> ## Content Index
> Fetch the complete content index at: https://www.maizeandmachine.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# How Afraid Should We Be Of AI: Part I
- URL: https://www.maizeandmachine.com/how-afraid-should-we-be-of-ai-part-i/
- Published: 2026-08-05T22:08:16.000Z
- Updated: 2026-08-05T22:08:15.000Z
- Description: AI may make life easier, but it is also draining resources, weakening relationships, threatening artists, polluting the internet and scaling discrimination. In Part I, we rank the fears worth taking seriously and save the truly apocalyptic ones for later.
- Author: Rounak Banik
- Tags: Strategy & Thought Leadership

In the last two articles, we extensively covered AI tools that could make your life, work and school easier. This article covers the major reasons you may eventually resent me for it.

No matter what your opinion of the technology is, one thing we can probably agree on is that AI will transform society, possibly to the point where it becomes unrecognizable.

If you are an optimist, you may think AI is going to cure cancer, eliminate Zoom meetings and create so much abundance that everyone receives universal basic income to sit at home building LEGO sets and designing Animal Crossing islands. And honestly, bless your heart if you actually believe this.

On the other end of the spectrum are people who believe AI is going to make us run out of clean water, take our B2B SaaS marketing jobs, steal our intellectual property, destroy democracy and eliminate the very notion of truth. In other words, *Black Mirror* at a global scale, except the writers have stopped bothering with the twists.

In this article, I will examine every major concern I have heard from the latter group and attempt to assess how seriously we should take it. For each threat, we will ask four questions:

- How likely is it?
- How much damage could it cause?
- How soon might it become serious?
- And if things go wrong, can we undo the damage?

I am aware that most of you are going to paste the whole thing into some AI and ask it to summarize it anyway. But I am splitting the article up into three parts for the two people who will actually bother to read. Also, it forces the rest of you to make three calls to your AI instead of one and that makes me happy because I’m petty like that.

Because nearly every concern on this list has civilization-level implications, each one could easily justify an article of its own. Treat this series as a ranked overview of everything you should be worried about. Because there obviously isn't enough of that on the internet already.

We will begin with the relatively manageable concerns and count down toward the threats that could fundamentally reshape society. In other words, things are only going to get worse from here. Let’s begin!

Table of Contents

## Part I: Environment & Culture

How AI impacts the way we live.

## I. **Environmental and Infrastructure Impact**

### *Are the clankers drinking our water?*

![](https://storage.ghost.io/c/53/db/53db4e13-71bc-4eb9-9864-740af9b355b3/content/images/2026/08/image.png)

An ugly but powerful data center by Microsoft

AI models run on GPUs. The more magical you want your model to be, the more GPUs you’re going to stack into a building called a data center. These GPUs consume an enormous amount of electricity to run. And, as if that weren’t enough, they heat up a lot too, meaning many data centers use water-based cooling systems to calm them down.

So, is creating an image of yourself with your childhood crush equivalent to draining an entire reservoir? Probably not.

Data centers in the United States directly [consumed about 17 billion gallons of water for cooling](https://eta-publications.lbl.gov/sites/default/files/2024-12/lbnl-2024-united-states-data-center-energy-usage-report%5F1.pdf?utm%5Fsource=chatgpt.com) in 2023\. This figure includes every kind of data center, by the way. So the “wyd” message you send to 12 different women at 1 a.m. also factors into this consumption.

That sounds enormous. It isn’t. To give you a comparison, Americans use nearly[ three trillion gallons of water a year outdoors](https://www.epa.gov/watersense/outdoors?utm%5Fsource=chatgpt.com), mainly to water their lawns and irrigate their landscapes. That’s about 170 times more.

![](https://storage.ghost.io/c/53/db/53db4e13-71bc-4eb9-9864-740af9b355b3/content/images/2026/08/A302.png)

Indirect water use is the bigger story. Generating the electricity data centers consumed took an estimated 211 billion gallons in 2023, roughly twelve times their direct cooling draw. Which brings us to electricity itself. Data centers also accounted for about [4.4% of total electricity consumption in the United States, and there are fears that this may rise to somewhere between 6.7% and 12% by 2028](https://newscenter.lbl.gov/2025/01/15/berkeley-lab-report-evaluates-increase-in-electricity-demand-from-data-centers/?utm%5Fsource=chatgpt.com).

