‘We value our crowd’: Appen boss Armughan Ahmad defends AI worker conditions

Appen’s boss says the group pays contractors above market rate, amid claims they’re being forced to complete 15-minute tasks in just two.

Appen boss Armughan Ahmad. Picture: Supplied

The boss of Australian AI specialist Appen has defended allegations that the company’s crowdsource workers are being forced to work too quickly, amid concerns that their working conditions could make AI chatbots including Google’s Bard inaccurate and potentially harmful.

Appen contractors last week told technology news publication The Register that they were in some cases being given two minutes to complete work that takes 15 minutes and were penalised for not completing tasks within the allotted time, with Google’s Bard generating less accurate information as a result.

The group has around 1 million crowdsource workers globally who help improve the quality of AI tools and large language models on behalf of the tech giants. Those workers are hired to verify and feedback the accuracy of responses produced by chatbots like Google’s Bard, with the human feedback used to improve the quality of future answers.

The Register reported Appen’s workers are generally limited to working 26 hours per week for up to $US14.50 ($21.7) per hour, and that some of the workers have filed an unfair labour practice complaint with America’s National Labor Relations Board.

“My comment would be that Appen really values our crowd, and we’ve had a really good track record of always paying our crowd above market rate,” Armughan Ahmad told The Australian in an interview.

“Our view is that we continue to value our crowd and that people will be employed to do this work and that’s very important to us, and it will continue to be a part of our strategy. We are building AI with human alignment, and we believe that in order for us to build and train AI with human alignment, you need to have humans engaged as part of the training.

“Once it’s finetuned, which is called reinforcement learning from human feedback, then you need to have instructive datasets, and you need to also then provide assurance services to ensure that it’s not biased. Then you need to do red teaming and benchmarking.”

Appen has around 1 million crowdsource workers globally who help improve the quality of AI tools and large language models on behalf of the tech giants. Picture: Josh Edelson/AFP

Appen shares dropped 10 per cent on Monday after the company announced its chief financial officer Helen Johnson would be departing the company after just two months, which Mr Ahmad said was due to family issues.

Mr Ahmad, who is based in Toronto, said Ms Johnson was unable to relocate from Arizona to Toronto.

The CEO is still deep in the midst of a turnaround effort, attempting to reverse a slide in fortunes for what was once the ASX’s strongest performer, and convince investors it can take advantage of new opportunities in generative AI.

Since Mr Ahmad took on the CEO role in December, Appen has added a chief revenue officer, a new chief marketing officer and chief technology officer and has tapped the market for $60m in fresh capital.

“I think we’ve added really great leaders, while we’ve also promoted some people within the organisation,” he said. “We’ve been very busy. And when you’re when you’re doing all this hiring, and you’re attracting some really great talent, they must see something that I see. A bunch of these people I’ve worked with in the past, and they really value finding a great opportunity and doing the hard work.

“Earlier this year we announced $US46m in cost reductions, we have already exceeded 60 per cent of that, and we are going to be cashflow positive by the end of the year.”