Labor isn’t what it used to be. Where in past years the expectation was that jobs were done at a certain place and time, now there are entire swaths of work that can be accomplished by anyone, anywhere.
Lukas Biewald, CEO of CrowdFlower and a speaker at next month’s Web 2.0 Expo in New York, is at the center of the labor shift. His company has found an interesting way to tap Internet-connected groups to get work done — think Mechanical Turk, but with additional tech and quality-assurance layers added on. What’s really surprising is that many of the groups CrowdFlower turns to would never define themselves as formal workforces.
Biewald covers a variety of topics in the full interview, including:
- He sees similarities between “labor on demand” and cloud computing: both keep costs down and reduce the risks associated with scale.
- “It’s hard to explain my business to my mother,” Biewald says. In the interview, he digs into CrowdFlower’s unusual — and somewhat complicated — business model.
- He provides further proof that virtual currency is a big deal: Around half of CrowdFlower’s work involves it in some fashion.
- He acknowledges that distributed work has a disruptive and negative affect on many businesses. However, Biewald believes it’s a “rising tide” that will “increase the GDP of the world.”
Read more after the jump.
What is “labor on demand”?
Lukas Biewald: What labor on demand means to us is that you can access tens of thousands, or hundreds of thousands, of people instantly. Truly instantly. You send us a job and we post it online through all of our different channels, and we get lots of people working on your job all at once. Or, we find the specific person that’s best for your job.
It’s exciting for businesses because they can scale up and scale down. Just as cloud computing made it so businesses didn’t have to predict how many servers they were going to need at any given time, labor on demand allows businesses to not have to predict how many people they’re going to need at any given minute.
Can you walk me through a typical CrowdFlower job?
LB: Let’s use a business listing verification as an example. Suppose Yelp, which isn’t one of our customers, gets a complaint saying that a business address is incorrect. Now, they don’t want to take the business down immediately, but they do want to respond to that complaint as fast as possible. So they send that complaint to our system, and we post the job. It gets sent all over the world to everyone who happens to be in the CrowdFlower system at the time. Anyone can grab that job, and they will call the business in question or visit the company’s website — whatever the specific instructions are.
Sometimes we have multiple users employ different strategies. So two people call the business and one person checks the website. That’s why we’re a technology company; our process involves redundancy. We’ll have workers spot-checking each other. We’ll use all kinds of automated systems to prevent fraud and errors.
Returning to the example, we get results back and decide we’re 98-percent confident that this specific business does exist. We shoot the information back to Yelp. The cool thing is all of this happens within five minutes or so. It means the client can always have updated information instead of waiting a day or a month for someone to get around to checking.
How do workers become aware of jobs?
LB: This is the thing that’s a little bit complicated about CrowdFlower’s business model. It’s also why it’s hard to explain my business to my mother.
We make deals with other companies that have lots of people around: outsourcing companies, e-rewards companies, even game companies. For every job that they get someone to do, CrowdFlower will pay the company money and that company can incentivize users with things like free seeds for a game or airline miles. We also put jobs inside Amazon Mechanical Turk, which then pays people small amounts of money for doing tasks. We have an open API where anyone can monetize people if they can get them to do these tasks.
We’re not in the business of actually collecting people. We go through channels, and most of our channels don’t consider themselves workforces.
What we do is add quality control on top of workforces. That was the core technology that started the business. Not every worker is going to be good at every task, but our system can figure out who’s going to be good and who’s going to be bad.
Is virtual currency a big part of your business?
LB: People think it’s a minor part of our business, but it’s actually a huge source of work for us: It’s about half virtual currency.
How do you address complaints from professionals who are disrupted by crowdsourced work?
LB: Any sort of important technology or shift has winners and losers, right? Look at something like 99Designs, which professional designers often complain about. They say the quality of designs is going down. Yet, at the same time, lots of businesses that were just designing things themselves are now getting things designed.
Look, I believe in minimum wage. I think people should have to pay a certain minimum amount of money. Employers shouldn’t be able to exploit people.
Overall, the amount of work out there is increasing. There are certainly tons of examples of people that are hurt by this, but I think it’s a rising tide. This will actually increase the GDP of the world.
What’s your take on the relationship between human work and machine learning? Are they at odds?
LB: With machine learning it’s easy to get to 90 percent a lot of the time, but it’s hard to get to 99 percent. And 99 percent is what you need for lots of applications.
If you look at machine learning implementations in the real world, they almost always use a technique called active learning to more efficiently collect data. Active learning is based on the idea that human beings and machine algorithms learn best when presented with confusing information.
For example, think of a machine classifier that’s trying to decide if a boy or a girl is the subject of a photo. If you show tons of pictures that are obviously boys and tons of pictures that are obviously girls, that’s not going to be as effective as showing a few obvious boys and a few obvious girls and then lots of examples where it’s tricky.
When the algorithm gets tripped up, companies can send us the examples where the classifier is confused. We then categorize the examples that will help the algorithm improve.
That’s a specific example of machines and people working together that excites me about my business. I think it’s the future of machine learning.
Related to that, there’s a discussion about why we have more digital work. The assumption was there’d be less because as more gets automated, you’d expect outsourcing to decrease.
Yet, outsourcing — digital work in particular — has taken off over the last 20 to 30 years. I don’t have a great sound byte answer to explain that phenomenon, but I have seen that it’s the companies that do the most machine learning that end up with the most of these kinds of tasks.
As we get used to automated processes working well, they leave a big trail of stuff that needs to be dealt with by people in their wake. So, I absolutely don’t fear artificial intelligence and I don’t view it as a competitor. Mechanical Turk uses this brilliant term: “artificial, artificial intelligence.” It has all of the benefits of people and computers.
What will you focus on in your Web 2.0 Expo New York keynote?
LB: There’s this amazing phenomenon going on. Companies are taking core parts of their business and they’re sending them not to outsourcing call centers — like they were doing 20 years ago — but sending them to millions of people distributed around the world.
I want to talk about not just how this phenomenon is affecting business, but also the social impact of anyone in the world who has access to a computer and broadband. It’s a surprisingly large number of people, and these people are now able to compete in a global information marketplace.
When you try to open a factory, it’s a tough process. You have to ship goods. You have to go through all kinds of regulations. But when you shift information back and forth, there’s very low overhead. There’s no need to send hundreds of people into an office and buy them computers and infrastructure. Many people already have the tools to be effective global workers. We’re seeing people in refugee camps actually finding useful, meaningful work and making money off of the refugee camp information infrastructure. This is changing the way business operates and also changing the fabric of the world we live in.
This interview was condensed and edited.
- How crowdsourcing helped Haiti’s relief efforts
- Mechanical Turk app on the iPhone Provides Work for Refugees
- Crowdsourcing and the challenge of payment
- What is data science?
- Data science cheat sheet
Lukas Biewald will discuss the business and social repercussions of distributed work at the Web 2.0 Expo in New York, being held Sep. 27-30. Save 20% off registration with the discount code “radar.”