Replacing Workers with AI Is a Big Mistake

I run an Artificial Intelligence (AI) startup and work with this technology regularly. It is in my best interest that more people start using this technology. Yet, here I am saying that replacing workers with AI is a big mistake. Why is this?

Well, over the past few months, I have been in discussions with organisations that show up with one agenda: “How can we use this technology to replace our (unproductive) workforce?” Well, they do not exactly say that word-for-word. But when you look past the diplomatic language and niceties, this is the essence.

Even though different organisations operate with different cultures, these discussions are remarkably similar. Competent executives with assertive demeanours show up with a surface-level understanding of how this tech works. When someone tries to explain the finer technicalities, the executives are usually quick to shut them off. “No science research here, please!” or variants of it are typical answers to hush any technical discussions.

“What else are they interested in?”, you ask? Well, they are interested in the fastest, cheapest way for them to implement this tech and replace their (unproductive) workforce. And one can literally sense the FOMO (Fear of Missing Out) fill the room. I often hear variants of the following sentence:

“We need to act fast before our competition races ahead!”

The more I look at this sentence, the more I feel that this is a race towards catastrophe. Having said that, the organisations that I have been talking with are not the worst of the lot. That honour is reserved for Quick Draw McGraw!

The Quick Draw McGraw

Quick Draw McGraw is a cartoon featuring a horse-sheriff that I enjoyed as a child (the merry days of Hanna-Barbera). In this context, I am using him as a metaphor for quick-acting organisations which aim to capitalise on the “AI Revolution”.

You see, the organisations that I have been talking with are still “considering” changes. On the contrary, the typical Quick Draw McGraw has already fired — sorry — laid off a significant part of his task force, and is forcing his remaining subordinates to somehow make this “AI-thing” dance — sorry — work.

I presume that this is the competition everyone is worried about. Now, to be fair, it could be that some of these organisations were genuinely quick to build and test pilot projects before they pulled the plug on their task force. But still, this, in my humble opinion, is a big mistake.

To begin understanding why I say this, we need to touch upon the nature of the models that are heading this revolution.

The Hamster that Powers the Wheel

I have no plans of boring you with technicalities. But I think anybody who has anything to do with AI needs to understand that these models work based on training datasets. Think about it like teaching a child how to do something by showing the correct way first.

This “showing” what is correct is largely done by the datasets of inputs (known as features in the biz) and correspondingly correct outputs (known as targets in the biz). Now, let me ask a simple question here.

Where do these datasets come from?

In the vast majority of the circumstances, these datasets come from human experts who currently perform that task; the very experts that the aforementioned organisations are trying to replace. This brings me to my next question:

If you fire your experts and replace them with AI, who has the expertise to check if the AI is AI-ing correctly?

Well, I guess that is a problem for future Quick Draw McGraw to solve.

At this point, I feel that I am straw-manning Quick Draw a little here. So, let me try to be fair and talk about things from his perspective as well.

---

The Goose that Lays the Golden Egg (and the Raccoon that Steals it)

Over the past few centuries, technological innovation has been rapid. A key factor that has been driving this innovation has been humanity’s collective interest in making money. After capitalism gained popularity, we have not looked back.

Any organisation constantly seeks better and more efficient ways of doing things. This is what happened during the Industrial Revolution and the Information Revolution (internet). One could argue that this is what is currently happening with the so-called AI Revolution.

I would say that this so-called AI Revolution has some ways to go before we compare it with the likes of the Industrial Revolution. But leaving that point aside, one could argue that Quick Draw is simply doing what he has always done best: optimise for efficiency and profits.

Even when industrial automation was a thing, people feared that the “machines” would replace them. Well, they did. But then, those human beings went on to find better things to do. Again, this is what we have always done: we off-load and automate stuff that we find repetitive and boring so that we may focus on the fun stuff.

But then, GenAI happened — the sort of AI that is currently automating (conventionally) creative stuff like painting, photography, video editing, etc. One could argue that while these activities appear fun on the surface, they are still ultimately boring tasks. Don’t get me wrong. You could always find a chap who enjoys doing something most don’t.

But when we look at the above professions as jobs, some of them were already well on their way out (mediocre paintings mass-painted by trained “painters” come to mind). I have done video-editing and I did not enjoy it a lot. When it comes to photography though, the birthday photographers and wedding photographers still seem to be safe since these skills cannot be automated that easily. Nonetheless, it is just a matter of time before they get automated away by AI-controlled drones.

Why am I so sure about this? Well, like I said, this is what Quick-Draw has always done! Now, knowing this, can we be mad at Quick-Draw? Well, I certainly cannot.

However, there are some major issues here. Recall that these AI models are trained based on datasets. These datasets carry features/inputs (say, data pre-video-edit) and targets/outputs (say, data post-video-edit done by an expert).

Question: Who did the video editing in the dataset?

Answer: Some video-editing expert!

Unlike conventionally automatable tasks, creative tasks do not have a clear definition of what is “correct”. This means that while GenAI models learn some arbitrarily correct way of video editing, the benchmark for video editing will still be pushed by field experts. Well, then, Quick Draw might say:

“Ha! All I need to do is just keep training my models on the current state-of-the-art video-editing techniques.”

Last time I checked, experts were not idiots. Once people start catching wind that their work is being used to train AI unfairly, they would simply stop supplying. Consequently, all the Quick Draws would be caught off-guard with yesterday’s trend/state-of-the-art, while all the true experts find better things to do with their lives.

This essay is supported by Generatebg

Besides this point, regulation in the field of AI is/will be a thing too.

At this point, I must admit that I am over-simplifying a lot here. My aim here is to not be thorough, but rather touch upon the reasons why I think this behaviour is a big mistake. But then, it does not help much if I call something a mistake and do not mention a remedy, does it?

What Should the Horse-Sheriff Do?

First things first, value your people. Do not treat them like numbers on a worksheet.

When I write this sentence, dear reader, I am doing my best to visualise YOU! That is right. I am talking to you; I do not think of you as “reader number 5352” who will be reading this essay in the future.

Secondly, I am not saying that AI or GenAI is a bad thing; it is a promising technology. Use the technology to help your people out and make their lives easier. Introduce the technology to them and let them “play” with it. They will figure out what to do with it.

Do not leave them out of the loop and develop a solution to replace them. Even if you succeed in replacing them, you will be left with fleeting success.

Sometime down the line, you will be stuck with powerful AI-models that deliver mediocre results, while the real competition benefits from your former employees.

---

If you’d like to get notified when interesting content gets published here, consider subscribing.

Further reading that might interest you: 

• How To Bear Responsibility For Artificial Intelligence?
• Science Fiction Has Lost Its Way
• How To Read The Future Of Large Language Models?