I’ve been writing this post for almost a year now, on and off, so some points here may be out of date.

AI, the biggest buzz in tech right now, continues to proliferate everywhere in my day to day life. Despite that, I still don’t use it, and I have zero intentions to ever pick it up and embrace it.

From environmental impacts to ethical concerns, the amount of reasons why I continue to actively avoid using AI continue to grow. In this blog post, I’m going to talk about my main reasons I remain highly skeptical of AI and what it will do to “rapidly evolve society”. Some of these claims might be debunked. I’m willing to accept that and consider alternatives here. But as it stands now, the amount of cons associated with AI continue to stack higher than the pros.

Disclaimer: I do not write AI software, or work on any code that interacts with AI. All of these thoughts are opinions of my own.

Ethical concerns

Truthiness

The land of misinformation

Case study: The Nintendo Switch 2 game price crisis

Note: The example about Switch 2 physical game prices are only relevant in the US. Physical copies in the EU do cost 10 more than a digital copy.

When it comes to AI, truth seems to be defined by what is most commonly agreed on by the mass, even if it’s not actually true. A great example of this I saw earlier this year was with the Nintendo Switch 2 announcement. A big rumor began that physical game copies were going to cost $90. This threw folks on the internet into a storm of anger, because how could a physical game ever cost more?

This rumor held very strong until information about Mario Kart World was shared. In reality, the physical copies did not cost $90 USD. Users on Reddit began noting that game copies instead only costed $79.

However, the damage had already been done. Around this time, and even before an official announcement was made, if you asked Google “How much are Switch 2 physical game prices”, you’d be met with a very confident AI Overview sharing that they would cost $90 USD.

The problem here is two-fold.

Firstly, by Google’s AI Overview regurgitating nonsense it heard online, now, anyone who’s curious about the rumor will believe it without any true evidence to back it up. In terms of what the AI believes is the truth is what the majority is saying. Even though the majority were wrong, lead by the false rumor. Because of this, the rumor has now spread even more. Now people who were simply inquiring about this have an “AI-certified” answer that does all of the hard work for them. Which in this case, was helping build their belief in the rumor. How does the rumor go away? Manual intervention - that being individual users reporting the AI information as incorrect until it’s retrained to know what’s actually true. But the damage continues to accumulate until that happens.

The second fold of this problem is by how much it impacts the one affected by the incorrect information. News flash, there are still people out there that believe Switch 2 physical game copies cost $90. Ever since they read that the games were going to be $90, from the AI, they’ve now held a strong negative bias against Nintendo as being a sell-out company. All of this happened, when Nintendo, in no way ever made the statement it would cost this much. How many people are now avoiding Nintendo because they don’t believe their company is worth their dollar?

How do we solve this? We don’t. Big tech trains their models across the internet, mostly without permission (so much for robots.txt). Because of this, misinformation will automatically proliferate in AI. Big tech has no incentive to slow the train down. What happens when something that isn’t the truth becomes the truth? That’s when our society begins to devolve. Evidence, reasoning, nuance, and context are all thrown out the door to artificially make us “more productive”. AI makes assumptions for us that we aren’t realizing we’re making when we accept it’s answers. Arguments against this will say “You should be questioning what you see on the internet more”. While I tend to agree with this statement, asking that of the common person isn’t much of a possibility. Humans are incentivized to always choose the shortest path to a solution, even if it means it’s the incorrect one. This acts as a great segue to the next section.

Learning

How do you learn?

How does a human learn something? Ask yourself that question and think about what it took for you to learn a new skill, a new hobby, a new trade.

When I think about what it took me to learn how to write code and software, I think back to when I first began. When I was in high school, I started with Python, very briefly. I wanted to help one of my teachers automate taking attendance. I had no idea where to start. I went to a very small school where coding was not taught. So I had to figure out how to teach myself what it meant to code. I would occasionally watch those YouTube series teaching you “Everything you need to know about [some language]”, but things weren’t quite clicking. I spent days trying to fix one line of code. I was incredibly inefficient at the time. At the same time, I was learning. I was beginning to crawl. I was very slow, doing many things wrong, but I can say I was still learning. Learning is less about reaching the end goal and more about the process in between - the headaches, the hours spent wasting time on something seemingly easy in hindsight, you name it. Learning is everything between the start and the end.