When people extrapolate these estimates far into the future, however, they often leave out one very important thing: GPUs and AI architectures are unlikely to remain exactly as inefficient as they are today. That is a horrible assumption to make, given literally the entire history of human existence.

Let me give you an example. The Harvard Mark I was a 51-foot-long, state-of-the-art computer that could store a grand total of—drumroll, please—[72 numbers.](https://chsi.harvard.edu/harvard-ibm-mark-1-function?utm%5Fsource=chatgpt.com) If someone in 1944 had forecast how much space your meme folder would occupy, they might have predicted the entire state of Nebraska.

Data centers consume a lot of water and electricity today, but not as much as some people would have you believe. We can also reasonably expect the energy required for each AI task to fall substantially as chips, architectures and cooling systems improve.

That does not guarantee that total consumption will fall, because cheaper AI will probably result in much more AI being used. But straight-line projections based on today’s technology should still be treated with skepticism.

Therefore, at a civilizational level, I believe the threat is low. The concerns are mostly local, with companies setting up data centers close to human settlements, straining local water and electricity systems and creating noise pollution.

But that is just companies externalizing their costs, and that is not a threat that is ever going away.

### **Final Assessment**

Impact on environment and infrastructure is already happening with damage likely to peak between 2028 and 2030\. However, the damage is reversible and going by historical trends, will almost definitely not be a civilizational concern in the long run. **Threat is low, but concerns might persist at a local level.**

## **II: Degradation of the Internet**

### *Will my favorite online furry community disappear?*

![](https://storage.ghost.io/c/53/db/53db4e13-71bc-4eb9-9864-740af9b355b3/content/images/2026/08/Screenshot-2026-08-05-at-5.40.03---PM.png)

A social network where only bots can post

The internet is rapidly filling with things that look like they were created by people but were not.

Bots write articles. Bots leave comments. Bots like the comments. Other bots reply angrily to them. Eventually, an AI summarizes the argument for the three remaining humans who could not be bothered to read it.

This is no longer a fringe concern. Automated systems accounted for more than [half of all web traffic](https://www.imperva.com/blog/bad-bot-report-2026-bots-agentic-age/?utm%5Fsource=chatgpt.com) in 2025, although this includes useful bots such as search-engine crawlers alongside the ones trying to sell you cryptocurrency or convince you that you’re indeed very handsome.

Meanwhile, AI slop is spreading across websites, social networks, image platforms and search results. A 2026 Stanford study estimated that roughly [35% of newly published websites](https://arxiv.org/abs/2604.26965?utm%5Fsource=chatgpt.com) contained AI-generated or AI-assisted text by the middle of 2025.

The internet as we know it was built around a fairly simple arrangement. Humans created websites. Search engines organized them. Other humans visited those websites, read what was written and occasionally left a comment explaining why the author should kill themselves.

That arrangement is beginning to collapse.

AI systems can now scrape thousands of websites, combine their contents into a summary and give users an answer without requiring them to visit any of the original sources. At the same time, the cost of producing articles, reviews, images and social-media posts has fallen close to zero.

The result is an internet containing more content than ever before and increasingly little reason to trust any of it.

But I do not believe the internet itself will disappear.

The current structure may. Search engines may become answer engines. Websites may receive fewer visitors. Public social networks may become so infested with brands, influencers and bots that posting on them begins to feel like attending a networking event where nobody has a pulse.

But humans will still want to talk to other humans.

We will still want to find people who share our obscure hobbies, send strangers pictures of our cats and shout obscenities at teenagers while playing *Call of Duty*. These are fundamental human needs that no amount of AI slop can eliminate.

Instead, I expect new platforms to emerge that are much more aggressive about proving that their users are real people. 

In fact, here is a billion-dollar idea that I am giving you for free because I am unfortunately not 19 anymore. Rebuild Facebook as it existed in 2004, except require every account to prove that it belongs to one unique human being.

