Is AI making us inefficient?

Amid the debate around generative artificial intelligence’s (AI) impact on the workplace, much of the discussion revolves around productivity. Some argue that AI is making individuals and organizations infinitely more productive. Others say that all it's really doing is allowing firms to lay off a lot of people.

A recent study from the London School of Economics (LSE) adds its voice to the mix. The researchers suggest that a lot of the discussion around AI's impact on the workplace is driven by hype rather than evidence. What's more, it may be distracting us from gaining a better understanding of the kinds of tasks AI will perform.

Tasks matter

For instance, if the AI is primarily deployed to take on largely inefficient tasks, it will simply result in them being performed more efficiently.

"If Al is deployed to undertake unproductive, superfluous tasks, the efficiency benefits will be reduced, even if these tasks are performed more efficiently than a human could, because the said tasks are inefficient to begin with," the researchers explain. "We call this eventuality 'efficient inefficiency’."

In other words, it will make the inefficiency more pronounced rather than make the organization more efficient. It underlines the importance of deploying AI correctly, which is as much an organizational challenge as it is a technical one.

Organizational slack

The researchers highlight that most organizations contain a large amount of slack, with managers creating work that may serve political purposes but does little to drive operational efficiency.

While Cambridge research calls into question David Graeber's notion of "bullshit jobs", the LSE team nonetheless found that there are many tasks performed at work that aren't either useful or necessary. As such, deploying AI on these tasks will simply make matters worse rather than better.

While we may assume that managers would inevitably want to strip out these inefficiencies whenever possible, the LSE research suggests that there may be motives to do anything but.

Even when there is motivation to reduce inefficiency, managers are limited by their ability to both spot it and deal with it. As a result, the researchers believe that many managers will simply deploy AI to replace existing tasks, regardless of whether they are efficient or not.

Reengineering the workplace

It links back to the so-called Solow paradox, which pointed out that although computers were becoming widespread in society, their impact wasn’t reflected in productivity statistics.

Michael Hammer and others famously argued that this was inevitable so long as managers deployed new technology on old processes. The better approach was to re-engineer organizations so that they tapped into the capabilities of digital technology.

The LSE paper chronicles this happening at various times over the past few generations.

For instance, the US State Department would usually print out telegrams using teleprinters, which were torturously slow, resulting in long-delayed decision-making. They attempted to fix this by buying more teleprinters, which helped speed up the printing of messages, except most of them were never actually read. The problem wasn't a technical one, but rather an organizational one.

This issue is becoming all too evident as we're getting glimpses into how generative AI is being deployed. For instance, one study that looked at how developers are using AI co-pilot found that coders were generally writing more code, but there was also an increase in broken code that had to be returned to and patched up. It's an example of what might look like efficiency gains at first glance being anything but.

Intentional change

As with previous technologies, it's not simply a case of deploying the tech and enjoying efficiency gains. Instead, it requires managers to deliberately and intentionally look at the way employees behave, identify inefficiencies that could be eradicated with the help of AI, and re-engineer the way people work.

None of this should be enormously surprising. After all, previous generations showed that it takes time for people to develop the skills required to effectively utilize new technologies.

We've already seen evidence that the most frequent users of generative AI are often those with the least understanding of it, so it's perhaps not too surprising that applications are patchy in terms of efficacy.

The study underlines that just deploying AI to replace certain human tasks isn't what's important. What really matters is that we're replacing tasks that are inefficient with those that are efficient. At the moment, it's quite likely that generative AI is being assessed on its technical capabilities rather than on what the organization actually needs.

As with previous generations of technology, if we don't re-engineer our teams and our organizations, then it's very unlikely that generative AI will achieve anywhere close to the incredibly ambitious goals both technologists and policymakers have for it.