‘Cold and callous’: The Uber Eats algorithm is sacking people automatically

One day this spring, Abiodun Ogunyemi, an Uber Eats driver with an approval rating of 94 per cent, peered into his phone and read the words: “we have made the difficult decision to end our partnership with you.”

Ogunyemi was being fired. Uber’s facial recognition software, named the Real-Time ID system and made by Microsoft, had failed to recognise him. This triggered the company’s safeguarding protocols – which stops people other than drivers using an account – and terminated his profile.

“We understand that this news can be upsetting, but our decision is final and Unchangeable,” the message went on, bold type included.

Ogunyemi, 46, was confused at first, then angry. He’d done nothing wrong. “I’ve always worked according to the rules,” he said. He spent weeks trying to clarify, appeal and escalate. But the company sent the same boilerplate messages. “It was as if I was talking to remote-controlled people, or a robot,” he told The Big Issue.

Ogunyemi is one of hundreds of couriers and private hire drivers – the majority of whom are people of colour – who say they’ve been unfairly fired by an algorithm. The Independent Workers Union of Great Britain (IWGB) said over 200 people had contacted the union over the last year complaining of unfair sackings from app-based companies, many relating to buggy facial recognition software. There are fears drivers could be thrown into poverty.

The sackings have provoked a backlash from unions and MPs. They are demanding a fairer termination process that would bake in the right of employees to explain themselves before automatic firing, to appeal after dismissal and to union representation.

“It’s a very untransparent process, which many times just involves switching off the app,” Ian Byrne, a Labour MP, told The Big Issue. Byrne tabled a parliamentary Early Day Motion on unfair dismissals last November, supported by 72 MPs.

“It’s an extremely cold and callous way to run an organisation,” he said.

Ogunyemi, who lives in Manchester, began working for Uber Eats after being laid off as a bus driver in March 2020.

“It started becoming a passion,” he said. The flexibility meant he could spend more time with his three children – aged 17, 14 and six – and gave him a chance to get out the house and talk to people during quarantine. “To keep your sanity in check and all that,” he said.

Then around October, Ogunyemi recalls failing Uber Eats’ ID system for the first time. His hair and beard had grown amid lockdown compared to his file picture. (“There was no avenue for you to go and barber your hair,” he said.) But the system has a human backup, who quickly verified him, so it was no obstacle. Over months this happened multiple times.

After Transport for London (TfL) suspended Uber’s license in 2017 over safeguarding concerns, the company in 2020 introduced its real-time facial verification system. This checks only verified drivers pick up passengers or orders.

But Microsoft’s facial recognition software is blasted by critics as ineffective and discriminatory. In a 2018 MIT study, the software was found to be more error-prone on darker skin, with an error rate of up to 21 per cent for dark-skinned women.

(A Microsoft spokesperson said: “Microsoft is committed to testing and improving Face API, paying special attention to fairness and its accuracy across demographic groups. We also provide our customers with detailed guidance for getting the best results and tools that help them to assess fairness in their system.”)

Uber’s London license was reinstated in September last year after Westminster Magistrates Court found its tweaks had satisfied TfL safety requirements.

“So TfL has been instrumental in driving the introduction of this tech,” James Farrar, general secretary of the App Drivers & Couriers Union (ADCU), which has beaten employers on appeal to reverse driver dismissals, told The Big Issue.

Uber was contacted for comment. The company told TechCrunch earlier this year: “While no tech or process is perfect and there is always room for improvement, we believe the technology, combined with the thorough process in place to ensure a minimum of two manual human reviews prior to any decision to remove a driver, is fair and important for the safety of our platform.”

Farrar added: “That’s not a good enough answer though, is it, if you’re the one who’s had your life turned upside down?” The firings sting especially painfully for Uber taxi drivers, he said, because they also automatically lose their private hire license – their livelihood.

In March this year, Ogunyemi failed the check again. He said Uber Eats asked if he was letting anyone else access his account. No, he replied, but perhaps it would be easier to add his wife, who also had an Uber account, to his profile. The next day, after 4,993 total deliveries, his account was suspended.

It’s stressful, frustrating, annoying – you feel like the whole world is crashing down on youAbiodun Ogunyemi

Abiodun Ogunyemi

“In a recent account audit, we were not able to confirm that the photo you submitted was actually you. The photo of the person signing in did not match the driver profile photo that we had on file, which we take as an indicator of account sharing,” Uber Eats told him, in documents seen by The Big Issue.

There was no opportunity to clarify or appeal with union representation. He contacted the IWGB to help fight the decision, which has protested app-based terminations, but nothing changed.

Byrne said automatic terminations like this have created a culture of fear in the gig economy. “You can imagine living under that sort of fear, this faceless entity that can pull the rug from people’s feet in seconds,” Byrne said. In other sectors your job becomes more secure after you pass probation. In the gig economy insecurity never goes away.

Facial recognition is just one way couriers and cab drivers complain of being fired by an algorithm. According to the IWGB, Deliveroo riders can be automatically dismissed over late deliveries, even during app malfunctions, and drivers for gig economy couriers Stuart can be deactivated for fraudulent activity when its GPS system plays up.

The impact on Ogunyemi turned his life upside down. “You want to provide for your family,” he said. And when you can’t, “it’s stressful, frustrating, annoying – you feel like the whole world is crashing down on you.”

Ogunyemi said he was still paying his mortgage plus £121 a month for his courier insurance and £167 on car finance repayments. He asked the latter for a payment deferral (they accepted, adding interest). Luckily, he managed to find work around six weeks later.

But cash was tight. Ogunyemi asked his 14-year-old son to borrow his 17-year-old brother’s bus pass one day to get to school. It was confiscated. “My boy got stranded in town,” he said.

Farrar said there’s been no debate about whether AI should be allowed. “In terms of facial recognition, we need to have a fundamental debate about whether or not we should be doing it at all,” he said. Farrar added Uber, a private company, has avoided the consultations and scrutiny that would accompany a decision by TfL to introduce the technology – AI by stealth.

A TfL spokesperson said Uber, not TfL, chose to use facial recognition software to fulfil its safety and verification requirements, and there is an appeals process for suspended private hire licenses.

But even when licenses are reinstated, the debt can linger for years. George Ibekwe, a 55-year-old former Uber driver, was fired in 2017 after a customer complaint and had his TfL license suspended. Three months later he won his appeal – but had in the interval gone onto universal credit and was forced to move rental accommodation amid financial pressure. “Income was very tough, there was nothing coming in,” he told The Big Issue. He said he’s still in debt today.

Byrne echoed calls for wider algorithmic transparency and accountability across all industries. “We’ve seen the failings of algorithms in many different facets of society,” Byrne said, from last year’s schools marking fiasco, to electoral targeting, to the gig economy. “That’s something for society to really look at, to tackle, and to make sure people are not unfairly targeted.”

Ogunyemi said the software is racially discriminatory and should be abolished until perfected. “AI is good, but it’s only good when it’s 100 per cent perfect,” he said. “It’s denying a lot of people daily earnings.”

You would never go ahead with an operation by AI, Ogunyemi went on, that had a failure rate of up to 21 per cent. He said by threatening livelihoods, Uber Eats’ algorithm poses a similar threat.

“Our lives are at risk when you continue to use it.”