Singapore app pays users ‘pocket money’ for data samples to train AI assistants

SINGAPORE – A Singapore start-up has turned to the public to build an army of data trainers, who earn “pocket money” by performing tasks such as recording their voices and labelling pictures.

And anyone with a smartphone – even those who are not tech-savvy – can perform these tasks using the Tictag app, earning several cents to a few dollars each time.

The data collected is used to train artificial intelligence (AI) voice assistants and image recognition technologies in speech and language, safety monitoring and computer vision.

The app has gained at least 20,000 users on Android and iOS since its 2021 release, and Tictag has paid out about $24,000 in all, said Mr Kevin Quah, chief executive and co-founder of Tagteam Technologies, the firm behind Tictag.

The start-up assists tech firms in collecting data, labelling datasets and training AI models.

For instance, some of Tagteam’s clients use AI-powered camera systems to identify whether their employees are wearing protective gear or entering hazardous zones. Other clients use voice recordings to train AI systems to estimate a user’s age based solely on their voice, for age verification and other purposes.

Data training has traditionally been the job of data scientists and annotators.

But amid a shortage of manpower to meet AI’s ever-growing demand for data, Tictag’s creators enlisted everyday users through its app, Mr Quah told The Straits Times in an interview at Tagteam’s office in Chinatown.

Founded in 2019, the firm works with tech companies, directly or through outsourced vendors. It is now among the start-ups accredited by the Infocomm Media Development Authority (IMDA), allowing it to do business with government agencies and large enterprises.

It has received US$3 million (S$3.96 million) in investments as at July, said Mr Quah.

Tictag users are assigned bite-sized tasks, such as recording words spoken in Hokkien or saying “Bixby” – the name of Samsung’s AI assistant on its mobile devices – at different volumes.

The tasks often entail annotating images. This requires users to circle components in pictures, such as the brand logo of a car, or to label the colour of an item. This provides the AI with accurate training data to recognise images.

The app cross-checks answers against samples from various users, to ensure consistency and spot any anomalies.

In addition to human checkers, the platform’s software is also able to filter answers that are blatantly wrong.

The more time-consuming tasks include taking pictures of wildlife or uploading images of e-commerce receipts to train an AI model’s ability to understand receipts, such as in the processing of claims.

Users earn digital coins that they can exchange for supermarket or ride-hailing vouchers, or cash that is transferred to their bank account.

App users can easily earn rewards worth several dollars by completing tasks, said Mr Quah.

Tagteam has also tapped more than 120 users from marginalised communities in Singapore and Asia as data annotators and checkers, to validate users’ answers.

By working with at least 14 social organisations in the region, the company provides users such as low-income seniors and people with autism and disabilities a chance to earn extra pocket money, said Mr Quah.

Mr Low Junin, a trainee at SPD Ability Centre, a disability support organisation that Tagteam has partnered, said he has earned more than $100 from the app since 2023, by completing tasks in his spare time.

The app offers a way to familiarise himself with data jobs, said Mr Low, 33, who hopes to find a job in the sector.

Freelance graphic designer Allan Tan, 55, uses the app casually as a way to make extra pocket money, also chalking up about $100 so far.

The tasks usually take around five to 15 minutes, allowing him to make up to $5 for each task completed, said Mr Tan.

“The app has helped me learn something new and provided some side income too, which is a bonus,” he said.

To encourage users to take the challenges seriously, coins are awarded based on the accuracy of the tasks completed, said Mr Quah. Users who complete 80 per cent or more of the tasks correctly will receive the full coin rewards, but those who score below 50 per cent earn nothing.

Users receive the rewards several days later, after the company vets the results manually and with the help of AI, said Mr Quah, adding that those with a good track record are rewarded with more tasks, which can unlock more coins.

Data is the lifeblood of AI models, enabling digital services to be more reliable. But it needs to be properly categorised in order for machine algorithms to put it to use, Mr Quah said.

“Everything you train for needs 1,000 to 3,000 samples of that one example, that’s why data is constantly required.”

He added: “The demands keep getting higher. Today, it’s labelling colour, but next, we could be asked to highlight dents in a car or the number of people in a car.”