, the matching task is strangely compelling, and before you know it, you'll have every one of your photos tagged with all the real names and e-mail addresses of each person that the photo features.

This real-name tagging is what makes Google's face recognition so creepy. Remember, all these photos aren't on your computer: they're on Google's server. And because e-mail addresses are unique, Google could use the tagged photos from all its Picasa users to create a global database matching photos to e-mail addresses. Doing so would not even violate Google's privacy policy, so long as Google only uses this information to make its service "better" and does not make the database generally available.

But what's really unsettling about Google's service is that it doesn't just stop at your friends. Before you know it, Google is asking you to identify all those other faces in your photographs--the people standing in the background, the faces in the crowds, even the faces on posters. This is certainly keeping with Google's corporate mission "to organize the world's information and make it universally accessible and useful." But is that what we really want from a photo-sharing website?

Our iPhoto experiences have been a delight: we've been excited and pleased to find so many pictures of our kids, family, and friends--and even ourselves. On the other hand, when we used the advanced tagging feature in Google's Picasa, we felt as though we were intelligence analysts working in the windowless lab of some totalitarian government.

We believe that consumer-driven face-recognition technology will fundamentally change public-policy debates about biometrics and mass video surveillance. After September 11, nobody really understood how this technology worked, what it got right, and what it got wrong. But before the end of this year, millions of Americans will have first-hand experience with some of the very best face-recognition systems ever deployed. Once the family-photo novelty wears off, we'll be watching to see if iPhoto and Picasa users ask their government to regulate this technology--or accelerate its deployment.

Friday, February 27, 2009

Flexible Screens Get Touchy-Feely

The first bendable, touch-screen display flexes will be used by the military.

By Duncan Graham-Rowe

Touch and feel: Bendable, touch-sensitive screens could lead to a new generation of more rugged and easy to use portable displays. Credit: Flexible Display Center

Researchers have developed the first computer display that is both flexible and touch sensitive. They say that the breakthrough could lead to more practical and easier-to-use portable devices.

Over the past few years, there has been a drive to develop displays that more closely mimic the properties of paper.

E Ink, based in Cambridge, MA, already supplies displays that are easy to read in direct sunlight and require little power for both the Amazon Kindle and the Sony Reader, compared to LCDs and plasma screens. E Ink's technology uses a layer of microcapsules filled with submicrometer black and white particles to create a low-power, reflective screen.

Ultimately, though, the goal is to make displays that are not only flexible, but that also respond to touch. The first flexible electronic-paper product, the Readius, is due to launch later this year. This electronic reader features a roll-out E Ink display made by Polymer Vision, based in the Netherlands.

Sri Peruvemba, VP of marketing at E Ink, says that adding touch sensing to this kind of display presents a whole new set of challenges. There are a number of ways to make screens touch sensitive, he says, but most are designed to work with a rigid screen.

Resistive touch screens--such as those used in the Nintendo DS games console--rely on points of contact being made between two separate conducting layers. Flexing these layers can lead to false inputs, says Jann Kaminski, display engineer at the Flexible Display Center(FDC), at Arizona State University, which codeveloped the new display with E Ink. "For resistive, you have an air gap that has to be maintained," he says.

Similarly, capacitive touch screens--for instance, the one used in the iPhone--rely on conductive transparent films made out of indium tin oxide (ITO), a material that doesn't like to be flexed. "It is a brittle, ceramic-like material," says Kaminski.

Touch screens that detect changes in light or vibrations across a screen as it is touched fare no better when flexed because the signals used become distorted, says Kaminski.

The only approach left, he says, is inductive touch-screen technology, although this is not without its challenges. Inductive screens typically use magnetized styluses to induce a field in a sensing layer at the back of the display. This kind of layer is not inherently sensitive to flexing, and some are already commercially available. But most flexible displays feature a very thin stainless-steel backplane that allows the display to flex while maintaining enough rigidity to protect it from damage. And these metallic backplanes act as a barrier to the electromagnetic fields that make inductive touch screens work.

To get around this, the FDC team uses an alternative backplane material: a thin-film plastic material made by DuPont called Teonex polyethylene napthalate (PEN). This material is already widely used in thin-film transistor manufacturing. It provides support for the display while allowing the inductive touch layer to work, says Kaminski.

Peruvemba adds that the approach doesn't degrade the quality of the image because the sensing is carried out behind the display. This is particularly important because E Ink products rely on the reflection of ambient light rather than on an energy-sapping backlight.

Prototypes have undergone rigorous testing (watch a video of the testing), and the first place that the displays are likely to be used is in the military, says Shawn O'Rourke, director of engineering at the FDC. The military interest comes from the need for portable displays that do not shatter, he says. Most displays are built on a glass backplane, so there is a real need to have something more robust. "They want thin, lightweight displays that are rugged and low powered," says O'Rourke.

Better computer-vision algorithms overlay digital information on the real world.

Tuesday, February 24, 2009

Microsoft Demos Augmented Vision

Better computer-vision algorithms overlay digital information on the real world.

