There's a reason the self-driving cars now being tested by dozens of auto and mobility companies aren't mainstream yet. They use some of the most advanced artificial intelligence ever built, requiring cutting-edge computer vision and billions of calculations per second. Pricey stuff. Yet to teach that computer to see demands something much less exotic: everyday citizens performing simple tasks for a few bucks an hour. 

MightyAI is a computer vision company that trains artificial intelligence programs to better see and understand the world. That way, it can better differentiate between, say, a paper bag blowing in the wind and a cat trying to cross the street. The startup pays people to perform simple image recognition tasks on their phones or desktops. For a few cents or dollars per task, a person might receive an image of a city street and be asked to outline all the pedestrians, cars, street signs, or whatever objects the system requests. That info is then used to strengthen data sets for firms developing computer vision technology, including self-driving car companies.

CEO and co-founder Daryn Nakhuda is cagey about the company's client list, citing the automotive industry's secrecy about autonomous vehicle technology. "It's the automakers themselves," he says. "It's the tier-one suppliers, some of the startups building hardware around perceptions, around cameras or other sensors."

One customer the company can mention is Mcity, the University of Michigan's vehicle testing facility. Members of Mcity get access to MightyAI's data set; the facility counts Ford, GM, Honda and Toyota among those members.

The Seattle-based MightyAI screens users before it allows them to join its platform, called Spare5. Each person gets a test, similar to the real thing, in which they have to outline specific objects in photos. It's not always mindless work: If a car is partially obscured, for example, the user must estimate where its border would run. 

Pass the screening test and you're given new batches of photos on which to perform similar tasks. Other assignments can include creating keywords to describe a photo or labeling specific objects within it. Pay ranges from 1 cent to a few dollars per task, depending on the complexity. Each can generally be performed in a few seconds.

Don't go thinking you've found an easy new full-time gig, though: There are a limited number of tasks available at any given time based on the company's needs. And MightyAI admits the average person would tire pretty quickly given the repetitious tasks, so spending a few hours a day on the app isn't realistic--it's meant to provide some extra change when you can spare a few minutes. (Hence the name Spare5.)

Nakhuda​ says that more than 500,000 people have signed up on the platform. The company targets users in places like Instagram, Facebook and Twitter.

Continually training the system with human input makes the company's data set stronger, and thus valuable to companies creating computer vision software.

"The least sexy but most important part of what most AI companies are working with is the data," Nakhuda says. "You need to have well labeled data. Even the best algorithm only performs as well as what you've trained it with."

Founded in 2014, MightyAI was initially launched for the purpose of developing language processing technology. After deciding that big players like Amazon and Google were already dominating that realm, the startup turned its primary focus to image processing with self-driving vehicles in mind. 

"We saw that the complexity and the quality requirements were so much higher in automotive because of safety and because of the nature of the data," Nakhuda says. "We thought it was a real opportunity for us."

The 85-employee company has $27 million in funding. Its investors include Intel, GV (formerly Google Ventures), and Brad Feld's Foundry Group.

While autonomous vehicles remain the company's main focus, MightyAI says its object identification tech is also used to help power online shopping platforms and delivery bots.

Published on: Dec 20, 2018