Over the past decade, the relationship between humans and technology has grown so deep that technology’s influence is now felt across almost every aspect of our lives.
Everyday tasks such as ordering groceries, navigating to your next destination or checking your bank balance now can all be done at the touch of a button, and from the comfort of our own homes. The pandemic has evolved technology from convenient to essential, as tech solutions have allowed us to continue performing everyday tasks without exposing ourselves to unnecessary health risks or contributing to the spread of the virus.
We may take these services for granted and, in the words of Arthur Clarke, they are almost “indistinguishable from magic” for us non-techy mortals. In reality, many of these apps and gadgets depend on artificial intelligence (AI) and machine learning (ML), which require vast quantities of data to develop and improve.
While we may understand that AI and ML play a role in our modern lives, what many don’t realise is that this sophisticated technology is supported by a secret army of real, physical workers. As our dependence on tech has grown, concerns have been raised around man vs. machine, namely, that robots could perform jobs that previously hired people. The truth is that machines rely on millions of humans from around the world to feed data to the algorithms that keep our favourite apps and gadgets running.
Specifically, humans label data for AI developers, and their algorithms’ appetite for data is bottomless. In 2019, analyst firm Cognilytica reported that the market for third-party data labelling was $150 million in 2018. The market is forecast to grow to more than $1 billion by 2023 as the technology continues to develop.
The physical workers operate on a cloud-based, crowdsourcing platform, choosing whenever they want to spend time completing tasks posted to the markeplace by companies from across the globe. Most of these tasks consist of some form of labelling and usually involve text or image classification or checking audio transcriptions. This data is then either fed back into AI/ML algorithms or used for other kinds of business process automation (such as human-in-the-loop moderation) or for field crowd tasks (such as checking local information to improve maps and navigation).
A wide range of innovations benefit from crowdsourced data labeling, including self-driving vehicles, voice assistants and smart speakers, banking models, ad targeting, search engine optimization and more.
Switzerland-incorporated Toloka is one of these platforms. After a steady start feeling its way into the industry, Toloka now serves the needs of some 2,400 businesses worldwide and continues to expand its reach. It boasts a worker base — who are referred to as ‘Tolokers’ — of around 9 million people. These are located in over 100 countries across all time zones, making this a real 24/7 operation. Toloka believes it is only just scratching the surface of its potential workforce, which it estimates to be 400 million-strong.
Toloka is not alone in this burgeoning space. Since 2019, no fewer than five other crowdsourcing services have raised capital in equity markets, including ScaleAI, who was recently offered funding that valued the company at over $3 billion. At present, one of the oldest industry players is the ubiquitous Amazon, which has long been operating in this market through its Amazon Mechanical Turk platform.
Not being the first to arrive at the high-stakes table may have had its own benefits for Toloka, which was set up in 2014. Having observed the criticisms levelled both at Amazon and the crowdsourcing model more broadly, the company has put two key issues at the forefront of its business model.
First, it’s set to invest more in its workforce, thereby empowering it. For example, at Toloka the performers give feedback on task clarity and fairness of pricing, so it’s not just the requesters that have a voice. This feedback shapes the ratings of requesters and helps improve the competitive task assignment system that directly benefits both requesters and performers.
Moreover, heeding claims that the crowdsourcing model could be exploitative, Toloka is determined to give its workers tangible educational and employment benefits. Its training process provides Tolokers with new, transferable skills, while its flexible part-time nature helps them earn extra disposable income while working from home.
The company has also taken steps to quell customer concerns that crowdsourced quality could be volatile and unreliable. To address this, Toloka has introduced a unique quality assurance mechanism where tasks are smartly matched with performers, using AI-based predictions of the completion probability and quality, resulting in a final product of comparable quality to what’s produced by in-house labelling teams, but at 10x the speed and at a significantly lower cost.
Although the industry faces scrutiny, it plays a vital role in both continuing society’s technological advancement and maintaining the applications and systems on which we rely. Addressing concerns around quality and working conditions requires an ethical, sustainable approach that puts workers at the core of the business model.
Crowdsourcing platforms cannot be exploitative, but must enrich workers. Companies like Toloka, ScaleAI and Amazon aim to find this balance, and some claim to have already found it. Whichever company is most successful in this regard will be positioned to dominate the market in an industry primed to boom over the coming years.