I eventually tried to learn Objective-C and Swift but could never quite wrap my head around object-oriented programming at the time. The end goal was to turn on and off Bluetooth on my school-provided MacBook. After nearly a month of trying anything I could, without knowing what to try, I eventually got it. The learning was complete, though I might not have actually known why it worked, at the very least, I was teaching myself how to look for different ways to solve problems.

Learning isn’t easy

So why do I bring all of this up? My concern with AI is that when we don’t incentivize learning in the tool itself, it will never be used for that purpose. This especially concerns me with new developers, that are just beginning to learn how to write software.

Tell me which is easier: Copying/pasting your code into ChatGPT and asking “Why doesn’t this work?”, or thoroughly looking at the code, getting a deep understanding of it, and resolving the problem yourself. This is rhetorical of course, but the main point I want to drive here is that easiest does not mean always mean the best course of action.

The best analogy I can put around this is the current obesity health crisis in the United States.

Note: I draw this comparison from one of Hank Green’s videos on YouTube where he outlines the idea of constantly starving for information.

To sum up Hank Green’s words, now, more than ever, we have more access to food than was ever conceivably possible. One hundred years ago, we were amidst the Great Depression. Bread lines and food scarcity was common. Now, we have more food than we know what to do with. Let’s think about a distinct part of the human condition: survival. One aspect of survival is food. We need food to survive. Well, in this day in age, we now have more food than ever. In no way is that a negative thing at all, in fact, life expectancy is much higher now than it ever was due to this. In the most logical scenario, if a human wants to maximize the most calories per dollar, and effectively feed themselves the most food for the least dollar, this can be done with relatively low effort. Go to the snack food aisle, grab cheap, high calorie foods, and purchase them. Just like that, you have your daily calorie intake covered. That’s great and all, but what happens when we go to the logical extreme - over-consuming high calorie food for as cheap as possible. That leads us to the American obesity problem we have today. We over-consume in cheap, low quality, high calorie foods - which, if our ancestors had known we were able to do, would have bewildered them. Today, we’re aware that this is a problem, that we, as individuals, are trying to solve for ourselves. We’re trying to maintain a balanced nutrition and healthy eating habits. It’s surprising, I know, that eating the most calories for the cheapest dollar isn’t what we want to do today. But it goes to show that the easiest solution for calorie intake is not the correct one. The new problem is hidden within the problem. By giving ourselves everything we need, and even more, we shift the goalpost and have to limit ourselves. As we all know, healthier food options tend to have less calories and are generally more expensive. When compared to the snack food aisle, it’s clear that this is the more difficult solution to food - being picky about what to consume, making more rational decisions about the best value food, while still maintaining health.

Learning is honest

How does this compare to AI? Well, AI, in my opinion, is always pointing you towards the easiest solution.

There’s a great clip from the satirical TV show Silicon Valley. One of the team’s developers builds an AI to help fix bugs in the code. After some time, the team realizes the entire codebase has been deleted. The developer of the AI hypothesizes that the AI might have deleted the entire codebase to delete all bugs. While I can only hope this hasn’t happened yet, there’s a non-zero chance this could happen.

All of this circles back to the original issue I’m bringing up - that we’re skipping the hard part - which is learning. AI does not try to lead you to learning unless you, as a user, explicitly ask for more context. The user itself has to request it. As I mentioned before about the Switch 2 games, a lot of people do not try to learn, or more pessimistically put, trust too much of what they first read online. I don’t think it’s our own fault though, or that we have much choice. Social media/big corporations tends to favor taking things at face value. Anything to gain angry clicks these days. If you ask anyone if they want to continue to learn through their life, you will always get a yes. The issue is that the environment isn’t conductive for asking questions or learning - it’s designed to keep you on it for as long as possible, like social media, to maximize profits. If the system were designed to help you learn, it wouldn’t give you the direct answer - rather, it would lead you to the answer you have to find yourself.

If AI could teach you (it could, it just doesn’t)

If teaching you things were profitable, AI wouldn’t give you the answer. It would teach you how to evaluate the problem, use reasoning, and come to an answer.

In a very simple example you can think of it like this: You ask AI “What’s 5+5?”.