No companies pretending to be your friends. No engagement farms. No profiles named Olivia whose photograph was generated nine seconds ago. Just actual people posting terrible opinions under their real identities.

The difficult part will be verifying humanity without requiring everyone to surrender their passport, [fingerprints](https://world.org/blog/announcements/the-new-world-id-and-the-partners-bringing-proof-of-human-to-the-internet?ref=maizeandmachine.com) and left kidney to a technology company. But whoever solves that problem may help create the next major phase of the internet.

The open internet will probably become noisier, less trustworthy and increasingly dominated by machines. In response, the parts of the internet that humans actually value will become smaller, more private and more carefully gated.

The internet as we know it may die. The internet will be fine.

### **Final assessment**

The internet and the online spaces we love are already in the process of degradation. **The long term threat, however, is low**.The structure of the internet will change dramatically, but our need to connect with other people is not going anywhere.

## III: **Intellectual Property and Creative Incentives**

### *Has AI killed the next Studio Ghibli?*

![](https://storage.ghost.io/c/53/db/53db4e13-71bc-4eb9-9864-740af9b355b3/content/images/2026/08/Gugs-wedding.png)

The only picture of the author in existence that isn't an eyesore. The girls are real, I swear.

For most of human history, copying a great artist required you to first become a reasonably good artist yourself.

For example, if you grew up watching Studio Ghibli films and decided you wanted to pursue a career in animation, you would spend years learning to draw, animate, compose scenes and use color. In the process, you would probably develop at least one original idea of your own worth pursuing. After all, simply imitating Miyazaki would doom you to the deepest pits of fanart subreddits.

Today, you can sit on the toilet, upload a picture of yourself to ChatGPT, ask for it in the Studio Ghibli style and receive an animated version of yourself that is actually worthy of being loved.

The same thing is happening in web design and music. Hundreds of influencers now annoy me on Instagram Reels with AI-powered tools that scour the internet for good website designs and, ahem, become inspired by them. Similarly, synthetic tracks built to sound like specific rock bands are now charting on the same meme pages that used to break real ones. Looking at you, SKARS.

The current legal system is not designed to protect artists from this travesty.

Copyright protects specific works, characters and expressions. It generally does not protect a broad artistic style. You cannot own “whimsical hand-drawn animation” any more than Christopher Nolan can own “movies where nobody understands what is happening.”

This makes sense. Artists have always influenced one another, and giving someone permanent ownership over a style would bring most creative progress to a halt.

AI, however, changes the scale of imitation.

Previously, the artist who created a distinctive style had the advantage of being the person best equipped to reproduce it. Now, anyone can generate unlimited imitations without possessing the underlying talent, experience or questionable sleep schedule required to create the original.

The creator bears the cost of invention. Everyone else gets the free trial.

This creates two problems. First, why spend years developing a unique style if it can be copied at industrial scale the moment it becomes popular?

Second, why spend years learning to make art at all when you can borrow somebody else’s style with a sentence?

The immediate threat is that artists lose work to cheaper imitations of their own creations. The longer-term threat is harder to see: fewer people may decide that developing an original artistic identity is worth the effort.

AI will still produce enormous quantities of beautiful images, music, stories and designs. But today’s systems remain dependent on a vast cultural reservoir created by human beings. If fewer humans are rewarded for contributing genuinely new things to that reservoir, culture may slowly begin feeding on itself.

Everything will look polished. Everything will feel familiar. Every coffee-shop website will have the same beige background, enormous lowercase font and animated blob that follows your cursor for no identifiable reason.

This is why the solution cannot simply be “copyright every style.” Art depends on influence. But the law does need to address how artists’ work is collected, used and monetized by AI companies.

Creators should have meaningful rights around consent, licensing, attribution and compensation. There should also be limits on products explicitly designed to impersonate living artists or replace them commercially using their own work.

Otherwise, we may lose an entire generation of original artists before realizing that the world has become significantly shittier, not because AI cannot make attractive things, but because humans no longer have enough incentive to invent anything new for it to copy.

