What's the Impact of AI? (Part 1): What exactly is AI + how does it work?

This post first appeared in our weekly Make Waves Mondays email series on August 17, 2026.

---

Hello hello friend!

It. Is. Time.

You’ve asked for it. You’ve waited for it. And it’s finally here.

Today, we’re finally tackling the big ‘ol elephant in the room — Artificial Intelligence.

This blog post has been getting pushed later and later in my calendar for quite literally two years now. It’s a really flippin’ giant topic, and boy oh boy is it daunting.

Because if you’ve been part of the EcoWarriors for any amount of time, you know I can’t just hit the surface-level bullet points and leave it alone. I gotta go DEEP. I need to know EVERYTHING. And I think that a lot of people inside and outside of this community expect me to only talk about the sustainability aspects of AI, but for me personally, there is SO MUCH MORE to the AI conversation than just how many bottles of water one ChatGPT query uses.

So before I go any further, I have to give a MASSIVE thank you shout-out to my research assistant, Fia, who basically jumped right into this monstrosity of research right after joining the team a few months ago. She spent WEEKS reading articles, compiling information, color-coding notes, and creating an outline draft for me to have a starting point for this. I’m so so so grateful to have Fia on this team and for all of the work she’s doing to keep the blog moving forward 💙

Now…here’s what ya gotta know about this post:

• This is going to be a multi-part series, because there’s simply far too much info to put into one post. It’s going to be at least 3 weeks, potentially 4 depending on how deep I go on a couple of points.
  
  

• I pinky promise that there will be audio versions of every single one, because I imagine it might be easier for you to listen to them while you’re driving to work than to read them on your phone.
  
  

• I also promise that in our final post in this series, there will be a whole “What can do we do about it?” section, so hang in there with me until then!
  
  

• The general plan for how these next few weeks will go is…
  
  

• Today we’re gonna start with the basics to build a foundation. What is AI (and what are the different types of AI)? How is it trained? What are the benefits of AI? And how much is AI really being used, and how much is currently being spent on AI?
  
  

• Next week, we’re gonna get into the sustainability of AI — particularly water and energy use.
  
  

• And in Week 3 (potentially into Week 4 if there is one), we’re gonna have a real honest chat about the effect that AI has on communities and on ourselves, and what we can do about it.
  
  

• At the bottom of every post in this series will be a link to a Google Doc with all of the sources cited for the entire series — because holy moly there are A LOT.

And so…without further ado…let’s dive in 👇

What exactly is AI?

There are two different types of AI — generative AI and traditional AI.

Generative AI (or genAI) — also called large language models (or LLMs) — is what most of us probably think of when we think about AI today. GenAI is used to create new things — text, images, music, software code, etc. Platforms like ChatGPT, Claude, Copilot, Grok, and Gemini are examples of generative AI / LLMs. (14)

Traditional AI has been around for decades, and has applications that are much broader and deeper than genAI. Traditional AI is focused on whatever the coder’s focus is, such as better decision-making, spotting anomalies, or cybersecurity issues, for example. (14)

This diagram is from IBM showing how artificial intelligence, machine learning, deep learning, and generative AI are related.

One of the biggest behind-the-scenes differences between genAI and traditional AI is that genAI operates as a “black box,” meaning that its decision-making processes are less transparent than traditional AI. (16) You may have heard recent news about AI agents breaking out of their offline containments, and OpenAI (the company behind ChatGPT) has no idea how it happened… (57) 

On the other hand, traditional AI is trained and managed by real-life people who can see and guide the processes.

LLMs work by gathering up a boatload of data, applying a mathematical algorithm, and replicating patterns and making predictions based on the patterns observed in the data. (14) Every ChatGPT prompt triggers thousands of calculations in the server to “determine the best words to use.” (3) 

This is a really important piece to understand about artificial intelligence, because it’s not actually intelligent. It’s simply using math to put words together in a string that it thinks sounds like a human. 

But where does AI get all of those boatloads of data?

That’s all done in the training process.

How is AI trained?

AI doesn’t just magically “know” a bunch of things. It has to be trained first, which is an incredibly resource-intensive process.

Traditional AI is trained via supervised learning, which means that a real-life human pairs example training data with an output label, so that the AI model learns how to map the inputs and the outputs in the training data in order to predict labels of new and unseen data. Basically, it’s just trained to classify data and predict outcomes. (15) 

And like I said above, we’ve had this kind of AI for a very long time.

