This is a free preview of NOTICE News+

Our Sunday Deep Dives go exclusively to our NOTICE News+ paying members. Our members not only get full access to this Deep Dive, but they also are the backbone of our support. They are the reason we can produce fearless, independent journalism every day.

💻 What is AI and how did we get here?

Before we can understand what artificial intelligence is doing to the world, we need to understand what it is—and why it marks a fundamental break from the kind of computing that came before.

THE BEFORE ERA: Modern computers have been around for roughly a century. And for most of that time, they’ve done what they were built for: math. Fast, precise, emotionless math.

• That’s always been their superpower—calculating, storing, and retrieving numbers far better than the human brain ever could.

But they’ve also had major limitations. Traditional computers need precise instructions—clear input, clear output.

• Want a spreadsheet to calculate your taxes? You had to tell it exactly how. Computers didn’t “think.” They followed rules.

For many computer scientists—and more than a few science fiction writers—the dream has always been something more: a machine that doesn’t just follow orders, but “thinks” for itself, asking the questions and finding the answers.

THE CHANGE: That kind of computing requires vast amounts of processing power, data, storage, and energy—four areas where we've seen exponential advances over the past two decades.

• For example, computer chips—and the way they work together—have advanced dramatically: a single chip today can have over 7 million times the processing power of the one that ran a Windows 95 machine.

At the same time, the amount of data available to computers has exploded—thanks to digitization projects like Google Books, which has scanned over 25 million volumes.

• And storage capacity has kept pace: A solid-state drive the size of your wallet can now store more information than libraries could just a few decades ago.

All of this would be meaningless without the energy to power it. Advances in green energy and efficient computing infrastructure have made it possible to keep the AI engine running—without collapsing under the weight of its own electricity demands.

THE RACE: Those new technologies fueled a race to develop this new style of computing. The public got a first glimpse of this with IBM’s Watson supercomputer, which became famous for beating humans on Jeopardy.

• Watson could analyze clues, parse natural language, and search through vast databases to find the right answer—impressive at the time, but still fundamentally limited to retrieval, not generation.

THE BREAKTHROUGH: The true breakthrough came more than a decade later with the release of ChatGPT—a chatbot built on OpenAI’s large language model, GPT-3.5.

• Unlike Watson, ChatGPT didn’t just find answers. It could write essays, compose emails, explain complex topics, mimic different writing styles—and even write its own computer code.

It wasn't just answering questions—it was creating them.

• Released to the public in November 2022, ChatGPT became the fastest-growing consumer app in history, reaching 100 million users in just two months.

It marked a turning point: artificial intelligence meant that computers could generate their own code—a major step towards being autonomous and “thinking” for themselves.

🧠 How AI works

To be fair, there are many different types of artificial intelligence—each designed to handle specific tasks like vision, movement, sound, or decision-making.

But in this Deep Dive, we’re focusing on the branch of AI that deals with language.

THE TECH: The AI technology that processes and handles language is called a large language model, or LLM for short.

• LLMs power tools like ChatGPT and Apple Intelligence, which are all over the news right now—and are likely to have the most immediate impact on your daily life.

They can write emails, summarize articles, generate code, answer complex questions, and even mimic your own writing style. But how do they actually work?

HOW IT WORKS: An LLM is “trained” on hundreds of billions of words from the internet—books, Wikipedia, news articles, Reddit threads, blog posts, and other websites.

• GPT-3, the LLM behind the first public version of ChatGPT, was trained on about 45 terabytes of text—or roughly a million books’ worth of content.

But “training” doesn’t just mean feeding all that information into a supercomputer. The model doesn’t store facts like a library.

Instead, it analyzes patterns across all that language—how words appear together, how sentences tend to start and end, and what kinds of words are likely to follow others.

IT’S ALL MATH: At the end of the day, a computer is still a computer and works best with math (just like the computers before). In an LLM, language gets boiled down into a massive web of statistical probabilities.

• This is a key difference between human thinking and artificial intelligence. The model doesn’t “understand” your question like a person would—it simply uses what it’s learned to guess what comes next.

