The Centaur of the Centaur

Automating your processes?

Hallucinating AI is a terrible co-pilot. It is an "automation blindness" that lead to screening officers meticulously spotting every harmless shampoo bottle on X-rays, while missing nearly every gun and bomb smuggled through their checkpoints.

My personal experience in automation is linked to driving an electric car - Tesla. Let me debunk the general perception of EV limitations 1. lack of infrastructure i.e. not enough charging stations, and 2. limited charging capacity i.e. not enough capacity to travel longer distances without needing to recharge.

As a Tesla driver, I find some common perceptions about electric vehicle (EV) charging to be misguided. Charging an EV is not drastically different from refueling a traditional car at a petrol station. It simply requires a shift in habits, visiting a charging station instead of a fuel pump. Many BP stations now include EV charging facilities, making the transition even easier.

Modern EVs, including Tesla models, boast impressive charging capacities. With ranges of 650 kilometers or more, these vehicles can easily handle extensive usage, such as taxi services, from dawn to dusk without issue. The convenience and practicality of EVs continue to improve, challenging outdated misconceptions about their limitations.

Limitations I came across with Tesla were largely related to its AI. The AI is well equipped to perform 360C scan of the surroundings of the motor vehicle. The visual in the vehicle is nothing short of fascinating. It is sophisticated enough to register shades; left alone the trucks the shades belong to. But it is not smart enough to distinguish the shade from the truck in the adjacent lane from the shade of the bridge ahead. As the shade gets darker in its RGB pallet Tesla does one of the two things: it slams on its brakes or slows down well below the speed limit. Neither of these actions are safe to perform on a highway of 80km/h. These are bound to create a chain reaction should an accident occur.

Similarly, the AI is sharp enough to register green lights, but not smart enough to consistently register red. It is naive enough to register glowing red and orange lights on the ambulance driving past and profoundly stupid to slam on the brakes right in the middle of moving traffic in response.

In short, the AI is the reason why the EV is not the safest motor vehicle to drive. Bear in mind this AI was the key safety feature in promoting Tesla to drivers.

Now getting back to the automation in business settings. Today's AI resembles having a chatbot that summarizes the emails to the boss in 2025 equivalent to the 2000s gag when the boss whose secretary printed out the boss’s emails and put them in his in-tray so he could go over them with a red pen and then dictate his reply. Seems convenient, smart, cost-efficient and almost productive.

Now on the productive note, let's get back to the screening example above.

It's good enough to get the job done much of the time. Not all the time. AI can also become the defective booby-trap code that is statistically guaranteed to look as plausible as the good code. AI would inevitably lead to human intervention, notwithstanding the irony of the reasons of adopting of AI was to remove costly human touch. AI was the "Centaur"! Though we now realize we need a "Reverse-Centaur", a human to decode the code to fix the bug.

This human is otherwise referred to a niche technical specialist if outsourced. This human costs more than combined payroll impost of those the AI replaced initially. In this scenario, "the-human-robot-nurse" is a "The Centaur of the Centaur". Because everything hangs on his ability to debug the very AI and its functions the business has become so reliant on.

Or this human is otherwise referred as "employee-in-the-loop" or "monkey-in-the-middle" or "moral-crumple-zone" who is to blame for all misdemeanors of the faulty script provided by an external tech company at the point of pitching the AI to the business. In this scenario it is most likely the monkey in the middle will have to engage "The Centaur of the Centaur" because the tech company would see no benefit in leaving the script or parts of it unprotected and robbing itself of the opportunity from selling "debugging" and "further training" services.

Figurative cost comparison here looks like this:

1. paying humans = 100k;
2. paying tech company = 120k

Net loss to the company, excluding the emotional agony of fixing the same thing again x3 times before is (20k).

The cost could be a lot higher than that though if we look at this scenarios:

Next (UK Retailer) fined millions of dollars by an equivalent of Fair Work in Australia for consistently underpaying employees because their software update failed to reflect correct awards in the system;

Tesco (UK Supermarket) ends up underpaying 140,000 workers due to inherent bugs in the payroll systems. The company was fined and ordered to pay over 10 million pounds to employees.

Bear in mind neither company went about robbing their staff intentionally. Nor were they given any discretion by government agencies trumpeting their horns in media portraying them as thieves. These companies thanklessly continue to employ thousands of people and provide for their livelihoods. This of course outside the subject line.

What is a typical business AI made up of? AI predicts the future by looking at the past data, and when that prediction becomes an automated decision, AI becomes a torrent of machine judgments based on human bias obtained a point in time.

Still want to replace boring humans with spicy autocomplete?

Tell me why in the comments section.

Your thoughts on the matter?

Truly Yours,

Narghiza Ergashova