Much of Accounting is deterministic. LLMs are probabilistic. Your job is probably both.

Do you know which elements of the job are deterministic and which are probabilistic?

Whether it’s your own job, the roles of your team members or if you are leading a technology-enabled transformation to change other peoples jobs, it is an important question. Critical, existential even . . .

We often assume we know, which might be the most dangerous mindset to hold . . . 

I wrote an article two and half years ago on the impact of “Dunning Kruger” on the application of AI (or GenAI as we used to call it in the old days 😉). 

The “Dunning Kruger Effect” is a cognitive bias where even otherwise smart people with limited knowledge or experience in a specific domain greatly overestimate their own judgement and decision-making in that domain.

Sound familiar?

We have all observed and even been at the sharp end of implementing decisions from “on high” only to experience the reaction of subject matter experts as a collective “WTF are they thinking of in trying THIS IDEA?” . . .

And sadly, as we ascend the greasy pole of seniority, we find ourselves making exactly the same mistake (not YOU, of course – the others . . . 😉)

The Guardian newspaper just published an article by Ramin Skibba that reported on a survey that finds that while many executives are claiming that AI boosts productivity, workers in their businesses say they’re drowning in ‘workslop’ . . .

It made me wonder whether we are seeing another resurgence of Dunning Kruger in the workplace . . .

“Much of Accounting is deterministic. LLMs are probabilistic. Your job is probably both.”

This dichotomy may be at the root of the misalignment between executives and those that do the real work of business.

Using the proverbial “sledgehammer to crack a nut” can appear highly effective at first sight. However, any expectation of devouring the delicious nut kernel may evaporate on examining the remains!  😉

Applying the wrong tools to the job can appear productive, unless you have to perform the subsequent tasks in the process . . .   Echoes of Denning Kruger again . . . . 

“Workslop” is one of the terms used to describe what happens when we use AI to quickly generate work that seems plausible and polished – at least superficially – but is in fact so flawed or inaccurate that it needs to be heavily corrected, cleaned up or even completely redone after it’s passed on to colleagues.

Been there?

A recent survey of 5,000 white-collar US workers found that 40% of non-managers say AI saves them no time at all at work, while 92% of high-level executives say it makes them more productive.

So what’s causing this workslop deluge? The answer is more complex than being simply a case of workers cutting corners. The real driving force connects back to the C-suite.

Companies have spent billions on enterprise investment in Generative AI. Some of them, like Block, Amazon, Dow, UPS, Pinterest and Target, have laid off employees at the same time, attributing the cuts to AI’s potential productivity. Workers who remain may feel pressured by their employers to use AI to produce more work, often with little guidance or training.

A disconnect separates executives giddy about Generative AI from workers – who are finding that AI sometimes makes their jobs harder and more intense.

A product designer commented “It seems to be common to just copy and paste a bot’s message directly into chats or emails”. At times, when confused by work a colleague sent, they will respond, saying: “Yeah, I’m not sure what AI meant by that” – meaning they’re effectively outsourcing judgment to the chatbot. 

It is a dangerous cycle but surprisingly common in the drive for “productivity”.

“The problem is, Generative AI is often being presented as a general-use tool that can do anything, but the reality doesn’t work that way. So, what could be creating part of the workslop is [AI’s] unclear mandate or use case”.

Much of Accounting is deterministic. LLMs are probabilistic. Your job is probably both.

Food for thought. Despite the massive potential of AI, there is a leadership responsibility to guide how best to apply it in your own business or team. That takes time, experience and thought.

The alternative might kick the can down the road for a while, but will probably end in tears . . . 

We should be setting clearer expectations for the “sweet spot” usage of LLMs or we may face a “productivity bomb” rather than a productivity boost!

You can read the original ageing 😉 article on GenAI, Dunning-Kruger and the “Emperors New Clothes” here . . 

You can read Ramin Skibba’s article in The Guardian “Bosses say AI boosts productivity – workers say they’re drowning in ‘workslop’” here   . . . .

You can also read the detailed report of the 5,000 person survey in the Wall Street Journal “CEOs Say AI Is Making Work More Efficient. Employees Tell a Different Story” here . . . 

If we have a serious interest in the future of work, business performance, the nature of our roles and that of others, this is a topic we need to think through and discuss. 

And all that before we get to Agents!

Thanks for reading . . . 

Much of Accounting is deterministic. LLMs are probabilistic. Your job is probably both.