Instead of it giving you the answer like a calculator, what if it responded “Well, what is something that there is five of?” You look down and see five fingers on your hand. You tell it, “I’ve got five fingers on one of my hands.” It responds “Cool, and how do you add?” You think about it, and you realize that you can count the number of fingers across both hands to get your answer.

While this example isn’t one where someone would be beating their head against a wall, the idea is generally the same: you work towards the solution so you know how to solve it in the future. AI does generally give you a breakdown in some situations, part of what I also fear is that users are going to ignore it because it’s all about getting the easiest, quickest answer. Our path of least resistance is growing shorter, but at what cost?

Profitability

AI, today, remains unprofitable. It’s a big gamble based around a ton of hype that we’re going to suddenly have another “breakthrough” that does even better than this first round.

Enshittification

Obviously it’s really really bad for us

This section was inevitable. The platforms we’ve came to “love” and “appreciate” are of course the same ones building these wonderful AI tools. Totally good for us, right? Right???

I recently discovered this talk. If you don’t want to listen to the whole thing, the big part I want to share is from 2:51-3:38. Isn’t it disgusting how companies have became so engrossed in creating higher amounts of revenue that they’d rather diminish their own customer’s experience without them ever knowing?

I realize that Google’s internal decision to enshittify it’s search was made before the large proliferation of AI as we know it. If this was a decision they were making before AI blew up to what it is today, what in the world are their decisions looking like today? Is it not safe to assume that they’re going to be making search even worse to incentivize using their latest and greatest zodiac sign? Don’t even get me started on the fact that they’re about to about to start pushing ads into these platforms. I hope you’re excited to begin paying for your monthly subscription to ads.

Enshittificaton is good, actually (apparently)

I found this case arguing for enshittification. On the surface, they agree that enshittification is bad. They, however, argue that by having it, we consumers realize that we need better alternatives. That by us even realizing enshittification is happening, we can ultimately choose what platforms to use.

Here’s the thing though about these platforms. They’re fighting for your attention for as long as possible. The longer you sit and watch your short form videos and get mid-roll ads, the more money they make. The more you help confirm their bias that their AI is good by simply using it, the more money they make.

That’s obvious, so let’s just drop these platforms.

Unfortunately as we all know it’s never that easy. If we reflect back to my previous section where I make comparisons to the obesity problem, anyone who’s ever tried to exit from social media platforms realize the struggle. You’re constantly being hinted that you’re missing out, that you’re anti-social and that you must be on these platforms to simply exist in the real world. Note: This couldn’t be further from the truth. It doesn’t help that even when we’re told that we’re being spied on, we still use the platforms, despite that. Keep in mind, this happened over 12 years ago.

Countless studies have been and are being done to expose how social media harms us. I’m a young adult, so my entire high school career involved social media. At the time, we didn’t have enough information to suggest social media negatively impacted youth. We know more about social media and how it affects us (and how companies are using it against us) than we knew even 5 years ago. Each day, we tend to learn a new ulterior motive behind each one of our favorite doom-scrolling centers. Despite having great evidence to suggest that we should absolutely never use them, we continue to use them.

The point I’m making

Unfortunately, I don’t think there’s ever a scenario where the common people have an uprising against ‘Big algorithm’ and take back their full, undivided attention. I think individuals are fully capable of it. But unfortunately, just like the obesity problem, the number of users will continue to grow as platforms keep finding ways to keep us hooked. This isn’t just social media, but AI is being designed to hook us as well (fortunately some people aren’t completely mindless to these awful attempts).

Another hooking attempt: Microsoft wants you addicted to their AI

Long tangent about social media aside, you might see how this parallels the AI industry. Sign people up for something cheap, free, quick, and seemingly perfect, then put your users back into reality where you profit off of them as much as possible. If we, as humans, weren’t as easily tricked as we are, we probably could jump ship from these platforms in large scale numbers. Including AI ones.

I don’t see us getting off this ship until our government stops being paid by big tech to simply exist.

AI doesn’t make you think. There’s studies out there suggesting that. The less you think, the more money they make. George Carlin must be rolling in his grave.

While writing this post, Cory Doctorow published his talk about artificial intelligence and his takes on it. Since reading it, I’ve been recommending it to friends and family, so that also means I’m recommending it to you as well: Pluralistic: The Reverse-Centaur’s Guide to Criticizing AI. One of my favorite pieces of this was where Cory talks about AI art image generation always leading to a feeling of something being “missing”. I think the summary finally provides finally starts making the water a little less muddy.

