← Back to Brain Food My Son Thinks AI Is Putting the World at Risk. So I Did the Homework. Mixed

My Son Thinks AI Is Putting the World at Risk. So I Did the Homework.

• 10 min read • brain-food

When someone believes AI is jeopardizing the world, do you argue or listen? I chose to listen, & discovered surprising truths about AI's real impact. Dive into a journey of facts, tough conversations, & a deeper understanding of technology’s role.

He's genuinely against AI. I build with it for a living. This is me trying to earn a real conversation with my son, on facts, truth, and respect, instead of winning an argument.

My son thinks I'm on the wrong side of this one.

He works for a company that repairs the giant diesel machines behind some of the largest mining operations in the nation. Brutal, skilled, essential work. He's sharp, he's principled, and he does not soften things to keep his dad comfortable. (No idea where he gets that.) For a while now, he's been frustrated with me for building so much of my work around AI, and not in an eye-rolling, kids-these-days way. In a this-actually-matters way. He believes AI is putting the world in real jeopardy. The energy. The water. The speed of the whole thing. The bills we can't see yet. And he's watched his father lean all the way into it.

Here's what I had to sit with. He wasn't being dramatic. He was being honest. And when one of your kids looks you in the eye and tells you they think you're part of something harmful, you've got two options. You can get defensive and recite your reasons (we had that round already). Or you can go do the homework, because they're worth the effort.

I chose the homework. Here's what I found.

The scary numbers got a lot less scary

The headline almost nobody mentions: the per-query numbers that anchored the whole 2024 panic came down by roughly 10x in about a year.

In 2023, the figure circulating was about 3 watt-hours per query. The current measurements fall within a tight band between 0.24 and 0.43 watt-hours. Google's own disclosure puts a median text query at 0.24 Wh. A peer-reviewed 2026 study in Joule from Microsoft Research clocked a median query at 0.31 Wh and found the popular estimates overstate real-world energy use by four to twenty times.

Why the drop? Better chips, leaner models, and an honest correction to some genuinely pessimistic guesses baked into the early math. That "AI uses ten times a Google search" line everyone repeats? It isn't true anymore. The people measuring this carefully keep landing lower, not higher.

It doesn't make AI free. It does make a lot of the headlines about eighteen months out of date.

The water stat everyone shares is just wrong

You've seen it: every AI prompt drinks 500 mL of water. Half a bottle, per question. It gets passed around in group chats and LinkedIn posts, and it sounds damning.

It was a misreading of the original study. The honest range is more like 10 to 30 mL per query, and some air-cooled data centers run effectively zero on-site water.

For context, here's roughly where America's water actually goes:

(source)

If you genuinely want to fight for water, the leverage lies in how we grow food and generate power. The chatbot is a rounding error at the bottom of the list.

We're two links in the same chain

Here's the thought I kept circling back to.

My son keeps those mining machines running. And those machines pull the copper and other metals out of the ground, which end up in the servers I build on. The chips I'm worried about him worrying about start their life in a mine, moved by a machine he keeps alive. His hands are upstream of mine.

I'm not saying that to score a point. The opposite. Seeing it stopped me from scoring any. Because the moment it clicked, the whole "AI versus the planet" framing fell apart in my hands. There's no clean side to stand on. Everything that builds the modern world has a footprint: the mine, the data center, the tractor, the server, his work, and mine. We're not on opposite teams in this story. We're two links in the same chain.

So the honest question was never "does AI use resources?" Everything uses resources. Mining does. Farming does. This article does. The question that actually survives the panic is the harder one: compared to what, and in service of what?

Because the alternative to an AI query is rarely nothing. It's the slow path: ten browser tabs, read them all, cross-check, synthesize, write it up. That burns energy too, plus a couple of hours of your actual life. One AI query compresses it all into a single event. "Compared to what" is the whole game.

The part where my son is right

Now the honest counterweight, because I'm not here to wave pom-poms, and because he'd catch me if I tried. There are two places where the worry holds up.