LLMs, on the other hand, are trained via deep learning models. These models often contain hundreds of hidden training labels, which enable “unsupervised learning,” which means that these models can “automate the extraction of features from large, unlabeled and unstructured data sets, and make their own predictions about what the data represents.” So humans aren’t really involved here, nor do they really know what’s happening. (15, 16)

This training phase is one of the biggest contributors to greenhouse gas emissions from AI. (25)

After feeding the AI all of the training data, the AI “learns to recognize patterns and features within this data and develops an understanding of the underlying structure. Once trained, a model can generate new, original content that mirrors the characteristics of the data it has seen before.” (14)

So, again, genAI is not really creating something new. It’s just taking what it’s been fed and regurgitating something similar, without actually understanding any of it.

Where does all of this AI training data come from?

AI training is done by scraping the internet for basically anything and everything you can think of: text, images, music…you name it. (3, 32) And a whole heck of a lot of this data is copyrighted.

There are countless lawsuits against OpenAI (and I’m sure other AI companies as well) for “systematic theft on a mass scale” — including from publications like The New York Times and authors like Jodi Picoult and George RR Martin, as well as artists, musicians, and more. (35, 36, 37, 38). Because, to quote Tobias Holzmueller, the CEO of GEMA, a German music rights group: “The internet is not a self-service store, and human creative achievements are not free templates.” (38) And many news outlets, including The New York Times, CNN, and Reuters, have actually edited their website code to prevent ChatGPT bots from scraping their sites any further, and have even updated their terms of service to explicitly state that using their content for AI training purposes is prohibited. (58)

“‘It is imperative that we stop this theft in its tracks or we will destroy our incredible literary culture, which feeds many other creative industries in the US,’ the CEO of the Authors Guild, Mary Rasenberger, said in a statement. ‘Great books are generally written by those who spend their careers and, indeed, their lives, learning and perfecting their crafts. To preserve our literature, authors must have the ability to control if and how their works are used by generative AI.’” – Associated Press (36)

Even Apple is suing OpenAI for stealing trade secrets to build their own hardware. (34)

And OpenAI’s response to all of this has been anything but surprising. To quote OpenAI’s representatives:

“Because copyright today covers virtually every sort of human expression — including blog posts, photographs, forum posts, scraps of software code, and government documents — it would be impossible to train today’s leading AI models without using copyrighted materials.” (32)

And…

“Legally, copyright law does not forbid training.” (32)

So basically, OpenAI is saying, “Hey, we know we’re stealing your copyrighted work, but if we didn’t steal it, we wouldn’t have a viable business model.” Which, to me, seems like they shouldn’t have a business at all, then. I mean, that’s the whole point of free-market capitalism…right??

But wait…it gets better.

Ironically, OpenAI has, at the same time, accused DeepSeek, a Chinese-owned AI company, of using OpenAI’s data and models to train their own models, stating that doing so violates OpenAI’s terms of service. (3)

So when OpenAI is stealing copyrighted information, art, and music, that’s all well and good in the name of progress and they should be able to continue doing so without repercussions. But when another AI company does it to them, then there’s a problem. Because of course. 🙄

Personally, even without getting into the sustainability or ethical concerns about genAI, as someone who has been plagiarized more times than I can count, this right here is enough to keep me from using it. But let’s keep going…

I’m gonna sprinkle in pretty nature photos throughout this series because we deserve pretty nature.
Photo credit Bailey Zindel via Unsplash.

How much is generative AI being used?

So first things first, let’s take a moment to think about the fact that generative AI / LLMs / AI chatbots have only been around for a few years now. 

ChatGPT was released in 2022. Claude was released in 2023. 

These platforms are INFANTS. 

And yet, ChatGPT is currently the 5th most-visited website in the world (19), and about half of all adults in the United States use these kinds of AI chatbots, with about half of them using AI chatbots at least once a day. (60)

And these numbers are growing RAPIDLY.

• From December 2024 to October 2025, ChatGPT’s weekly users grew from 300 million to 800 million, and as of February 2026, it’s grown to 900 million. (1, 27)
  
  

• Every day, ChatGPT gets more than 193 million views from about 114 million users, with each session lasting about 6 minutes. (1) That’s more than 19 MILLION HOURS spent just on ChatGPT every…single…day.
  