Environment

AI “data” centers

You mean AI slop machines, right?

I’m not going to touch on this topic all that much since common discourse is fairly aware of the issues. If you aren’t aware, here’s the summary:

• They require tons of energy - so much so that Lake Tahoe might be losing it’s entire energy supply
• They require a lot of water
• Residents nearby are almost always impacted by negative side effects (noise pollution, light pollution, water pollution)

Impacts to society

The two moral panics

There seems to be two moral panics in the land of AI - the sentient program problem, or the one where AI makes all of our lives a little bit worse all around.

“AI will become sentient”

World-ending outcomes of AI seem to be the biggest fear for some folks today. I don’t quite get it. For people who haven’t done it yet - look into the history of artificial intelligence. The development cycle for AI tends to follow a pattern. That pattern is:

1. Artificial intelligence research continues as normal
2. Some revolutionary discovery is found in how we can further optimize our existing artificial intelligence systems
3. The hype train starts rolling, getting everyone and their dogs hyped
4. Reality settles in and people begin to move onto the next better thing

As humans, we need to remind ourselves of how we survive today. We do something, it turns out good or bad, and we use that as a guide for our next decision. If we see patterns, we need to use those patterns to make informed decisions about where to go next. It’s super easy to get excited about something that seems mystical or out of this world, but we need to remain in reality.

If we truly did have a silver bullet, would any of us have jobs right now?

A sloppier world

A lot of this post is centered around the thought that artificial intelligence is inherently bad, and the badness of that comes from the outcome of using these products. Those outcomes include more bugs in your favorite programs and less reliable sources of truth for everyone. We have evidence of that, which I’ve already went over.

The software crisis

Degrading experience

When you think about the apps, games, programs - any software you use - do they feel like they’ve improved in quality over the years or degraded?

In some ways you can likely say they’ve improved. One less button to click, more realistic graphics. I can agree on these points.

In other ways you can probably say we have regressed. Maybe less features that cost more money (Netflix/Amazon providing ad-based subscription models because it turns out RTB is more expensive than initially thought).

The crisis

If you aren’t familiar with the software crisis, the idea is that software is progressively decreasing in quality as time goes on due to a number of factors. Primarily I think this is being driven by speed. How quickly we can go from a concept to actual code in the real world.

Here’s the problem that the crisis presents. In the modern world, we have very complex processes. Dependencies. Inherently things are complex regardless of whether we like it or not. Due to this fact, the things we build will also be just as complex at minimum. This is what I look at as “essential complexity”, as it’s an essential part of how the solution itself was designed. Maybe there’s a simpler way to do things, but this part isn’t typically the programmer’s job to fix.

Then we have accidental complexity. This is where programming choices are made incorrectly. For example, a developer writing their own algorithm in because they aren’t familiar with using an existing function of a library. An abstraction that confuses future developers. These are things that could be prevented given enough time and space to conceptualize the problem at hand, and could even be avoided had the essential complexity been reduced.

Timelines

As time goes on, developers have been asked to do more, and quicker. We need Feature A out by Q2. We need 3 bug fixes by tomorrow. I have no doubt that the feature and bug fixes can be out by those timelines. But do those timelines make sense based on the complexity of the project?

In my experience they haven’t, and I think many other developers can agree with me. Programmers are being rushed to ship code as fast as possible, which, in some ways, can be a good thing, but in other ways, is a detriment at the cost of the quality of the system. The system begins to accumulate tech debt. Your system reaches a whack-a-mole state: fix one bug, cause another.

Many companies face this problem today, and as it turns out, AI’s “accelerated productivity” doesn’t actually accelerate anything except for how quickly buggy, untested code reaches production.

So on one hand, sure, code lands in production faster. At the end of the day many companies will call this a win. But if the system barely gets by, and is almost always failing in some capacity, at what point is it really a win?

Sure one could argue that software will always have bugs, yada yada yada. But when more bugs are becoming more common in software as a whole, across the entire industry, it’s becoming more and more difficult to say we aren’t moving too quickly.

My anecdotes

Programming

Lack of awareness

I’ve had a few occasions where I’ve been asked to look at pull requests at my current organization where the code was generated. In those scenarios, it’s been blatantly obvious that the code was not authored by a human. So what? Well, in these scenarios, the code attempting to be merged in would have caused devastating issues with the already fragile system we’re working with.