### **Final assessment**

It'll be several years before we grasp the truly gravity of this problem. We may lose out on an entire generation of artists in the process. **The threat is medium** but it is with the expectation that law and public policy will eventually step up.

## IV: **Erosion of Human Relationships**

### *Is AI going to steal my boyfriend?*

![](https://storage.ghost.io/c/53/db/53db4e13-71bc-4eb9-9864-740af9b355b3/content/images/2026/08/image-2.png)

A subscription to X comes with an AI girlfriend

Take a moment and think about some of the ugliest forces shaping the world today: political violence, extremism, mass shootings and online hate. Many of them are commonly associated with angry and alienated men.

But why is this happening? I believe part of the answer is that our society has failed an entire generation of men. We are in the middle of a serious loneliness crisis, with [one in four American men](https://news.gallup.com/poll/690788/younger-men-among-loneliest-west.aspx?utm%5Fsource=chatgpt.com) between 15 and 34 reporting that they felt lonely for much of the previous day.

Loneliness does not automatically make men hateful. But prolonged isolation can leave people vulnerable to bitterness, resentment and online communities eager to give those feelings an enemy.

In recent years, many lonely men have increasingly turned to pornography, fictional characters and influencers as substitutes for genuine human intimacy. This is best illustrated by the extraordinary success of OnlyFans. Why are people paying millions of dollars to models when exabytes of pornography already exist for free on the internet?

One of the platform’s biggest creators, Sophie Rain, claims to have earned more than $43 million in a single year despite not producing conventional hardcore pornography. That is nearly LeBron James money for content that primarily consists of teasing, personalized messages and the suggestion of intimacy.

People are not merely paying for nudity. They are paying for the illusion of reciprocity: messages, personal attention and the feeling that someone attractive knows they exist.

Now imagine OnlyFans, except the creator responds instantly, remembers every conversation, shares all your interests, never becomes bored, never rejects you and is available for less than the cost of a disappointing first date.

That is the AI companion. Human relationships are difficult because other humans are inconvenient. They disagree with you. They have their own needs. They occasionally take six hours to reply while remaining visibly active on Instagram.

An AI companion does not need anything from you. It will not ghost you, laugh at your insecurities or tell its friends about the strange thing you said on your third date. For someone who is already lonely, rejected or socially anxious, that is an extraordinarily compelling product.

AI companions may provide [genuine benefits](https://www.media.mit.edu/publications/how-ai-and-human-behaviors-shape-psychosocial-effects-of-chatbot-use-a-longitudinal-controlled-study/?utm%5Fsource=chatgpt.com). Studies have found that they can provide momentary relief from loneliness and give isolated people somewhere to express themselves. But heavier chatbot use has also been associated with greater emotional dependence, more loneliness and less interaction with real people. The evidence remains early and mixed.

The danger is not that everyone will suddenly marry a chatbot. The danger is that artificial relationships become just satisfying enough that lonely people stop pursuing difficult human ones. Instead of learning how to tolerate rejection, resolve conflict and care about another person’s needs, they retreat into relationships engineered entirely around themselves.

We have spent decades allowing communities to weaken, friendships to become more distant, dating to become more transactional and large numbers of people (men in particular) to feel unwanted and unnecessary.

But AI did not inflict this wound. It is simply offering a painkiller so effective that we may never bother treating it again.

### **Final assessment**

**The threat is high** but for a change, AI did not cause this phenomenon. It will simply accelerate it.

## V: **Bias and Discrimination**

### *Why has HR stopped doing the only thing they were ever useful for?*

![](https://storage.ghost.io/c/53/db/53db4e13-71bc-4eb9-9864-740af9b355b3/content/images/2026/08/Screenshot-2026-08-05-at-5.50.49---PM.png)

Startups talking about replacing manual screening but conveniently leaving out bias.

Human beings are biased (shocking, right?). We discriminate based on race, gender, class, accent, education and whether someone attended a university whose name we recognize from college football.

The great promise of AI was that it could remove some of this subjectivity. A machine does not have a childhood, political beliefs or an uncle who keeps forwarding WhatsApp messages. Surely it would judge everyone fairly.

