“How old do I look?”
It’s a question we’re dying to ask other people, but we hesitate because we’re afraid of the answer we’ll get. And when someone asks us how old they look, we run for the hills. Tell some 30 year-olds they look 40 and you risk wrecking their days — as well as your relationships.
But robots can tread where people dare not.
Trust the Data or the Mirror?
Joseph Sirosh, corporate vice president, Information Management and Machine Learning at Microsoft, and his team built a bot that can tell how old you look.
They leveraged APIs and features from the Cortana Analytics Suite, big data, machine learning, the elasticity of the Cloud and a boatload of photos in the process. They then invited engineers who came to the Microsoft Build last Spring to check it out in a beta setup of sorts. They weren’t supposed to share the link.
But they did.
Face Rating Goes Viral
“What began as a fun, interactive, demo to show the power of real time analytics and machine learning turned out to be so much more,” said Sirosh. How-Old.net went viral overnight as 50 million people uploaded half a billion images, over seven days. To date, 80 million people have uploaded 500 million images.
Sirosh said that the participants were checking out Microsoft’s algorithms, which in the most literal sense is true.
But chances are good that very few of them were thinking about the technology at all. Instead they were busy sharing their results and the link on Facebook at Twitter at such a high rate that social media sites reacted as if they were being attacked by a bot.
While the Microsoft team learned a great deal from what was supposed to be an experiment (including how to look younger), and we’ll share those below, they also relaunched a new algorithm at Strata + Hadoop World this morning.
Instead of finding out how old the data says you look, you can have a stranger answer the question. You can be one of the first to check it out now,
How to Look Younger
But before you upload that photo, Sirosh and his team gleaned some important non-geeky insights from the experiment. If you want look younger in a picture, shave, don’t smile, and don’t wear glasses. Want a prescription for depression? Do all three.
Your Big Data App
If you’re a big data developer building a big data app, Sirosh had some “great learnings” to share as well:
- Build cloud-first. The web can turn a small data application into a big data one in a hurry. The cloud provides the elasticity you need to auto-scale to your traffic and data storage needs. Developing in the cloud can be very agile because of the variety of supporting services you can leverage. It also has compelling economics.
- Watch the metrics in real-time. Live monitoring, real-time analytics and dashboards aren’t optional. They are necessary to know how, and what, your app is doing and how customers are responding to it. They are also pretty incredible to watch if your application takes off.
- Start with big data analytics in the cloud. Cloud-based fully managed services make it easy to set up big data analytics.
- Test and learn. Experimenting is the new normal for app development. How-old.net was built mobile-first, cloud-first which let us rapidly iterate using controlled (or uncontrolled) experiments, with very little developer effort.
Title image by Fade Qu.