Thanks for coming to the 2012 Terrarium Holiday party. We hope you enjoyed it as much as we did. See you next year? Fully embarrassing, full color images available on Facebook.
Frank was an experiment, an opportunity for our team to see how far we could push the application of Amazon Mechanical Turk. Along the way we learned some interesting things about optimizing usage of Mechanical Turk that we wanted to share with the larger community.
There is a lot to talk about so before the tl;dr chorus starts up here is a high level summary:
- Utilizing Mechanical Turk is not a pure engineering task. It is as much social science as computer science.
- To optimize, Mechanical Turk requires three hats: developer, social engineer, and administrator.
- Great interaction design can help mitigate price sensitivity of Mechanical Turk workers.
- Constant monitoring is the only reliable approach to consistent high quality.
Mechanical Turk is challenging (and interesting) because it is fundamentally different from the model that computer science is based upon. A developer is accustomed to working with deterministic resources. Within that paradigm everything is the result of an input/output process.
Frank was a grand experiment in the limits of Amazon Mechanical Turk (see his best work here). How complex of a task can Mechanical Turk handle? Can humor be crowdsourced? At what volume can a complex task be handled efficiently with reliable quality by Mechanical Turk?
With Frank, we put some serious task volume toward answering those questions. Over the last weekend Frank had a throughput of a photo every two seconds and continued to accelerate. At that rate we were able to acquire enough usage data to draw key conclusions. And with experimentation being the primary goal, today we decided it was time to pull the plug on Frank.
With a heavy heart, we announce that the grand experiment that was Frank has reached its conclusion. Hope you had as much fun as we did.
We learned a great deal about Amazon Mechanical Turk in the process (that was our goal from the outset). Stay tuned to this space for a longer post detailing our conclusions.
In our last blog post we introduced you to our latest experiment: Frank (Frank Said What? in the App Store). As Frank has made the rounds and countless new friends one question has come up over and over: How did Frank work?
For the weeks following Frank’s launch we played our cards close to the vest, pointing people to this video in response to any “how does it work” questions.
Today is the big reveal.
Frank is our newest iOS app. We found Frank while doing UX research in preparation for our next big project. As hard as we tried, we couldn’t leave him behind. Frank is hard to quit.
So what does Frank do? Frank is the most sophisticated photo captioning app you’ve ever used. More than that, Frank is unpredictable, hilarious and addicting.
The secret to Frank’s magnetism? Instead of randomly assigning pre-written captions, Frank looks at every photo and writes an individual response. Even we are surprised from time to time by the things that Frank says. The end result is the perfect mash-up of Instagram and your favorite meme.
At a party?
Impress Frank with your friends.
Bored and need a quick pick-me-up?
Send Frank a photo and spice things up.
Find something strange on your walk home?
See what Frank thinks.
Let’s take a quick tour.
After you login through Facebook or Twitter (don’t worry, Frank won’t post on your behalf unless you want him to), you are brought to the main feed of recent and popular Frank photos. If you see one that you love, you can tap the mustache icon (1) to “stache” the photo.
Sending a photo to Frank is always one tap away. Tap the camera button, either take a photo on the spot or select one from your phone. Then all you have to do is wait.
It typically takes 3-5 minutes and while you wait you can browse around the app, or jump over to something else and wait for the push notification (2) that Frank has finished your photo. All of your photos are collected in your personal tab so you can get back to them easily.
It’s easy to share a Frank: just click any of the share links (3) and you can zip it off to your Facebook, Twitter, e-mail (4) or receive a standalone link (like Instagram but way more fun).
Why would Terrarium make an app like Frank? We have been experimenting with different approaches to object classification for our next big project. One approach had the unintended consequence of returning descriptions of photos that were often hilarious. Naturally, this quickly turned into a competition to see who could produce the best image and description combination. And thus Frank was born.
If the technical piece piques your curiosity, check back soon as we will be talking more about Frank’s inner workings and behavioral oddities.
Hello. Fantastic to meet you.
This is our new blog. Don’t be a stranger, we have big things planned.
Terrarium is a growing team born from the MIT Media Lab and inspired by the intersection between culture and technology. We built Peddl as an alternative model for buying and selling online. Peddl required deep collaboration between disciplines that don’t usually cross paths. That process laid the philosophical foundation for Terrarium.
Why Terrarium? Terrariums are used to cultivate unique ecosystems on a miniature scale. You can grow all kinds of strange new things in a Terrarium. New growth is our goal.
And when your goal is that simple it enables you to operate a little differently. You can fly in the face of conventional wisdom. We are scientists, artists, engineers, designers and writers working together without titles or roles. We are all different and that is what matters. We chase creative tension. We make a mess.
The only expectation is invention, growth. And we don’t constrain that with industry, category or expectations. If someone has a good idea they run with it. See what sticks. See what grows.
Check back soon and expect the unexpected.