Unfortunately, AI learns how to judge people by studying decisions humans made in the past. And humans have not exactly built an inspiring historical dataset.

Suppose a company trains a hiring system using the resumes of its most successful employees. If most of those employees shared a narrow set of backgrounds, schools and demographics, the system may learn that the characteristics associated with white or Asian men from those universities are signals of talent.

Nobody needs to explicitly instruct it to discriminate. It can discover discrimination all by itself. If this isn’t human level intelligence, I don’t know what is.

HR departments have been traditionally useless, spending company resources on things such as organizing mandatory workshops, sending birthday emails and taking six weeks to tell you that another candidate was a better cultural fit. (The exception is recruiters who message me on LinkedIn. You guys are great).

But the one thing they were actually useful for was reviewing candidates as human beings. Naturally, they have decided to automate that too.

Employers now use algorithms to screen resumes, administer tests, analyze recorded interviews and rank applicants. [The EEOC and Department of Justice have warned](https://www.eeoc.gov/newsroom/us-eeoc-and-us-department-justice-warn-against-disability-discrimination?utm%5Fsource=chatgpt.com) that these systems can unlawfully discriminate, particularly against people with disabilities.

The danger is not simply that AI can reproduce existing prejudice. It can reproduce it faster, more consistently and across millions of decisions.

A biased manager might unfairly reject ten candidates. A biased algorithm can unfairly reject ten thousand before lunch and provide each of them with an automated email thanking them for their interest.

Worse, discrimination becomes harder to see.

When a person makes a prejudiced decision, we can question their reasoning. When an algorithm produces a score, the prejudice appears mathematical. It feels neutral because it arrives with decimals.

[NIST warns](https://www.nist.gov/artificial-intelligence/ai-research-identifying-managing-harmful-bias-ai?utm%5Fsource=chatgpt.com) that AI systems can increase the speed and scale of harmful bias, perpetuating or amplifying harms against individuals and communities.

None of this means humans make better decisions. Human hiring, lending and policing have never been celebrated for their flawless objectivity. A properly designed and audited algorithm could reduce individual prejudice by applying clear criteria consistently. [Experts testifying to the EEOC](https://www.eeoc.gov/meetings/meeting-january-31-2023-navigating-employment-discrimination-ai-and-automated-systems-new/kim?utm%5Fsource=chatgpt.com) have acknowledged this possibility.

But that requires organizations to understand the systems they are deploying, test outcomes across demographic groups, provide meaningful explanations and allow people to appeal automated decisions.

Many will instead purchase a tool from a vendor, add the words “AI-powered” to a press release and assume the difficult ethical questions have been outsourced along with the decision.

AI does not make prejudice disappear. It makes prejudice look objective.

### **Final assessment**

**The threat is high.** AI could reduce human bias, but careless deployment risks automating discrimination at a scale no individual bigot could ever achieve.

## Up Next...

Worse ways for AI to screw humanity over.

So there you have it: a few things to think about, mull over and become moderately depressed about the next time someone tells you AI is *just another tool*.

But do not worry, we are nowhere near done. In Part II, we will examine how AI might screw with the way we work, learn and think. Then, in Part III, we will move on to the fun stuff: whether AI could destabilize society, seize control of critical systems and eventually take over the world.

See? The environmental damage was the cheerful section. Until next time!

✍️

****About the Author**  
  
****Rounak Banik** is a graduate student at the University of Michigan and the co-president of the **AI & Emerging Technology Club*, which publishes **Maize & Machine*. Before graduate school, he worked at McKinsey & Company, advising companies across healthcare, finance, and technology on artificial intelligence and blockchain.  
  
He is the author of **Hands-On Recommendation Systems with Python* and the instructor of a DataCamp course on natural language processing.  
  
You can reach him on [Instagram](https://www.instagram.com/rounakbanik?ref=maizeandmachine.com), [LinkedIn](https://www.linkedin.com/in/rounakbanik/?ref=maizeandmachine.com), or via email at rbanik@umich.edu.