How AI-generated "Workslop" makes us less productive

AI was supposed to boost workplace productivity, but a new Stanford study reveals it's often having the opposite effect. "Workslop"– low-quality AI-generated content passed between colleagues – is costing organizations millions while damaging relationships and trust. The culprit isn't the technology itself, but how we're choosing to use it.

In a recent article, I looked at USC research into the most effective uses of AI in education. It boiled down to whether students used AI to help them think or whether they used it to replace their own thoughts.

A recent report from the Stanford Social Media Lab found that similar issues are at play in the workplace, with employees often using AI to take shortcuts that ultimately lead to more work down the road due to the poor quality of what's produced.

Minimal viable effort

The analysis found that people would often use AI to produce work that was passable and that required the least amount of effort possible. In a nod to the "AI slop" that is cluttering up social media feeds, the researchers refer to it as "workslop."

The study shows that as the capabilities of AI have grown, there has been a division of applications broadly similar to that found by the USC researchers. Some will use AI as a learning support tool that helps them make their own work better. Others will take the easier approach and get AI to do the work for them.

Unfortunately, despite the marketing hype from the vendors, the outputs of such shortcuts are often littered with problems. We're not just talking about the oft-cited hallucinations either, as the analysis found work that was unhelpful and incomplete, often because context around the project was missing.

Minimal gains

The researchers believe this could be a big reason why so few organizations are seeing any kind of productivity gains from AI. These "lazy" applications are simply shifting the workload downstream, as someone else has to tidy it up and make sense of it.

The study found that 40% of employees had been on the receiving end of workslop in the last month. Indeed, these poor souls went on to say that over 15% of the content they received probably qualified as AI-generated muck.

Most of the slop was sent between colleagues at the same level, but people were far from unwilling to send it to their manager, either. While workslop was common across industries, it was especially so in professional services and tech.

Obviously, concerns around our use of technology to do our thinking for us have been around for many years, but the new generation of AI tools makes matters worse because we're spreading the cognitive offloading to someone else.

In other words, whereas using a GPS map or a search engine may dull our own senses, it's largely confined to us, whereas when we use ChatGPT to produce something to pass on to someone else, we're forcing them to do more work to make sense of the muck we're sending them.

Tidying costs

This matters, as while the mental costs associated with producing the AI-generated content are minimal, the study found that tidying up what was produced took recipients an average of just under two hours, or approximately $186 per incident.

When extrapolated across the typical organization in the study, this worked out to a productivity hit of over $9 million per year.

The costs were not just financial either, as the sharing of workslop also carried social and emotional costs. For instance, respondents reported feeling really unsure about how to respond when a colleague sent them AI-generated work, with nearly all saying they felt annoyed and even offended by it.

This carried inevitable social costs to the relationship, as people who sent workslop were subsequently deemed less capable and reliable. Indeed, around half of the respondents went as far as to view senders as less trustworthy. This should come as no surprise, as previous research has shown that we view users of AI as both less competent and creative.

Building guidelines

Various studies have shown that many employees today are using AI in secret, which perhaps plays a role in their inappropriate usage. In academia, the USC researchers advocate the creation of guidelines and training in how to use AI not just effectively but ethically.

Part of such guidelines should be a more discerning approach to the technology. The Stanford researchers argue that rather than suggesting AI should be used everywhere, all the time, we should be far more selective about where and how we use it.

Culture is also a factor, with the study showing that employees who believe they have agency and optimism were not only more likely to use AI but were also more likely to use it in a way that augments rather than replaces their intelligence.

Given the huge investment in AI, it's inevitable that there will be a period of figuring out what works and what doesn't. A comprehensive study of AI adoption in China found that deploying AI in order to cut costs is almost always a bad idea. The best results come when it's used to augment what we do rather than replace it. It's not clear that this message has quite reached workforces yet, but hopefully it will.

In the end, the problem isn’t the technology, it’s how we choose to use it. AI can sharpen our thinking or it can churn out workslop, but it can’t do both. Until workplaces stop treating it as a cheap shortcut and start treating it as a tool for real augmentation, those promised productivity gains will stay stubbornly out of reach.

Unlock more exclusive Cybernews content on YouTube.