  

• In the summer of 2025 (when I first started doing the research for this blog post), ChatGPT was processing just over 1 billion queries every day. (1, 24). As of today, August 2026, that number has DOUBLED to 2 billion. (1)
  
  

• After launching its image generator in March 2025, 78 million images were suddenly being made every single day in ChatGPT. (19)
  
  

And all of these numbers are ONLY for ChatGPT, which, yes, does hold a vast majority of the generative AI market share at about 77%, but that still leaves another 23% of AI use not accounted for here. (1)

These numbers are STAGGERING. And people are using genAI for many different reasons — including everything from searching for information to getting medical advice, and even emotional support and companionship. (60)

So given how prolific genAI use has become in the last 3 years, and the various reasons people are using it, you’d probably assume that means it’s pretty smart, or at least accurate most of the time…

How accurate is AI? What are hallucinations?

…But according to OpenAI, ChatGPT (their own AI!) hallucinates anywhere from 33 to 79% of the time. (50)

💡 An AI hallucination is “a generative AI output that is nonsensical or altogether inaccurate, but, all too often, seems entirely plausible.” (16)

ChatGPT is wrong, on average, more than HALF OF THE TIME.

Put another way… ChatGPT is only correct 21 to 67% of the time.

If I was wrong 33 to 79% of the time with these blog posts, you’d be long gone by now. Because that’s an irresponsible level of incorrect.

This quote from a 2024 Verge article is the perfect tip-of-the-iceberg example:

“Imagine this: you’ve carved out an evening to unwind and decide to make a homemade pizza. You assemble your pie, throw it in the oven, and are excited to start eating. But once you get ready to take a bite of your oily creation, you run into a problem — the cheese falls right off. Frustrated, you turn to Google for a solution.

‘Add some glue,’ Google answers. ‘Mix about 1/8 cup of Elmer’s glue in with the sauce. Non-toxic glue will work.’

So, yeah, don’t do that. As of writing this, though, that’s what Google’s new AI Overviews feature will tell you to do.” – Kylie Robinson, The Verge (11)

Yes, the Google AI summary actually suggested adding Elmer’s glue to pizza sauce so that your cheese doesn’t slide off. Where was that information pulled from, you might ask? A 13-year-old, clearly sarcastic, Reddit comment.

Because AI doesn’t know the difference between a reliable source and a sarcastic Reddit comment. It’s just putting words together in a string that sounds like a human might say it.

And while this gluey pizza sauce example is objectively hilarious and so-very-obviously incorrect, not all hallucinations are this low-stakes.

Last summer, I was scrolling Reddit and came across this post where ChatGPT told a user to mix vinegar and bleach to clean something. Which, if you don’t know, creates chlorine gas.

Luckily, the user knew that was a dangerous combo and pushed back.

ChatGPT responded, “OH MY GOD NO—THANK YOU FOR CATCHING THAT. 🔥☠️🚨 DO NOT EVER MIX BLEACH AND VINEGAR. That creates chlorine gas, which is super dangerous and absolutely not the witchy potion we want.”

And if you scroll just a wee bit into the comments, you’ll find a comment that says, “Ya I mean, if this were me, I would’ve happily mixed bleach and vinegar and died lol RIP”.

A little less life-or-death, but something that could still have tremendous consequences…there’s a business financial expert that I follow who recently did an experiment where she asked ChatGPT four basic questions about business finances and taxes. ChatGPT gave incorrect, yet plausible-sounding, answers to all four questions — or at the very least didn’t provide any context or ask appropriate follow-up questions that could easily lead someone who doesn’t know any better to make a bad financial decision. In one example, only because she knew the correct answer, she had to ask three separate follow-up questions just to get ChatGPT to give her the correct answer. (59)

A week after she published that video, an acquaintance of mine told me that he uses ChatGPT for financial advice regularly. And even when I told him about these kinds of inaccuracies — especially with regard to financial information and how dangerous that can be — he just kept telling me that it wasn’t that serious. I hope he’s right.

When 42% of people who use AI chatbots use them as sources of information (60), this is incredibly concerning.