By Kate Greene

Altered vision: This laptop is running augmented-reality software developed by Microsoft engineers. It can recognize a person’s location using the built-in camera. In this demonstration, virtual bubbles lead to a virtual pot of gold. Credit: Microsoft

Today, Microsoft researchers will demonstrate software that can, in real time, superimpose computer-generated information on top of a digitized view of the real world.

Adding additional visual data to a video display is a technique known as augmented reality. Michael Cohen, principal researcher at Microsoft Research, in Redmond, WA, says that the approach could add another dimension to future smart phones. "You could be out on the street, hold the device up, and it could recognize a restaurant and deliver ratings and the menu," he says. A smart phone featuring an augmented-reality display could also overlay a bus route and an estimate of when the next bus is due on top of a particular street. "It essentially becomes your portal to information," Cohen says.

Cohen and his colleagues will demo the augmented-reality technology at TechFest, an annual showcase of Microsoft's research projects, in Redmond. Their software, which runs on a small portable computer, analyzes scenes from a camera, matches to those stored in a database, and overlays supplementary information on the display. The researchers note that a smart phone with augmented reality could help allow engineers to "see" the pipes or electrical cables below a street. In the demonstration given at TechFest, the software will be used to lead people on a treasure hunt to a hidden prize of a (virtual) pot of gold.

Augmented reality has been an active area of research for more than a decade, although it has often required a head-mounted display and a backpack's worth of computing equipment. In recent years, cell phones and portable computers with cameras and other sensors have become powerful enough to handle the computational workload needed to run an augmented-reality system. Researchers at Nokia and Columbia University, for instance, are also developing augmented-reality systems, and a Japanese startup called Tonchidothopes to turn the concept into a product.

Most augmented-reality systems must be able to orient themselves accurately in order to function reliably. Some locate their position using GPS or by triangulating several Wi-Fi signals, and determine which way they are pointing using an accelerometer and a digital compass. Microsoft's augmented-reality device focuses on being able to recognize objects within a scene, using sophisticated computer-vision algorithms. Since the demonstration is carried out in the controlled environment of the conference hall, the researchers are not using location-detecting sensors; instead, they're relying solely on computer vision.

Recognizing elements of a scene regardless of the angle or lighting is a significant challenge. Cohen and his colleague Simon Winder, a senior research engineer at Microsoft, have developed algorithms that perform this task frame by frame, for a video feed in real time. The algorithm instantly matches frames to previously analyzed images stored in a database. In developing the algorithm, the researchers determined the best parameters or characteristics to help the system match each scene. Cohen explains that they used machine learning to quickly test different parameters and determine the ones that will provide the best matches.

For today's demo, Cohen's team took pictures of the conference hall in which TechFest is being held. The photos were analyzed using the computer-vision software, and the key features were stored in a database on a laptop computer that employs a built-in video camera to capture a scene.

"In about a tenth or a fifteenth of a second, the software is able to recognize a scene and look it up in a database," says Cohen. For the treasure-hunt game demoed during TechFest, the software displays a trail of bubbles that point to the direction in which the user should walk to find the prize.

Since it is just a research project, Cohen stresses that there is still plenty of room for improvement. For one thing, the parameters used to identify physical features of objects could be refined to make matching even more accurate, he says.

Another challenge to consider is how this kind of system would work in a less controlled environment, says Kari Pulli, a research fellow at Nokia. "The most common augmented-reality application is to use it as a museum guide," he says. "That's easy to do because the environment is fixed." The challenge is to make sure that such systems can work in an unfamiliar context, like a city street. But Pulli believes that this could become possible thanks to databases owned by Microsoft, Google, and Navteq that contain images of street views.

Cohen says he's optimistic that the computer-vision algorithms developed by his team could have myriad uses--from augmented-reality systems to gaming and robotics--but he doesn't foresee them being used in a specific Microsoft product anytime soon.

We need a new language for artificial intelligence

March/April 2009

Solving AI

We need a new language for artificial intelligence.

By Pedro Domingos

The goal of artificial intelligence (at least according to the field's founders) is to create computers whose intelligence equals or surpasses humans'. Achieving this goal is the famous "AI problem." To some, AI is the manifest destiny of computer science. To others, it's a failure: clearly, the AI problem is nowhere near being solved. Why? For the most part, the answer is simple: no one is really trying to solve it. This may come as a surprise to people outside the field. What have all those AI researchers been doing all these years? The reality is that they have largely given up on the grand ambitions of AI and are instead working on increasingly specialized subproblems: not just machine learning or natural-language understanding, say, but issues within those areas, like classifying objects or parsing sentences.

I think that this "divide and conquer" approach won't work. In AI, the best solution to a problem viewed in isolation often gets in the way of solving the larger problem. To make real progress, we need to work on "end to end" problems--self-contained tasks, like reading text and answering questions, that entail a number of subtasks (see "Intelligent Software Assistant"). Until now, it hasn't really been possible to do this, because the necessary computing power was not available. But within a decade or so, computers will surpass the computing power of the human brain. (While computers are extremely efficient at specific tasks, such as arithmetic, human brains are still ahead in terms of the number of operations they can perform per second. When this is applied to things that people are good at, like vision and language understanding, computers lose.)

Computing power is not the whole answer, though. Previous attempts to solve end-to-end AI problems have failed in one of two ways. Some oversimplified the problems to the point that the solutions did not transfer to the real world. Others ran into a wall of engineering complexity: too many things to put together, too many interactions between them, too many bugs.

To do better, we need a new mathe matical language for artificial intelligence. Examples from other fields of science and technology demonstrate just how powerful this can be: mechanics, for example, benefited from calculus; alternating current from complex numbers; and digital circuits from Boolean logic. Today these things seem like second nature to their practitioners, but at the time they were far from obvious. The key is finding the right language in which to formulate and solve problems.

What should be the language of AI? At the least, we need a language that combines logic and probability. Logic can handle the complexity of the real world--large numbers of interacting objects, say, or multiple types of objects--but not its uncertainty. Probabilistic graphical models have emerged as a general language for dealing with uncertainty, but they can't handle real-world complexity.

The last decade has seen real progress in this direction, but these are still early days. It's unlikely that we'll find the language of AI until we have more experience with end-to-end AI problems. But this is how we're ultimately going to solve AI: through the interplay between addressing real problems and inventing a language that makes them simpler.

Pedro Domingos is an associate professor of computer science and engineering at the University of Washington in Seattle.

The Evolution of Redundancy

IndustrialNetworking.net

Users Demand More Flexibility by Pushing for Open Standards. Don't Use Less Costly, Commercially Available Ethernet Components to Maximize System Availability

By Bob Wagner

As control systems grew in scope and became more distributed, the number of components in the system grew as well. Every component, to some extent, affects overall reliability of the system. To minimize the downtime due to system failures most major suppliers have developed products and systems to provide redundancy in an attempt to maximize system availability. These products range from redundant power supplies and cabling schemes to redundant processor and I/O systems with automatic failover, should an error occur.

External factors such as power source failures, electrical noise and potential physical damage to equipment also need to be considered. The installed system is typically a trade-off between cost and potential risk. Numerous potential points of failure will still exist, so system design must be carefully thought out to minimize their impact.

Network connection of controllers and I/O has resulted in significant installation savings in both material and labor. As these networks evolved, users demanded more flexibility by pushing for open standards. The proprietary networks gave way to more-open networks such as DeviceNet, ControlNet, Profibus, Modbus and Foundation fieldbus, to name a few. These networks provided the potential to integrate products from multiple vendors. Along with increased flexibility came the ability to access a wealth of configuration and diagnostic information directly from individual field devices. The status data help to pinpoint specific problems, potentially decreasing downtime. Automatic device replacement, available on some networks, enables the automatic download of specific configuration information to a replacement device, decreasing the time to replace a failed or failing device, as well as limiting the level of expertise required to perform the replacement. Diagnostic data also can be used to trigger preventive maintenance to help ward off potential failures.

Most of the networks noted require properly connected termination resistors at the end points to avoid excessive reflected signals that can interfere with normal signal transmission and result in intermittent or complete network failure. Connectors with individual resistors and wire connections are the most problematic. Molded type connectors are typically most reliable. Several vendors have an extensive selection of connector sets designed for reliably implementing the various networks in specific plant environments.

Profibus requires an active termination at the end nodes. Not only do the proper resistors need to be in place, but power must be maintained to those end devices, as well. If an end node loses power or is shut down for repair or replacement, resulting network errors can shut down the entire network. For critical applications, Siemens offers an active termination device that is separate from the normal network nodes to provide more flexibility in maintaining the system and limit potential failures. Of course, this device itself is a potential single point of failure.

ControlNet is the only network I’m aware of that provides for redundant cable connections integral to the network adapters. If redundant media is implemented, these adapters can detect a failure and automatically switch to the backup media path.

The implementation of Ethernet in a plant environment is a bit unique from the previously discussed networks, in that it comes in several different flavors, many of which can coexist on the same wire. A properly implemented Ethernet-hardware base potentially can support several different Ethernet protocols, each of which requires hardware that incorporates specific attributes or functionalities in order to operate optimally.

Ethernet also requires active intermediate devices to complete the network connections. These devices, such as switches and routers, are integral active components in the network and require additional planning to minimize the effect of single potential failures. Some newer Ethernet components include an integral switch with two or more ports. This allows for Ethernet installations to be set up in a daisy-chain fashion, which is often less complex and more cost-effective. Should one of these integral switches fail, you probably will lose that component and all components downstream.

To maximize system availability, don’t use less costly, commercially available Ethernet components. Any cost savings will disappear quickly the first time a less-hardened component fails.

Robert D. Wagner, P.E., is a controls consultant and system integrator in the Cleveland, Ohio area, with more than 30 years of industrial experience.

March/April 2009

Buzz Meter

Data mining sheds light on what makes news.

By Matt Mahoney

Multimedia

A chart (also below, here as a pdf) of online buzz words.

Political scientists have long studied the news cycle, tracking which people and topics drive coverage and for how long. But the sheer volume of news outlets made it hard to quantify their results.

Researchers at Cornell University are trying to get a quantitative handle on how news stories proliferate. Computer scientist Jon ­Kleinberg reasoned that instead of trying to sort items from blogs and news sites into arbitrary categories, he could home in on quotes to identify their topics computationally. But references to a quote might extract different phrases from it, change its tense, or paraphrase it, resulting in dozens of different versions. So Kleinberg and his colleagues developed algorithms that determine family resemblances between strings of words in different articles.

The researchers are now canvassing about a million online news items a day. Focusing on quotes might exclude some relevant items, but it helps identify the types of stories that prove most popu­lar and the websites that report on them first. The researchers have found that with the exception of a handful of professional political blogs that are the fastest to sniff out a story, mainstream media sites drive coverage, converging on a story two and a half hours before blogs react. But mainstream sites are also quick to abandon stories, while blog interest can persist for days.

Pop chart: This graph depicts the 50 phrases that generated the most buzz online in the last three months of the 2008 presidential campaign. The vertical axis indicates the number of Web items featuring some version of the phrase posted hourly; the horizontal axis shows fluctuation over time. Each phrase has an associated color, some labeled with phrase excerpts.

Credit: Jure Leskovec, Lars Backstrom, and Jon Kleinberg

March/April 2009

From the Labs: Information Technology

New publications, experiments and breakthroughs in information technology--and what they mean.

By Kate Greene

Brrr: A thermoelectric cooler, at the center, on a copper plate.
Credit: Macmillan Publishers, Ltd: Nature Nanotechnology vol. 4, issue 1 © 2009

Chip Chiller
On-chip cooling could increase performance and decrease power consumption

Source: "On-chip cooling by superlattice-based thin-film thermoelectrics"
Ravi Prasher et al.
Nature Nanotechnology online, January 25, 2009

Results: Researchers at Intel, Arizona State University, and Nextreme Thermal Solutions and RTI International, both located in North Carolina, have integrated a thermo­electric cooler into a computer chip for the first time. The semiconductor-based device, which uses electric current to move heat from one place to another, cooled a targeted region in a chip by 15 °C.

Why it matters: When microprocessors and opto­electronics operate, they generate heat; too much can inhibit performance and reduce reliability. Today's cooling systems use flat metal plates attached to a chip to disperse the heat, and metal heat sinks, fans, and liquid-based cooling systems to remove it. But these tech­nologies are bulky and inefficient. If small thermoelectric coolers could be built onto the heat-dissipating metal plates to target hot spots in the chip, they could replace other cooling systems and save space. Such focused cooling might also consume less energy.

Methods: The researchers selected thermoelectric coolers made from nanostructured thin films whose cooling properties had been proved superior to those of bulk thermoelectric materials. To attach a cooler to a copper plate already incorporated into the chip packaging, they applied an insulating material to the copper and deposited metal lines to serve as electrical connections to the cooler. Then they filled the spaces between the lines with a ­polymer for mechanical ­stability and soldered the cooler to the lines.

Next steps: Thermal resistance in the contact point between the cooler and the copper plate keeps the integrated device from cooling as effectively as a stand-alone device would. To reduce this resistance, the researchers are exploring alternative connectors, such as special types of solder and carbon nanotubes. They also plan to use more thermoelectric coolers to cover all the hot spots on a chip.

The water needed to produce everyday goods and beverages

Water use

Thirsty work

Feb 25th 2009
From Economist.com

The water needed to produce everyday goods and beverages

WATER is a precious commodity, as any farmer in drought-besieged parts of China, America or Kenya knows only too well. Consumers may already be aware of the environmental impact of producing goods in terms of energy or pollution, but they might be surprised to learn how much water is needed to create some daily goods. A cup of coffee, for example, needs a great deal more water than that poured into the pot. According to a new book on the subject, 1,120 litres of water go into producing a single litre of the beverage, once growing the beans, packaging and so on are measured. Only 120 litres go into making the same amount of tea. As many as four litres of water are used to make a litre of the bottled stuff. Household items are even thirstier. Thousands of litres are needed to make shoes, hamburgers and microchips.

Shutterstock

Wednesday, February 25, 2009

Website Maps Global Web Blocks

Herdict uses volunteers to monitor Internet censorship, filtering, and outages.

By David Talbot

Internet censorship occurs worldwide for a broad range of reasons, but the exact dimensions of the problem can be hard to document. A website launching today aims to tap the power of "crowd sourcing"--fielding and aggregating reports from volunteers--to provide real-time data on the state of Net filtering.

"We hope to have an ongoing and increasingly refined map of blockages around the world, as they happen," says Jonathan Zittrain, a Harvard law professor and cofounder of the Berkman Center for Internet and Society. "It might not just be blocks occasioned by a government trying to get rid of content," he adds. "It might be a company filtering stuff from its employees, and it might also be takedowns of content"--for example, videos that are alleged to infringe on copyright.

The site extends the work of the OpenNet Initiative, a joint project coordinated by institutes at Harvard, the University of Toronto, the University of Oxford, and the University of Cambridge to document Web filtering around the world. To date, the efforts of this project have been labor intensive: researchers typically dispatched volunteers within certain countries to check the accessibility of websites containing sensitive information. This method contributed to a study called Access Denied that produced a map documenting evidence of filtering within about 40 nations three years ago.

Such efforts will continue, but they will now be augmented by citizen reports, contributed via a site called Herdict (the name is a play on "the verdict of the herd," Zittrain says). The site aggregates reports and displays trends over time, showing which sites are blocked in which countries. Herdict is only designed to show whether or not a site is accessible; it cannot say whether the problem is due to government censorship, an Internet service provider's filtering policy, or nothing more than a severed cable. However, the researchers argue that such answers can be inferred by monitoring broader trends. "We are operating completely in the dark right now," says Zittrain. "But it's something that is hidden in plain sight, because any individual knows his or her own experience."

In theory, data collected by the site could also be used to compare the service quality of different ISPs, he adds.

The effort does present a potential catch-22, however: what happens if the site itself is blocked? Zittrain notes that reports could still reach Herdict via other means, such as Twitter. He adds, "When we get blocked, it will be our first milestone of success: we've done enough that we are worth blocking!"

But Who's Counting?

No one really knows how many people visit websites. A San Francisco startup and Google are both working to change that.

By Jason Pontin

In August 2006, when Roger McNamee invested in Forbes, he did so in part because its Web audience was thought to be huge. McNamee is a founder of Elevation Partners, a Silicon Valley private-equity firm that counts Bono of the rock band U2 as one of its managing partners; it specializes in big, bold investments in media and technology. Onstage at EmTech, Technology Review's annual conference, he said, "Look: I'm not investing in Forbes for its dead-trees business."

He counts: Konrad Feldman, a cofounder of San Francisco–based startup Quantcast, sees big business in audience measurement. Credit: Toby Burditt

RESOURCES: View graphs of U.S. advertising spending and U.S. online ad spending.

At the time, Jim Spanfeller, the chief executive of Forbes.com, claimed that more than 15 million readers around the globe had visited his site in February, making Forbes the world's leading business site. He supported his boast with research from ­ComScore Media Metrix, one of the two leading suppliers of third-party traffic data for the Web. The numbers seemed safe enough: Forbes­.com's internal server logs showed even greater Web traffic. It was embarrassing, therefore, when ComScore announced that it had changed the methods it used to estimate worldwide audiences, and that little more than seven million people had visited Forbes.com in July. That placed Forbes's online audience below those of Dow Jones (whose sites include WSJ.com) and CNN Money (whose sites include Fortune). Bitchy press accounts suggested that McNamee had been overcharged--if not actually robbed--for his investment, which was variously reported at between $250 million and $300 million.

More than two years later, McNamee claims he always knew there were broad discrepancies between what the internal sever logs of Forbes.com showed and what third parties reported. "To be a headache, it would have to be surprising," he says. Instead, he suggests, he invested with no very precise idea of Forbes.com's audience: "I looked at every indicator that was out there. They were all bad. In the end, I had to think about it differently. I invested in Forbes because I thought the market was underserved, and because they had made fewer mistakes than anyone else." (To this day, McNamee declines to say how much he paid for how large an equity stake.)

People still can't agree on how many readers visit Forbes.com. "According to ComScore, we have six to seven million visitors [per month]; our own logs say 18 to 20 million," says Spanfeller. But while the difference between third-party and internal measurements is, for a variety of reasons, particularly striking in the case of Forbes, confusion about the size of online audiences is universal.

No one really knows how many people visit websites. No established third-party supplier of audience measurement data is trusted. Internal Web logs exaggerate audiences. This matters to more people than investors, like McNamee, who worry that they have no way to evaluate new-media businesses. The issues involved are technical, and occluded by ugly jargon, but they concern anyone anxious about the future of media as print and broadcast television and radio shrink in importance.

Happily, a California startup and Google are working to measure Web audiences in new and better ways.

The Price of Journalism
Why care about something as arcane as dodgy audience measurement? Here's why: where content is free, as it is on most websites, the only thing that will pay for quality journalism--or, really, anything valuable at all--is advertising. For most new-media businesses, "display" or banner advertising is the main source of operating revenues. But the general inability to agree on audience numbers is stunting the growth of display advertising.

Every year, advertisers spend billions of dollars online; ­eMarketer, a research firm, predicts $25.7 billion in 2009 in the United States alone. Marketers study Web audiences to help them decide which sites to spend money on: they try to divine the number of people who visit a site every month, demographic details about those visitors, the length of time they stay on the site, the number of pages they view, and the relationship, if any, between the ads they see and the way they behave. The people who actually buy ads--media buyers and planners at advertising agencies--use this information to choose appropriate sites for campaigns. Finally, publishers use the data to set advertising rates.

However, the correlation between the size of Web audiences and their value to advertisers is not direct. In print, the relationship between audience size and advertising spending is simple, because the prices of ads derive largely from a publisher's audited statement of circulation; media planners buy the total audience. Online, it's more complicated because the currency of display advertising is ad impressions, or the number of times a specific ad is served to a particular part of a website. "Audience numbers don't affect my buying decisions very much," explains David L. Smith, the chief executive and founder of Mediasmith, an ­interactive-media planning and buying agency whose clients include the National Geographic Channel and Sega. "If we were buying the total audience of a site, it would be different. But most of the time we buy packages of impressions."

Jim Spanfeller, who is a past chair and current board member of the Interactive Advertising Bureau (IAB), the industry association that represents sellers of online advertising, agrees with Smith that unreliable audience measurement doesn't directly affect ad spending, at least at larger sites: "If you're an established site like Forbes.com, you're selling on an ad-impression basis. The problem arises when an agency is thinking about moving money from one medium, like print or television, onto the Web." Then, Spanfeller says, media planners can't show their clients whether Web audiences replicate or complement the audiences that advertisers are reaching through traditional media. "We need believable numbers so that we can do cross-media comparisons," he says. Additionally, bad audience measurement "hurts smaller sites with more targeted audiences that don't have a lot of impressions"--the class of sites that Spanfeller, like many digerati, says occupies "the long tail."

Thus, the real consequence of the audience measurement problem is a chilling effect on the transfer of advertising from older media to new. Meanwhile, another form of online advertising is growing quickly--but it's not the ads publishers sell. The numbers clarify. Spending on "keyword" or search advertising (the sponsored links that appear near search results on Google.com and other search sites) grew 21 percent in 2008, mostly at the expense of print, local television and radio, and Yellow Pages advertising; it now constitutes 45 percent of all online advertising. That's because the effectiveness of keywords is unambiguous: advertisers pay directly for click-throughs or purchases. There's no need to appeal to anything so disputed as the size or composition of Web audiences. This growth in keyword advertising has mainly bene­fited the search firms. By comparison, the display advertising that media companies sell grew only 4 percent the same year.

Four percent growth might sound all right to some, but it occurs at the same time that advertising revenues in print are falling rapidly. For instance, ad spending in newspapers will decline from $50.8 billion in 2007 to $45 billion by 2012, according to Borrell Associates, a research firm. Even Forbes is sweating. As a private company, it does not disclose its revenues, but the number of ad pages in its magazine has been shrinking since 2000. At the same time, the company's online advertising revenues are reported to be between $55 million and $70 million, a figure Spanfeller did not dispute. That's not so much for a publication with an audience of 20 (or even seven!) million. In the glory days of print advertising, publications with much smaller audiences earned as much or more: Red Herring, which I once edited, earned more than $50 million in print advertising revenue in 2000, and its circulation was only 350,000 readers, according to Ted Gramkow, the magazine's former publisher.

Display advertising was meant to fund the great shift of readers to new media. It's not happening. For more than 100 years, advertising paid publishers and underwrote their production of great journalism; now, those ad monies are being funneled to search firms that create nothing but code. As Roger McNamee says: "Getting this right is absolutely necessary for publishers to be able to continue to do interesting things."

Panel Discussion
What's wrong with existing methods of measuring Web audiences? Lots.

ComScore and Nielsen Online, a division of the Nielsen Company, are the established leaders in the field of audience measurement and the sale, to advertisers, agencies, and publishers, of the data that audience measurement produces.

These third-party audience measurement firms exist because the internal logs of publishers are notoriously unreliable in quantifying user activity on a given site. "When publishers use their log files, there are many limitations," David Smith says. He says that the limitations of using these internal logs (a practice sometimes called "census measurement") include, in ascending order of impact, overcounting individuals with multiple computers or Web browsers; counting "mechanical visits" by Web "bots" and "spiders" (for example, when Google crawls the Web to estimate the popularity of sites) as visits by real people; and overcounting individuals who periodically flush out the "cookies" of code that sites stash on browsers so that returning visitors can be recognized.

To create more-accurate audience numbers, ComScore and Nielsen Online rely on a methodology inherited from television audience research: the panel. Nielsen, for instance, has recruited nearly 30,000 panelists for its flagship product, called Netview. Panelists agree to have their Web browsing monitored through interviews and through "meters," or spyware, installed on their personal computers.

But what worked with television doesn't work nearly so well with the Web. "Panels are always problematical," says Spanfeller, "but on the Web they're super-problematical. Panels undercount by one-third to one-half." In short, publishers simply can't accept that their audiences are as small as panel-based measurements suggest they are.