The code in question completely lacked a foundational understanding of how our system works today.

The blindfolded painter

A good way of visualizing this is by using a home. I have a room I’d like to have repainted to match the colors of all rooms in my house. I hire a painter to come and paint the single room in my home. I ask the painter to be blindfolded as they walk through the house to the room. Once they reach the room, I ask them to paint the walls the color they think the other rooms in the house are. They think for a few seconds, and decide on the color they want. They blindfold themselves again, leave the home, get in their car, then drive to the home improvement store to get the paint. They come back, do the same blindfold routine, and then start painting.

The painter can probably make some really good guesses at what color the walls should be. Maybe my house was recently built and the trend that can be seen in most houses this age are white walls. Perhaps he was able to peek beneath the blindfold and realized I have a wood-finish vinyl floor and can use that to make a judgement. There are millions of factors that go into the reasoning of this painter’s decision.

But the decision is not bulletproof. The decision is based off of hypotheticals and best guesses. The painter doesn’t have enough awareness to see anything else except for the room they’re painting. The existing wall color is a dark blue.

The painter paints the room white to match what he thinks is the correct color based on everything he’s gained from his context clues. After finishing painting, I let him see the rest of the house. The color of every room in the house is a color that cannot be picked up from a home improvement store. It’s a light gray that cannot be found anywhere locally.

So what, he’s doomed to fail? Precisely.

The blindfolded programmer

The same here applies to code. The code generated has almost no holistic understanding of the system at play, and even if you do give it that context, it still has the ability to hallucinate a decision without any context. Maybe the painter knew you were trying to trick him, so he painted the walls bright pink. Artificial intelligence is a breeding ground for bugs, as very complex systems require the complex understanding in order to prevent bugs from even occurring.

At least in an enterprise environment, pull requests are almost pretty small. A few lines of code changed. But not an entire rewrite of a system. These real-world scenarios I mentioned earlier almost always change the foundation in a way that cannot be undone and do not have awareness of their own impact of the codebase.

Lack of ownership

When [insert AI tool here] tells a programmer to do something, they program it, and the solution causes a bug, the bug is never the developer’s fault, is it?

“I asked [insert AI tool here] to do [x] and it told me to do [y].”

It’s a way of blame shifting.

Now, don’t get me wrong here: I don’t think serious bugs, programming mistakes, or huge incidents should lead to a developer being severely punished or penalized. I also don’t think the mistake should be forgotten though, nor should ownership of said issue be avoided. A good developer will take full ownership of the problem they started and do everything they can to mitigate it.

But using AI as an excuse is a great cop-out to ownership. When you accept an AI answer in programming, you are wholeheartedly agreeing with the output regardless of what it produces. Whatever you say the answer is, even if it’s regurgitated straight from an LLM, is effectively your answer. When you use a calculator, whatever the answer is, that’s now your answer, even if you typed the formula incorrectly.

The fun

Is work always fun? Not always. Is programming always fun? No way. But is there something fun about spending hours beating your head against a problem only to solve it? Some might say no, but I’d say it’s the bread and butter.

What I’ve learned over the years is that programming is just a lot of trial and error. You try something, see if it works, and then act appropriately. As time goes on though, you learn how to do proper trial and error. Once you land yourself in an error, you revise better each time to where you get to the correct solution quicker.

When I first started programming, it took me days to solve a syntax problem in Objective-C. I didn’t know what I was doing, nor did I know how to actually go back and revise my trial to actually get closer to the answer. But when I finally figured out the solution, it felt great. The dopamine rush of solving a problem isn’t quite as high anymore since I’ve had so many, but a good challenge is always fun.

What fun would there be if your only challenge was how to actually test the code and not how to implement it?

Conclusion

I’m tired of AI just as much as some of you are. If you aren’t tired of it, then I hope you can use the tool to empower yourself to learn. I have yet to find any evidence of any actual learning going on.

For now, and until I see fit, I will be avoiding LLMs, and will be left behind like all of these headlines want me to believe. With me being left behind and all, I don’t think I’ll ever be able to do anything in the real world, and I might as well just never write code ever again for how left-behind I am.

@middleclassfancy on Instagram