By Catherine Thorbecke

MOST OF US have encountered “AI slop,” the deluge of low-quality content produced by generative artificial intelligence (AI) tools that has inundated the internet. But is this computer-made hogwash taking over our work as well?

News that Deloitte Australia will partially refund the government for a report sprinkled with apparent AI-generated errors has caused a local furor and spurred international headlines. Barbara Pocock, an Australian senator, said in a radio interview that the A$440,000 ($289,000) taxpayer-funded document misquoted a judge and cited non-existent references. The alleged AI mistakes are “the kinds of things that a first-year university student would be in deep trouble for,” she said.

Deloitte Australia didn’t immediately respond to my request for comment, but has said the corrections didn’t impact the report’s substance or recommendations, and told other outlets that: “The matter has been resolved directly with the client.” Besides being a bad look for the Big Four firm at a time when Australians’ trust in government-use of private consulting firms was already fraught, there’s a reason it has struck such a nerve.

It has re-opened a global debate on the current limitations — and high cost — of the technology backfiring in the workplace. It isn’t the first case of AI hallucinations or chatbots making things up, to surface in viral ways. It likely won’t be the last.

The tech industry’s promises that AI will make us all more productive are part of what is propping up their hundreds of billions of dollars in spending. But the jury is still out on how much of a difference it’s actually making in the office. Markets were rattled in August after researchers at the Massachusetts Institute of Technology said that 95% of firms surveyed have not seen returns on investments into generative AI. A separate study from McKinsey found that while nearly eight in 10 companies are using the technology, just as many report “no significant bottom-line impact.”

Some of this can be attributed to growing pains as business leaders work out the kinks in the early days of deploying it in their organizations. Tech companies have responded by putting out their own findings suggesting AI is helping with repetitive office tasks and highlighting its economic value.

But fresh research suggests some of the tension may be due to the proliferation of “workslop.” The Harvard Business Review (HBR) defines this as: “AI generated work content that masquerades as good work, but lacks the substance to meaningfully advance a given task.” It encapsulates the experience of trying to use AI to help with your job, only to find it has created more work for you or your colleagues.

Some 40% of US desk workers have received workslop over the past month, according to a September survey from researchers at BetterUp and the Stanford Social Media Lab. The average time it takes to resolve each incident is two hours, and this phenomenon can cost $9 million annually for a 10,000-person company.

It can also risk eroding trust at the office, something that’s harder to rebuild once it’s gone. Roughly a third of people (34%) who receive workslop notify their teammates or managers, according to HBR, and about the same share (32%) say they’re less likely to want to work with the sender in the future.

There are ways to smooth out this transition. Implementing clear policies is essential. Disclosures of when and how AI was used during workflows can also help restore trust. Managers must make sure that employees are trained in the technology’s limitations, and understand that they are ultimately responsible for the quality of their work regardless of whether they used a machine’s assistance. Blaming AI for mistakes just doesn’t cut it.

But the growing cases of workslop should also be a broader wake-up call. At this nascent stage of the technology, there are serious hindrances to the “intelligence” part of AI. These tools may seem very good at writing because they recognize patterns in language and mimic them in their outputs, but that shouldn’t be equated with a true understanding of materials. In addition, they’re sycophantic — they are designed to engage and please users — even if that means getting important things wrong. As mesmerizing as it can be to see chatbots instantaneously create polished slides or savvy-sounding reports, they aren’t reliable shortcuts. They still require fact-checking and human oversight.

And despite the big assurances that AI will improve productivity, and is thus worth businesses paying big bucks for, people seem to be using it more for lower-stakes tasks. Data suggests that consumers are increasingly turning to these tools outside of the office. A majority of ChatGPT queries (73%) in June were non-work related, according to a study published last month from OpenAI’s own economic research team and a Harvard economist. That’s up from 53% in 2024.

An irony is that all this may end up being good news for some staff at consulting giants like the one caught up in the current backlash. It turns out AI might not be so good at their jobs just yet.

The more workslop piles up in the office, the more valuable human intelligence will become.

BLOOMBERG OPINION