First, reasoning models. When you ask a model to "think" hard, involving heavy, multi-step, agentic processes, energy use jumps by an order of magnitude or more. One measurement clocked a medium reasoning response around 18 Wh, up to 40 under extended thinking, against a fraction of a watt-hour for a normal question (details). So when someone fires up the most powerful model on earth to ask what time zone Denver is in, that genuinely is wasteful. The fix is boring and it works: match the tool to the job. Don't bring a freight train to carry a grocery bag. (Right-sizing the model instead of reaching for the biggest one is, honestly, most of what I do for clients at Digital Ignitor. It's better for the bill and the planet at the same time.)

Second, location. The national average that makes data centers look tiny hides a real local story. A server farm running on hydro power in a wet climate is a different animal than one sitting in a desert, drawing on a stressed grid and a stressed watershed. I live in the Arizona desert. Watching water-hungry facilities go up next to a shrinking Colorado River is a fair thing to be uneasy about, and I'm not going to pretend otherwise to win a point with my kid. He's right that where we build matters.

The part that actually keeps me up at night

Here's where the research stopped being about the environment for me.

There's an old idea called the Jevons paradox, named after the 1865 case in which coal became more efficient and total coal use went up, not down. Cheaper coal just meant everyone burned more of it. The same trap is sitting right in front of us. If AI lets me do a task in 20% of the time, but I cram 5x more tasks into the same hours, I haven't saved a thing. I've multiplied it.

The technology was never the decision. The decision is what I do with the time it hands back.

I see this at work, too. At Rocket Media, when we swap a stack of manual steps for a couple of AI ones, the real win isn't cranking out five times the output. It's the team logging off at a decent hour. Same tool, two completely different outcomes, and the only variable is what we choose to do with the time we freed up.

Last fall, I built a treehouse with my younger kids over a break. The old me would have burned days researching designs, calculating materials, and second-guessing whether the thing would hold a child. AI did that planning in about two hours. And the temptation, the instant those hours opened up, was to shovel them straight back into work and call it productivity.

Instead, I spent five days hammering, sawing, and high-fiving with my kids. The AI didn't build anything. It gave me back the time to build the thing that mattered, with the people who matter.

That's the whole game. There's a real difference between efficient and good. Efficiency that buys you a treehouse afternoon is a gift. Efficiency that just lets you grind harder is a trap wearing a productivity costume.

If you're having this argument too, here's what actually holds up

A few things worth putting on the table, whether you're the one worried about AI or the one defending it:

The conversation I actually want

So here's where I've landed, and where I'm headed.

I'm not writing this to prove my son wrong. I want to say that plainly, because he's going to read it someday, maybe? I did the homework, and the homework says the planet math is better than the headlines and improving every year, and the water panic was off by a couple orders of magnitude. Those are the facts. They matter, and I won't pretend otherwise to keep the peace.

But facts were never really the thing between us. The thing is that my son cares enough about the world to get angry about it, and I would take a son like that over one who shrugs, every single day. His concern doesn't embarrass me. It makes me proud, even when I'm the one in his crosshairs.

So the conversation I want isn't "here's why you're wrong." It's "here's what I found, here's what still worries me too, and here's how I'm trying to use this stuff so it gives more than it takes." The reasoning model wastes. The grind-harder trap. The data centers are going up in our own dry corner of the world. I'll own all of it. Because the honest version of my position was never "AI is fine." It's "AI has a real cost, and that cost is only worth paying if I do something good with what it gives back."

That's the father I want to be standing in front of him. Not the one who won the argument. The one who took him seriously enough to learn, and respected him enough to tell the truth.

If you're reading this, son, and I hope you are: pull up a chair. I've got the sources. You bring the hard questions. I'll bring the coffee.

This whole rabbit hole started with a sharp, deeply sourced piece by my friend Sebastian Chedal at Fountain City: "Is AI as Bad for the Environment as People Say? A 2026 Read." He did the real work of pulling the current first-party data together, and he framed the question this whole post is built around: "what we do with the time the technology gives back?" If you want the full technical version with all the sources, go read his. It's worth your time.

Most of what I write here is about exactly this kind of question: where AI fits in a life and a business without costing you what makes either one good. If you're having your own version of this conversation with someone you love, join me.

Created with ❤️ by humans + AI assistance 🤖