The problem with AI

For months, I’ve been getting emails asking me to share my thoughts on AI for developers. Today is that article.

Spoiler alert: if you’re a fan of AI, you’re going to hate today’s article.

The problem with Artificial Intelligence is that only the first word in its name is true.

AI is just expensive autocomplete. It doesn’t “learn.” It doesn’t get “smarter” the more you “train” it. And “prompt engineering” isn’t a real thing.

AI guesses at the correct answer based on other questions like the one you’ve posed and common responses to that question, and spits out a confident sounding answer stripped of all context and attribution.

AI makes the StackOverflow problem, where developers just copy/paste code they don’t understand, worse.

On StackOverflow or some other forum, you often have comments from other users adding context, sharing updates, and offering corrections for mistakes. Blog posts often provide context about how the developer got to their solution or how it works.

AI offers none of that. And it’s often wrong. A lot of times, it just makes shit up.

Generative AI tools are theft.

They steal creative work from others, strip of its licensing and attribution, and spit it back out as original work.

AI tools are accelerating climate change.

They consume massive amounts of water (for cooling) and electricity to do basically the same tasks you could do before AI existed, but worse.

AI isn’t totally useless.

LLMs can be helpful for matching patterns that a human might miss, or doing boring, rote tasks that a human would find painfully boring. But these use cases are narrow and limited.

Generally, AI isn’t just useless. It’s actively harmful and should be resisted at every chance.