How does a machine learn? Like anything else, it gets trained.
But unlike animals – and human animals – which still learn from each other, most machines haven’t reached the stage where they can train themselves. Instead, they rely on humans. Since machines learn by establishing patterns they create by being exposed to enormous amounts of data, they still need a lot of help.
Say a machine or algorithm needs to learn how to distinguish people wearing hats from people not wearing hats, for example. If it’s exposed to millions of images of people wearing hats and of people not wearing hats, it will develop a method to distinguish one from the other and then transfer that inference to new images.
The people who power that ability to distinguish by lending their already-advanced discernment to the machine learning process are becoming an increasingly important part of the workforce.
Earlier this month, Appen, a global data annotation company headquartered in Chatswood, Australia, announcedOpens a new window that it was acquiring machine learning and artificial intelligence company Figure Eight for $300 million.
Figure Eight, based in San Francisco, got its start in 2007Opens a new window when its founders asked remote workers at the most well-known crowdsource work platform – Amazon’s Mechanical Turk – to make judgements about personality characteristics based on the way people looked in photos.
The project was mostly undertaken on a whim, but it eventually led to the company taking on increasingly relevant tasks. And that’s where crowdsourcing labor comes into the picture (pardon the pun).
For Appen and Figure Eight, some of the most important assets of their business models are the enormous, disparate labor pools they tap into in order to power their services – whether for annotation or transcription, the work often serves a dual purpose: completing the project at hand while also training algorithms to do the work in the future.
In some cases, the machine training is the entire purpose, while in others, like transcription, there’s a dual application. But whatever the end use of the labor, the people performing it are essential.
Appen, for example, boasts accessOpens a new window to a “curated crowd†of more than one million flexible workers scattered around the world, and to whom the company turns for its many projects. Appen says it has reached three billion judgements and processed 500,000 hours of audio, with a core employee base of only 500 people in nine offices around the world.
It’s not the only example of a firm with such a large labor pool: another company called ClickworkerOpens a new window , which has workers listen and rate responses in conversations, has built a million-person crowd of flexible workers as well.
The Appen model is playing a growing role not only in the organization of companies, but also in the work structures of the people powering their projects. However, we’re not talking only tech companies that rely on human perception to help them develop better machines.
Crowdsourcing labor has become ubiquitous. Ride-sharing apps and delivery services, for example, depend on flexible labor pools. Even digital security operationsOpens a new window have made use of crowdsourced labor.
Yet, some of these companies have run into trouble over labor laws, and there have been serious and legitimate concerns about fair compensation. One studyOpens a new window of Mechanical Turk, for example, found that workers earn on average less than $2 per hour.
On the other hand, the rise of crowdsourced labor pools could offer significant lessons for human resources, especially in tech.
Among the advantages of flexible work programs, contributors can work when they want – and not work when don’t. They often can choose where they work, for how long, and their general working conditions.
In some cases, they can work from home with a family as they’re earning money online. In addition to machine learning companies that require human labor, options to work from home on a freelance basis have exploded.
Depending on the nature of the work, many aspects of crowdsourced labor can be incorporated into the workplace, from the worker’s perspective as well as from the employer. Delineating tasks that employees across a company could complete according to their own timetable would benefit both companies and laborers, especially for projects based on company self-evaluation or performance.
Crowdsourcing labor can be a benefit to many companies – at least until the machines take over.