When authors Ajay Agrawal and colleagues said “math is cheap” in their book Prediction Machines, they didn’t mean math lacks value. Rather, they highlighted how advancements like calculators and computers made math easier to use—and therefore more accessible to the public. As math became more available, it became a foundational tool for innovation: from data analytics to digital music storage and beyond.
In a similar way, artificial intelligence (AI) is becoming cheaper—not in worth, but in accessibility. And as AI becomes more available, it will be used by more people, in more ways, across more industries. We’ll soon see machines that not only program other machines, but also rely on other systems to manage their own energy efficiency.
So even if you don’t plan to use AI personally, it’s important to understand how it works—and where it’s being used.
The Basics of Prediction
At its core, AI is about making predictions. Predictive modeling is largely about building decision trees—systems that simulate step-by-step logic based on input data. A decision tree is just a series of decisions the model makes to reach a prediction. These predictions are tested against known data to see how accurate they are.
Let me break it down.
Imagine you’re building a model to predict Titanic survivors. You start with a known dataset: passenger names, ages, genders—and whether or not they survived. You remove the “survived” column, then feed the remaining information into a model, asking it to predict survival. Once the model makes its predictions, you compare them with the actual data. If it’s only 10% accurate, the model needs retraining. You keep refining it until the accuracy hits a reliable threshold—typically around 80%. Why not 100%? Because a model that’s too accurate may have simply memorized your training data. It won’t generalize well to new data.
Sites like GeeksforGeeks offer great overviews of decision trees. Experimenting with different decision trees helps you identify which algorithms work best for your data. Eventually, you’ll narrow it down to a few reliable models—saving time and resources in the long run.
So What? Why It Matters.
As prediction becomes cheaper, platforms like ChatGPT, Copilot, and Grammarly are giving everyday people access to AI-powered tools for writing, researching, and more. But it’s crucial to understand: these tools are not omniscient, objective, or sentient.
They’re not “thinking” or “understanding” the way humans do. They’re drawing on enormous datasets—often scraped from the internet years ago—and producing results based on statistical likelihood, not truth or morality.
That means tools like ChatGPT are more trustworthy when solving math problems than offering opinions on human relationships, ethics, or spirituality. They don’t know truth. They assemble patterns. In essence, they’re calculators with words.
Banned Books and Biased Bots
A 2023 Business Insider article reported that two award-winning authors sued OpenAI, claiming it trained ChatGPT using their books without permission. The concern? That these platforms can reproduce content from copyrighted works without crediting the source.
Why is this a problem?
If AI is restricted from using well-researched, peer-reviewed material like books and scholarly articles—and instead can only access open-source blogs, forums, and social media—then it risks becoming less accurate and more inflammatory. The internet is filled with sensationalism, misinformation, and divisive voices. If that becomes the main training data for AI, then we’re teaching it to sound like the angriest corners of the web.
Thanks, Moderators?
It’s also worth noting that behind every AI model is a group of humans deciding what’s acceptable or not. These moderators shape the training data—and their personal or political leanings inevitably influence outcomes. If someone with extremist views (left or right) decides what stays and what gets banned, the AI will start to reflect those extremes.
We’ve seen this before: Twitter’s shadow bans, Grok’s (XAI’s) controversial outputs, and alleged censorship tied to COVID narratives. The AI doesn’t “believe” anything—but it reflects what it’s been fed and what it’s allowed to say.
That’s why it’s dangerous to blindly trust AI platforms. Not because they’re evil. But because they’re machines built by people with agendas—good or bad.
In Summary
I hope this post helps you better understand predictive models. They’re powerful, useful tools—but they’re not wise, not moral, and definitely not your friend. They can amplify both good and bad depending on what they’re trained on—and who is training them.
Be informed. Be discerning. And always ask yourself: Who’s doing the predicting, and who’s writing the rules?
References
Rivera, G. (2023, July 9). 2 authors say OpenAI “ingested” their books to train ChatGPT. Now they’re suing, and a “wave” of similar court cases may follow. Business Insider. https://www.businessinsider.com/openai-copyright-lawsuit-authors-chatgpt-trained-on-books-2023-7?op=1
GeeksforGeeks. (n.d.). Decision Tree Introduction with Examples. Retrieved July 25, 2025, from https://www.geeksforgeeks.org/machine-learning/decision-tree/
Agrawal, A., Gans, J., & Goldfarb, A. (2018). Prediction Machines: The Simple Economics of Artificial Intelligence. Harvard Business Review Press.
Leaders Check 