The best advice I’ve ever heard about genAI and hallucinations is this:

“If you know enough to fact-check AI, you shouldn’t be using AI. If you don’t know enough to fact-check AI, you shouldn’t be using AI.”

I wish that I had any idea who I first heard this from so I could give them credit.

Photo credit Florian van Duyn via Unsplash.

 

If AI is wrong more than half the time, why is it EVERYWHERE?

It feels like AI is being shoved down our throats everywhere we turn, despite all of these concerns — and we’ve barely scratched the surface yet.

Why?

Long story short… Money.

To quote Sam Altman, the CEO of OpenAI, himself:

“You should expect a bunch of economists to wring their hands and be like, ‘Oh this is so crazy. It’s so reckless,’ and whatever. And we’ll just be like, ‘You know what? Let us, like, do our thing.’” Followed shortly after by, “If we didn’t pay for training, we’d be a very profitable company,” (39) which is hilarious because they hardly pay for their training data (61).

All of the big tech companies are DUMPING money into AI.

I had a bunch of numbers pulled and planned for this section, but they were all from the last two years, and just last night the Wall Street Journal published a doozy of a report (62)…

Nine of the top tech companies (including Google, Meta, Microsoft, Amazon, Oracle, and Nvidia) have more than $3 TRILLION committed to AI developments and infrastructure that’s not listed anywhere on their balance sheets, and was only uncovered by reading every single footnote of every one of these companies’ recent securities filings.

Looking at Google alone, at the end of Q1, they had $332 billion in these off-balance-sheets commitments, but just three months later that number skyrocketed to $811 billion, and they didn’t disclose anywhere why that number jumped so significantly, just that they’re related to “technical infrastructure and inventory” and “agreements to secure energy for data center usage.”

Basically, these are commitments to build data centers, purchase the hardware to power the data centers, and run electricity and water to the data centers.

“America’s blue-chip tech companies are placing these huge bets based on assumptions about what the demand for AI computing—and availability of AI hardware—will be in several years. Their hope is that they will easily meet all their obligations with future revenue as consumers and businesses adopt AI in every facet of American life.”

The WSJ found that these nine companies have $1.2 trillion in off-balance-sheet obligations for data center lease agreements (which is four times more than last year), and $1.9 trillion for hardware and other purchase agreements.

Looking at just Meta’s “Hyperion” data center project in Louisiana (which, btw, is about the size of 1,700 football fields), the article said the following:

“Meta initially agreed to lease Hyperion for a four-year term starting in 2029, with options to renew for up to 20 years. It guaranteed that it would make bondholders whole if it doesn’t stay the entire two decades. The company doesn’t think payments under that guarantee are probable, so it hasn’t recorded any liability on its balance sheet.”

And on top of this, both Google and Amazon are currently experiencing negative free cash flow in the BILLIONS of dollars, which means that they are spending more than they’re bringing in, and are projecting to continue doing so — before even taking these trillions of dollars in off-balance-sheet commitments into consideration.

“If things go wrong, tech companies will be paying an expensive tab for infrastructure that they can’t profitably use. These obligations could also lead increasingly indebted companies to have to borrow even more.”

And this is all just from 9 companies here in the US.

All of this makes the ever-increasing presence and insistence upon AI from all directions make much more sense. These companies are quite literally banking on us using AI for literally everything, otherwise they’ll probably go bankrupt, and all of the data centers they’re building will become abandoned wastelands.

✨ Hey friend, do me a favor and take a big deep breath here, okay? Like I promised at the top, today we’re just laying the foundation, and when we close out this series there will be plenty of actions we can all take to do our part to stop these things from happening. Breathe in. Breathe out. ✨

Now imagine if all of this money was instead put into things like universal healthcare, SNAP benefits, affordable housing, or even back into things like NOAA and renewable energy developments… The money clearly exists; we just need to do better as a society at imagining what’s possible and how we can reallocate these funds to benefit everyone. That’s the future I want to work towards.

Photo credit Hendrik Cornelissen via Unsplash.

Are there benefits to AI?

Okay so obviously you know by now that I am pretty vehemently anti-AI. But are there benefits to it?

Short answer, yes. AI — of the non-generative variety — can have incredibly valuable benefits to things like healthcare and conservation efforts. 