When the AI sees a prompt—say, part of a sentence—it predicts the next word. Then it predicts the next one, and the next, building a full response one word at a time.

• It repeats this process trillions of times during training, adjusting itself based on whether its guesses were right or wrong.

All that guessing makes LLMs really great at guessing—so much so that if you give it a question like, “write a poem about the sky,” it will guess a very good answer that looks a lot like poems about the sky that it’s seen elsewhere.

BUT BUT BUT: LLMs don’t know if what they’re saying is true or false. They aren’t thinking or fact-checking—they’re predicting, based on everything they’ve seen before.

• Think of it like autocomplete on steroids: you type a prompt, and the AI fills in what it thinks is the most likely continuation, again and again, until you have a sentence, a paragraph, or an entire essay.

It’s not magic. It’s not sentient. But it is very powerful—because it turns raw data into fluent, convincing language in real time.

And it’s only just getting started.

🛑 The risks of AI

For science fiction writers, the biggest risk of AI is that it will one day become smarter than us—and either accidentally, or by design, try to wipe us out.

QUICK ASIDE: Experts disagree wildly on whether that’s something we should actually worry about.

• Some say it’s a far-off fantasy. Others—like Geoffrey Hinton, one of the so-called “Godfathers of AI”—have left their jobs in protest, warning that the technology could soon outpace our ability to control it.

BACK TO NOW: But whatever the future holds, there are two immediate, real-world risks AI poses today—and they affect everyone:

1. Massive job loss, especially in white-collar industries

2. Enormous environmental impact, driven by energy-hungry data centers

These aren’t speculative. They’re already happening.

JOB LOSSES: Technology has been replacing workers for centuries. The Industrial Revolution displaced weavers and artisans. The 20th century brought robots to factories and devastated mining communities.

But what makes AI different is who it's coming for.

• In past waves of automation, it was blue-collar workers who paid the price—factory hands, warehouse workers, delivery drivers.

But large language models like ChatGPT are targeting white-collar jobs once considered "safe" from machines: copywriters, paralegals, accountants, customer service reps, journalists, even software engineers.

THE IMPACT: According to a 2023 report from Goldman Sachs, generative AI could impact up to 300 million full-time jobs worldwide. 

• In the U.S., the (evil) consulting firm McKinsey estimates that AI and automation could eliminate 30% of hours worked across the economy—by 2030.

The loss of all of those once-safe jobs could continue to drive social unrest both here and abroad. Unless workers fight for protections and their fair share of the profits, the rewards will go almost entirely to the top.

THERE’S MORE: Training and running AI requires staggering amounts of electricity.

• For example, training GPT-3 consumed an estimated 1.3 gigawatt-hours of electricity—the same as what 120 U.S. homes use in a year.

And that’s just one model. The day-to-day operation of AI depends on sprawling data centers powered by massive energy loads. 

• In 2022, data centers consumed nearly 2% of the world’s energy in total. That number is expected to double by 2026, largely because of AI.

Tech companies love to boast about “green AI,” but many data centers still rely on fossil fuel-heavy grids, especially in the U.S. and Asia. And the race to build even larger models means demand is only accelerating.

Unless we put pressure on the government to ban fossil fuels, our embrace of AI will only accelerate the climate crisis.

OTHER RISKS: Beyond jobs and energy, AI threatens to amplify inequality, entrench corporate power, and become a tool for surveillance and social control—threatening democracies and freedom everywhere.

😎 How you can use AI today

That being said, AI isn’t going away. The technology is already being built into the tools we use every day—whether we choose it or not.

• And while the risks are real, the only path forward is to stay informed, push for regulation, and learn how to use this technology wisely.

Because if AI is going to reshape the world, we need to understand how it works—not just to protect ourselves, but to reclaim some power in a system that’s moving fast without our consent.

So here’s a breakdown of AI tools you can start using today—what they do, how to use them, and what they cost.

The rest of this Deep Dive is for NOTICE News+ members

The vast majority of our journalism is free—except the end of our Deep Dives. By becoming a subscriber to NOTICE News+, you’ll not only get access to the rest of this Deep Dive, but you’ll also be supporting our independent, progressive journalism.