When it comes to generative AI, people are scared of the wrong things

I’m taking a break from writing about physics this week to talk about Generative Artificial Intelligence. This is where neural networks generate text, images, and video based on prompts from a user. We are familiar with Large Language Models (LLM) such as ChatGPT, Grok, Deepseek, and so on, and everybody’s talking about them.

Unfortunately, I have been seeing a lot half-truths and alarmist claims that are not getting at the crux of what the issues are.

But Tim, aren’t you a physicist? What do you know about AI?

I’m a mathematician, but I’m not going to quibble. I work on problems in physics, but I also write signal-processing software.

Part of that latter activity has involved pitching ideas for using AI, managing AI projects, and sometimes writing code to communicate with AI engines.

In college 25 years ago, I did research in AI groups starting in freshman year and did my senior honors thesis in neuroevolution. I worked with Ken Stanley who was a PhD student at the time and, among many other things, worked at OpenAI, the developer of ChatGPT.

Many people are worried about AI but don’t understand the issues. They think of it as something like the Atomic Bomb, a potentially world-destroying technology that, if in the wrong hands, could bring about the extinction of the human race.

A popular talk called the AI Dilemma suggested that in a survey “half of … over 700 top academics and researchers … [said] there was a 10 percent or greater chance of human extinction from future AI systems” or “human inability to control future AI systems.” This is a bogus claim based on a dubious study as an article in Techdirt explains.

Such predictions are overblown and also ignore the subtle ways that AI can undermine society. These ways are not that different from the ways that social media has already undermined society. After all, the problem with AI isn’t AI acting alone. It is AI combined with social media that is the problem. AI simply exacerbates problems that already exist.

These include: