Microsoft debuted a new website and iPhone app this week. The app, called “Fetch!,” and the site, found at “What-dog.net,” offer users the chance to have a submitted photo matched to a specific dog breed. Over the course of the day Thursday, I started seeing posts trickling through my Facebook feed showing the real entertainment value of the programs, to wit, having pictures of yourself and your friends comically compared to dogs. If you love dogs, laughing, or are just looking for a new way to waste time on the Internet, it’s a winner.
If you’re looking for a fool-proof way to tell what breed your dog is … well, let’s say there’s still quite a bit of work still to be done on that front. I have a distinct feeling that we’re going to be seeing a lot of posts and screengrabs from the Fetch! app across all of our social media sites over the next several weeks. I’ve been road-testing both the web portal and the iPhone app with a number of photos — of my own dog and assorted other subjects — and I’m ready to report on its accuracy and fun value. Let’s get started and see how effective the Microsoft Garage’s experimental artificial intelligence really is!
Can the “What-Dog.net” site recognize my dog?
My dog, Baby, is a 2-year-old Bluetick Coonhound mix. What is the “mix”? I have no idea. Can Microsoft’s AI reliably inform me? I first tested the web portal by uploading a photo of Baby that I took a couple of weeks ago. Here is the result that the site, and its #WhatDogRobot, returned:
Interesting result. A Brittany is a bit smaller than a Bluetick Coonhound, but the site’s artificial intelligence certainly noticed Baby’s floppy ears and voluminous coat. They’re both scent-driven hunting dogs, so there’s that, I suppose! As you can see at the bottom, the web portal offers users the chance to report inaccuracies. I chose not to offer corrections during my experiments to avoid influencing the AI. Let’s try another, this time with the iPhone/iOS app.
Is the Fetch! dog app any more accurate?
Installing the Fetch! app was very easy. It’s free, and there’s no sign-up or registration. I opened it, uploaded a different, more head-on photo of Baby, and got a completely different result.
Another miss for the Microsoft dog-recognition robot. A fully grown female Anatolian Shepherd weighs anywhere between 85 and 120 pounds. I shudder to think how hard Baby would pull me during our forest hikes if she had an additional 20 pounds of muscle on her! Let’s give the app another photo and another attempt to correctly identify my dog. Perhaps I’m not giving the app a sufficiently clear look at her.
Within the iPhone app, you can zoom in on the image for a different perspective on the same photo. I ran this same snow photo again, this time, zooming in so that the frame included only my dog. The second attempt was no more helpful, giving me Beagle as a result. A bit closer, maybe, than Airedale Terrier, but not by much. We’ll give it one more roll with yet another photo!
Fabulous! Bluetick Coonhound is correct! Success at last! Clearly the artificial intelligence’s algorithms respond to the angle and clarity of the photos that users offer it, as well as the lighting and perspective.
Differences between what-dog.net and Fetch!
After a series of trials, there are distinct differences between the web portal and the iPhone app. The app is much more responsive and interactive than the “What-Dog.net” site. The website offers only a result, without any of the additional functionality that makes the smartphone app so much fun.
- Tapping on the 99 percent match badge, for instance, reveals the variety of breeds that the app attempted to choose between. Drilling down, the app tells me it’s actually 100 percent certain that Baby is a Bluetick Coonhound, with 0 percent matches for Australian Cattle Dog, Bloodhound, Border Terrier, and Airedale Terrier.
- Tapping the small orange triangle at the bottom right gives the user a breed profile of the top match. This includes a brief history, physical characteristics, and temperament.
- The three buttons you see along the bottom of the phone screen give users the opportunity to report inaccuracies, share on social networks, or save to your mobile device.
What does Fetch! make of a painting?
In September, I visited the National Gallery of Art in Washington, DC. Even though I was nominally on vacation, I’m always on my grind, and ended up documenting a number of paintings that included dogs. The one I liked best was Francisco de Goya’s portrait of “The Marquesa de Pontejos” (c. 1786), which depicts the subject with her little Pug. Could the Fetch! app determine the breed of a dog in a painting? Let’s find out!
A rousing success! Of course, it doesn’t take an artificial intelligence to know a Pug, but being able to recognize even a well-known breed from a 230-year-old oil painting is pretty boss for an iPhone app. It also means the detail photo I took while I was at the museum was not too bad.
What else can Fetch! do?
If my social media feeds are any indication, the most fun part of the Fetch! app and the What-Dog.net site is using a dog-identification app to have a good time gently mocking ourselves, our friends, and family members. In point of fact, all of the screengrabs I’ve seen posted over the last day were photos comparing humans and children to dogs and puppies.
For instance, the app thought my friend Julia resembled nothing quite so much as an Irish Water Spaniel! Immediately I could see the true value of this technology. I’m not going to throw any of my own pals under the bus, so instead, I offer you Fetch’s interpretation of myself. What dog does the app think I am while singing karaoke at the Pinhook in Durham, North Carolina?
I’m a Staffordshire Bull Terrier! “Strong, driven to prey on smaller creatures, affectionate”: That’s almost a perfect description of my karaoke performance style.
Share your funniest results!
Its accuracy with dogs is still questionable at this early stage, but who knows? The more that people use and correct the results for dogs they know and own, perhaps Fetch! will gain greater precision. Hopefully, the artificial intelligence will not become too knowledgeable. The last thing we need is a dog-recognition app taking over the world. As it stands now, it is one of the best dog apps out there, ranked on entertainment value alone.
Have you tried the What-Dog website or the Fetch! app? Can it tell what breed of dog you own? What manner of canid does it say you look like? Have you used it to mercilessly poke fun at your friends and family? Share your experiences with the app in the comments!
About the author: Melvin Peña trained as a scholar and teacher of 18th-century British literature before turning his research and writing skills to puppies and kittens. He enjoys making art, hiking, and concert-going, as well as dazzling crowds with operatic karaoke performances. He has a two-year-old female Bluetick Coonhound mix named Baby, and his online life is conveniently encapsulated here.