Among the problems with panel-based audience research, according to both Spanfeller and Smith, is that it tends to undercount people who look at sites at work, because most companies' information technology managers won't install strange spyware on their computers. Sometimes, panelists lie to interviewers. Also, both say, there is a straightforward "sampling error" (what statisticians consider the misprisions that derive from sampling too small a portion of a general population): with as few as 30,000 panelists, the audiences of smaller sites are often grossly underestimated or missed entirely.

A final problem with panel-based measurements is that at the moment, neither Nielsen nor ComScore has itself been audited by an independent party. Who knows, both Spanfeller and Smith asked darkly, how valid the firms' reporting methods really are?

Nielsen defends its panels. "I guarantee you, if our numbers were higher than the publishers' server data, we wouldn't be having this argument," says Manish Bhatia, the president of global services at Nielsen Online. Bhatia notes that Nielsen does sell products, such as SiteCensus, that install software tags on publishers' websites and measure server logs. "In combination with panels, they're useful," he says. "But panels are more reliable, they provide demographic information, and they tell you what people do after they've seen an online ad."

For its part, ComScore also concedes that server logs have their place: they disclose which Web pages a publisher served, and when. But like Nielsen, the company insists that only panels provide an accurate measurement of audiences and their demographic makeup. "Servers don't measure people," says Andrew Lipsman, director of industry analysis at ComScore.

Why are Nielsen and ComScore so wedded to panels? According to David Smith, "The incumbents have a huge amount of money invested in their methodologies--and getting them to admit they have a problem isn't easy."

Roger McNamee is more blunt. "I understand why Nielsen is so bad," he says. "But why isn't there anything better? There's a huge market opportunity for any venture capitalist who is willing to fund a system that audits actual traffic."

"What we need is a third-party Omniture," says Spanfeller, referring to the website analytics software that many publishers (including Technology Review) use to log their own traffic.

Measure for Measure
Recently, I visited Quantcast, a San Francisco-based startup that is hoping to provide just such a service. Founded in 2005 and funded with $26 million, mainly from Polaris Ventures and Founders Fund, the company wants its service, which launched in 2006, to overthrow traditional panel-based Web audience measurement.

Konrad Feldman, the company's youthful, redheaded, British-­born chief executive and cofounder, met me at the company's headquarters overlooking the Yerba Buena Gardens and the Moscone Center. In a large conference room with a cement floor, decorated according to the precepts of venture-capitalist high minimalism, he asked whether Technology Review was "quantified"--that is, whether its online visitors were tracked by the startup's software tags. After we confirmed that our site had been quantified for some time, he opened his laptop and searched for our URL at Quantcast.com.

An elegant dashboard of audience information was swiftly served: TechnologyReview.com, it said, had 342,000 "global ­people" and 205,000 "U.S. people." These numbers, which measured monthly visitors to our site, were not so low as those reported by traditional third-party audience measurement firms, but they seemed suspicious: throughout 2008, Omniture reported around 650,000 unique visitors to TechnologyReview.com every month. But we also learned that 32 percent of TechnologyReview.com's readers earned more than $100,000 a year and that 24 percent had postgraduate degrees, which seemed about right. (A peek at Forbes.com, which is not quantified but whose numbers the startup had extrapolated, showed that the business site had 4.9 million "U.S. people," who were richer than TechnologyReview.com's readers, although not as highly educated. Because Forbes was not quantified, Quantcast didn't supply Forbes.com's total worldwide audience.)

Quantcast's service, like that of existing audience measurement firms, begins with panels--or, more precisely, panel-like data in the form of "reference samples," provided to the company by third parties such as market research firms, Internet service providers, and toolbar companies, among other sources. These statistical methods create a basic model of U.S. Web traffic. But when publishers install Quantcast's tags on their servers, Quantcast gets more details; the startup adjusts for spiders and bots, people with multiple computers, and cookie flushers. The two methodologies are combined using something Quantcast calls its "mass inference algorithm," created with the aid of two Stanford University mathematicians and refined by the seven mathematically minded PhDs who work at the company. This algorithmic analysis of panel research and server-based measurement is unique in Web audience measurement (although Nielsen more coarsely combines the two methodologies with a service called VideoCensus, which tracks online video viewing). The resulting audience information, says Feldman, is much more reliable than anything offered by ComScore or Nielsen.

"Publishers and advertisers have used panel-based research for nearly 75 years," says Feldman. "So there's obviously an established way of doing things. But equally, there's a pretty clear recognition in the marketplace that something has got to change."

Because Quantcast's audience information is free (where ­ComScore's and Nielsen's measurements are not), the company hopes to make money by charging publishers who enroll in Media Planner, a service launched last May that helps media planners spend their clients' cash. Although Media Planner is wholly free for now, Quantcast wants to expand the service so that it can finely describe demographic subsets within websites' audiences, a ­utility for which the company believes the sites themselves will pay. ­Feldman explains this tricky idea: "You have a sales force at ­TechnologyReview.com, and they can't possibly speak to everyone who might value your audience. But if you can expose that audience to buyers, then you can create a way whereby buyers can discover the parts of your audience they find particularly valuable." Feldman says that Media Planner allows media buyers to find appropriate audiences, "but it's the publishers that should pay, as they're the ones getting higher rates for their audience segments." More ambitiously, Feldman hopes Quantcast's audience data, in combination with ad impressions, will create a new currency for advertisers, advertising agencies, and publishers that will make display ads more effective and therefore more valuable.