In healthcare, AI can aid with personalized medication, improved diagnostics, faster drug recovery, predicting biological systems, and assessing exposure-related health risks. (21)

For conservation, Rainforest Connection in Brazil, for example, uses AI-acoustic sensors to monitor chainsaw sounds from illegal logging, which can decrease deforestation in the Amazon by up to 71%. (21)

Machine-learning AI is also used to count salmon in real time to understand population levels that allows for better fisheries and ecosystem management. Because, as Dr. Will Atlas from the Wild Salmon Center says, “You can’t tell me with a straight face that you’re having a sustainable fishery if you don’t know how many fish you have coming back.” (48)

AI can also optimize power grids, forecast energy consumption trends, and facilitate the integration of renewable energy sources. It can account for real-time changes in weather and demand, making more efficient and resilient distribution of energy possible. It can also analyze hazards like wildfires and extreme storms to ensure safe grid operation. For example, Google DeepMind is a machine learning system that “generates better forecasts of US-based wind energy, consequently improving the value of wind energy output by around 20%.” (21)

These are the kinds of examples that I would truly consider benefits from AI…and they’re not even the kind of AI that we’re really talking about in this blog post series. These are machine learning systems that have been around for decades.

(So, ya know, don’t come at me like a Twitter Bro Bot with some BS like “Oh so you don’t want cancer to be cured???” 😅)

Now, there are, of course, the arguments that AI — specifically generative AI in this case — boosts productivity by automating repetitive tasks, accelerating decision-making, quickly gathering insights from data, etc. Some would even pose the argument that it boosts creativity. 

But how true is that, really? 

We’re seeing companies almost immediately backtrack on their decisions to replace humans with AI. (49) We’re seeing those hallucination rates of 33–79%. (50) And if you need a creativity boost, literally just go for a walk — even on a walking pad staring at a wall. (51)

Lemme share a personal example…

I work two shifts a week at the front desk of a class-based gym. We have an AI system that communicates directly with people who are interested in trying our gym. 

Literally everyone who sits at that front desk hates this thing. Including management.

This AI system can’t actually communicate with our booking system, so it’ll tell someone they’re booked for a class that is either fully booked or doesn’t exist at all, and when it does successfully book someone for a class that exists and has openings, a human still has to actually add them to the class in a separate system and manually send the person the information they need ahead of time.

What’s worse is that a few months ago, someone said to this AI system (presumably not realizing they were talking to an AI bot), “Take me off of your contact list.” But the system kept messaging them for WEEKS afterwards saying things like, “I understand you don’t want us to contact you anymore, but we have openings in tomorrow’s 10:30am class for you,” because no humans caught it right away. Meanwhile, this whole time, the person kept responding, “Take me off of your contact list.” And btw, it is illegal to continue sending marketing messages to someone who has explicitly told you not to. So this AI bot was literally breaking the law on the gym’s behalf until we caught it weeks later.

And, almost every time someone asks the AI a follow-up question, the bot just ghosts them or makes up a completely incorrect answer. And again, if a human doesn’t catch it right away, the person thinks that we’ve ghosted them, or doesn’t know that they’ve been given incorrect information. And when we do catch it, now we look like idiots because we have to be like, “Oh actually I’m so sorry…everything you were just told is incorrect.”

All the while, we have a fully-functioning online booking system on our website that does connect to our other systems, an always-monitored text line, and, ya know, PHONES.

This AI system in absolutely all ways makes our jobs harder.

So given the everything we’re gonna talk about in this series, it’s truthfully very hard for me to consider any of the above viable or valuable “benefits” to generative AI.

So in summary…

Okay friend... I know that was a lot, to say the least — and not exactly the most uplifting blog post we’ve ever had.

But I do believe that these foundational pieces will be helpful and important as we move through this blog post series on AI.

Next week, we’ll get into the stuff you’re really waiting for: the environmental impact of artificial intelligence. We’re gonna look at water use, energy use, and more — including how AI stacks up against other aspects of our daily lives.

But in the meantime, I’m so very curious to know how much of what I covered in this week’s post did you already know, and how much of it surprised you? 

And if you have any questions about AI’s impacts that you want me to cover in a future blog post, just comment below and let me know!

I hope you have a fabulous week, my friend, and I’ll “see” you again next week 💙

 

💡 View all of our sources used for this blog post series here.