It's Impossible to Be Average, and AI Can't Help You

(Photo: the “Norma” sculpture — an averaged female body — standing next to a real woman on a pedestal.)

Have you ever heard of Norma?

Okay, let me tell you a story. In 1942, a guy named Dr. Robert L. Dickson took a lot of measurements of female bodies. He averaged out those measurements and asked a sculptor named Abram Belskie to create a sculpture of this “average” woman. Thus, Norma was born. She was average in height, width, length, and size. She was an absolute midpoint in everything.

So, in 1945, a well-known American newspaper, The Plain Dealer, started an interesting contest. They invited women from the U.S. to send in their measurements to see who would match Norma as closely as possible. The winner would receive $100 (a big deal in 1945). A total of 3,863 contestants sent in their measurements. One might think that was a tough fight, since Norma was a depiction of the average woman — so a lot of contestants should have been close to a match.

Well, it turned out that out of 3,863 contestants, none matched perfectly. The closest one was Martha Skidmore. And if you click on the source link and see the pictures, you’ll find out she wasn’t that close at all.

So, out of 3,863 contestants, NONE exactly matched the average. Everyone had their something that kept them from being an absolute average.

Now, AI in general — and different chat models (like ChatGPT) specifically — are designed to produce the most average output they can come up with. GPT stands for Generative Pre-trained Transformer (bet most of you never cleared that one).

The idea of ChatGPT is a complicated one if you look at it from the viewpoint of a mathematician or a programmer. But we don’t have to be complicated snobs. We can get a simple, workable explanation.

Generative means that ChatGPT can generate a text output word by word.

Transformer means that it generates the next word based on whatever it has already generated — thus constantly transforming its input so its output is more coherent.

Pre-trained means that programmers fed a lot of textual data into the machinery of ChatGPT, so it generates its output based on millions of texts found on the internet.

Got it? If not, here’s a much simpler explanation: ChatGPT is a glorified autocomplete on your cellphone. If you type in “Hello how are,” it’s most likely going to say “you?” as its output.

ChatGPT is going to give you the most average answer from everything that has been said on the internet. It can’t have an opinion. It does not understand what is being said. It can’t acknowledge anything. It doesn’t see anything but a number that tells it what word is the most plausible in a given set of words.

Any other “better” version of ChatGPT, “thinking” models, or “more advanced” models are just that — glorified autocompletes. And they produce the most average answer one can get.

But guess what?

We know that humans can’t be average. The sizes of our waists are not average. Our behavior is not average. Our thoughts and reactions are not average. It’s against being human.

So what do I observe right now?

AI in health. A terrible abomination that assumes “good health” is something average. You can’t find two people who have “good health” and have similar measurements.

My wife loves buckwheat. She eats it every day. I look at her as if she’s an alien every time she consumes that hell-born resemblance of food. I can’t live without any scorpion sauce. My wife won’t even put black pepper in her mouth. When I eat something, I’ll run for about 6–8 hours on one portion of food. She can’t last 4.

Every diet must be tailored to a specific human being. Keto, Paleo, Dunkin’ Donuts diet, Carnivore diet — all of these are different every time. Some people will benefit; others will suffer.

One has to go and explore every single diet to find the one that fits.

So what about that “breakthrough in AI healthcare”? Well, that’s for the birds. A big pile of crap, to say it straight. You just can’t average out the diet between different people, much less everything else beyond diet.

Gosh, I get heartburn just looking at sourdough bread. Yet I don’t give a damn about any electrical waves or turbulence around my body. I can be totally comfortable in a power plant near 10-kilovolt wires and sparkling electricity. Yet I see people refusing to install a microwave in their apartment. I get them. We’re just totally different.

What’s next? AI for mental care? Oh for gawd’s sake! Don’t give me that bollocks. If you agree with the concept that a computer can handle someone’s mental problems, I can definitely say you’ve never attempted to help a single human being in your life.

A computer, by definition, is a machine that can collect, store, process, transfer, and produce information on demand. It is an extremely capable device that can augment human thinking in terms of mathematics and data storage. ANY computer fits this definition. This applies to your laptop, watch, iPhone — or ChatGPT. And the funny thing is: the computer cannot exceed the boundaries of this definition.

And when someone tells me the opposite, I can sure point out a rather blind person in the room.

So, don’t rely on marketing-bubble statements that your company needs AI. If someone tells you that your work can be replaced by AI, he most likely doesn’t know what your work is. If you believe your work can be replaced by AI, then either you don’t know what your work is — or you know you’ve been doing no work at all this entire time.

Stop praying to something you don’t understand. Start learning it and figuring out how to use it in your work.

And if you are unable to figure out whether you should or should not use AI in your business — contact me. I’ve been doing software development for over 25 years and working with AI for over 10. I know what it can or cannot do. I can get those guys trying to nibble at your budgets off your lines with a few simple questions about their AI products.