Feldman and his cofounder, Paul Sutter, the company's president, do not approach the problem of audience measurement as veterans of media. Feldman, a computer scientist, was the cofounder of Searchspace (now Fortent), which developed software to help financial-services firms detect money laundering and the financing of terrorists. Sutter founded the network optimization company Orbital Data (later acquired by Citrix) as an expert on high-performance computer architectures, a background that has proved useful as Quantcast processes the thousands of terabytes of data it has collected.

When the founders first conceived the company, Sutter says, "we just asked the most simple, kindergarten questions, and it soon became clear that the language that media buyers and planners were speaking was nothing like the language of Internet advertising, with its cost-per-clicks and so forth. Media planners liked to talk about audiences, demographics, and lifestyles. So the answer was quantcasting, which means just reaching the people you want to reach." Today, the company claims that 85,000 broadly defined "publishers" have elected to be directly measured by Quantcast, including the Disney-ABC Television Group, NBC, CBS, MTV Networks, Fox, BusinessWeek, and Time's SI.com and CNNMoney.com.

Quantcast is not the only company with the bright idea of replacing panel-based audience measurement. Last June, Google announced a new service, Google Ad Planner, which uses the company's detailed knowledge of Web traffic to provide interested parties with a more accurate understanding of Web audiences. Wayne Lin, Ad Planner's product manager, demonstrated the service to me when I visited the GooglePlex in Mountain View, CA. Because Google owns DoubleClick, one of the two dominant systems for serving ads, Web audience data can be combined with the ad-serving system so that media planners know which sites are best suited for which ads. The combination should be powerfully attractive for media planners and marketers, says Lin.

How do media planners regard the two new audience measurement services? "We use Quantcast now at Mediasmith, but they are not complete enough yet to be a total solution," says David Smith, who briefly advised the startup during its formation. The difficulty, according to Smith, is that the site's audience information won't be really useful--let alone a new currency--until more publishers elect to be quantified. Jim Spanfeller agrees. "They're to be commended for working hard on the problem," he says. "But it's very much a chicken-and-egg thing."

As for Google's Ad Planner, Smith says, "the agencies will never stand for it." Smith, like everyone I spoke to, argued that media planners will resist Google's audience information because no one wants one company to be so dominant in online advertising: were Ad Planner to be widely adopted, Google would be selling keywords through its search advertising network, AdWords; selling banner advertising through its display advertising network, AdSense; serving those ads through DoubleClick; and advising media planners on where to spend their advertising dollars.

Ad Planner also lacks a number of important features that an advertising agency might expect from an audience measurement service. According to Smith, it offers neither very detailed demographics nor a full explanation of its methodologies. Patrick Viera, TechnologyReview.com's own digital strategist and West Coast advertising manager, said disdainfully when I asked his opinion: "Yeah, I looked at it. It doesn't do anything you want. It's just a tool for selling AdSense."

Still, says Smith, there's demand for something new. "Publishers have to use third-party measurements, but third parties [such as ComScore and Nielsen] may underestimate audiences, and the truth is probably somewhere in between. That's why new companies like Quantcast have a chance."

Growing Pains
But neither Quantcast nor Google nor improved products from ComScore and Nielsen Online could, by themselves or in combination, fix display advertising and thereby ensure the future health of media.

Whatever audience measurement tools are adopted, they will themselves have to be validated by an independent party. Quantcast, ComScore, and Nielsen Online (but not Google) are all in the process of being audited by the Media Rating Council (MRC), which was established by the U.S. Congress in the 1960s to audit and accredit the ratings of broadcasters. Accreditation will smooth disputes about the different audience measurement methodologies, according to George Ivie, the chief executive of the MRC: "It will help bring the numbers closer together; and it will explain and make transparent the differences between the census and panel systems."

In addition to the disagreements about the size of Web audiences, though, online advertising suffers from deep structural problems that must be addressed before media planners and their advertising clients will spend really large sums. These are various and dauntingly technical, but according to David Smith, they all involve, in one way or another, the absence of commonly accepted, automated means to create, sell, serve, and track the performance of online ads.

Fixing all that will take years, as will the adoption of undisputed audience measurement methods. "This industry is only 13 years old," says Smith. "It grew rapidly with few standards for six years. Then it collapsed, with very little research and development for four years, and has just been getting back to the right kind of R&D and standards in the past three."

Still, by any estimate, the general confusion about Web audiences is the reason why the online medium has matured in so ungainly a fashion. "It's an amazing topic," wrote Roger McNamee in a conversation using the messaging service of the social network Facebook. "You could see it coming a mile away. Unfortunately, the remedy is not yet obvious."

Jason Pontin is the Editor in Chief and Publisher of Technology Review.

Copyright Technology Review 2009.