AI and the Art of Repeating the Same Mistakes

May 12, 2025

Thursday I was at a world-famous café, enjoying a tea and a sliver of cake with an overly expressive work client. This particular café didn’t have soft sofas, but was filled with hard steel chairs and tables. Still, there were groups of people on every side of us studiously typing away on their laptops. I happened to glance at the closest woman, who had a web browser open to a minimalist AI input bar. I watched as she carefully typed ‘What is the website for [company name]’. I thought to myself, “Oh no! We are doing it again!”

The 'Ground-Hog-Day' cycle I am referring to is the treating of new technologies as if they were the tools that preceded them. Certainly, we should expect the general public needs a period of transition to fully understand what AI changes; But, those of us in the know have an obligation to them to help them along to understand what those changes are going to be.

If we relegate them to getting their information from government and corporate mouth pieces, who see AI, like they do everything, from the perspective of its opportunity to control. We will be repeating, most importantly, the criminal appropriation and transmutation of the internet when it was first sprung on the general population.

When they dug up my street in the 90’s to lay the first internet cables, they promised the internet would level the playing field, giving everyone a voice, creating a perfect democratic society. This changed quickly. ISP's became the entry point for most users. Control of those nodes, and thus the network itself, switched from academic to a commercialized one. Those who I knew who had their own node had it priced out of their range to reasonably afford it. Peer-to-peer networks became ostracized. Cloud computing arrived, which wrestled the very machine you stored and processed data on, into the hands of others. On social media, dissenting voices were shadow banned so they could only preach to their own choir. We now know that omputer processors were installed with the Intel Management Engine and AMD’s Platform Security Processor to grant unwitting access to the computer in your own home to others as long as it was plugged in.

Government is pretending it is still deciding how to treat AI technology. In one hand they are investing in data-centers, but at the same time say they want to protect people from ‘AI slop’, which is a euphemism for they don’t want the public to have powerful AI that could be used against those with power. They will shackle AI, just as they have everything else, enacting regulations whose end is to restrict regular people from possessing powerful AI Models. Anything else they promise is theatre that they will say we all mis-understood.

Today's first lesson is that you should become a curator of your own personal AI models.

Where was I?; Another mistake we are repeating is using technology as though it was the devices that they replace. One of my first jobs in computers was for a large multi-national sales organization. Our staff all had computers equipped with the latest MS Office suite with word processor, spreadsheet, that cursed Power Point, email, and some had Access, the database making and management program.

A department head would schedule a meeting with his staff, a report on the subject of the meeting was generated, and often hand crafted. When the report was complete, it was printed out, one copy to verify the layout was attractive, another for the first revision, another for the final draft, and then finally one to every person in the team. The stack of reports were walked up to reception, where everyone’s mailbox was. By mailbox, I mean a physical plastic shelf with the recipients name on a printed Avery label. Of course, those boxes were full and you had to pick up three other peoples mailbox that rested on top of your target, and then place the report on top of the teetering paper stack, and then gingerly return the rest of the shelves on top of the leaning tower. In a few days, at the group meeting, errors were found. And the report would be printed once again to provide a record of the meeting.

Why wasn’t the report just emailed to everyone with a call for revisions and input? There was probably no real need for a meeting. Well maybe other than to verify that everyone had actually reviewed the information.

Other things the staff would do is:

• Enter each spreadsheet cell data manually. They didn’t use formulas.
• Price lists in the master database were updated manually.
• They never did digitize their filing system while I was there. Maybe its for the best I wasn’t there.
• The Canadian division’s website was basically just a picture of our offices with a dot ca domain name. I don’t remember it being used for anything productive. We did get the odd sales lead from it though, all of them asking for things so obscure the writer was too embarrassed to ask for in person or even on the phone. The web hosting was part of the package for giving our sales staff dial-up internet access so maybe I shouldn’t complain too much.

Hopefully none of you or your staff are still doing these things, but please don't make the same mistakes with AI. AI is still in its early days, what I write may change, but this is what it can do at the present time. AI can do more than search the internet. It can assemble its searches into a semi-coherent package. Offering options for you to choose a way forward. Do not allow it to choose, and if ever it recommends something, always ask if there are other options. Its decision making abilities are nowhere near as amazing as yours are. Even if you make bad choices, you probably have a reason you can use as an excuse. My opinion is that right now, AI just picks options to pursue randomly. If it recommends an option, its because it is the first it found, and stopped looking for alternative options.

Just to give an example, when I had a higher version of a binary than a new package I was installing was expecting, AI told me to create symbolic links to the binary with older version names as a link name. To make things worse, a binary of the older version was available. I could have both on the system without problems.

AI is a master at html, css, and javascript. You cannot ask them to generate a full page, if you do, you will have to significantly correct it. But it can handle creating functions very well. As the operator you will have to bring the vision and the plan, as with decision making its is not very creative and while can generate a plan if it knows the goal, asking it to make a series of goals is futile.

Not all AI are created equal. The one they have bundled in with whatsapp is good for nothing more than summarizing news articles, and asking questions you could use a search engine for. It does not have a grasp of most things physical, often talks in generalities, or continually asks for more information.

Here are some exercises to show you how AI can speed up your activities.

• Summarize a news article, get it to tell you so a child will understand, or in a different language.
• Ask it to write a more concise version of some of your writing, tell it to make it funnier, or sound like it was written by a gold prospector.
• Invent a character for a play. Tell the AI the theme of your play, what you want the person’s role to be. Then ask it to brainstorm ideas about the characters name, background, family life, work history, style and behaviors so it fits the theme of the work and the characters role in the performance.
• Decide to write an essay you know something about. Plan what you want to say. Ask the AI to tell you the contrary opinion to your thesis.
• Ask the AI for the operations manual, or code of conduct, or mission statement for a group of time traveling deer that steal cell phones and leave them in the past to be found by archaeologists.
• Ask it to teach you C pointers.
• Tell it you are certain you lived a past life, could it help you remember who you were.

Unless you are Stack Overflow, you have absolutely nothing to fear from AI. It is a tool to do things at a faster speed. Sometimes the gained time is that time you would waste reading through the very question you need answered, only to find out that the admin has judged that the questioner really wants to know something else, or should not do what they are asking, or is not phrasing the question in a way that pleases the admins, or some other non-answer, or wrong answer. Using AI frees up your research time to be more productive.

Zaarin -- Founder n·h·g