We Can Now Manufacture Information. We Must Manage Its Pollution.

There is a new type of pollution, and we should call it that: information pollution.

This is the pollution that contaminates the quality of decisions, making them unhealthy like polluted water or air cause illness in our bodies. This matters now, more than ever.

Radio, television, telephony, and the internet exponentially increased the distribution and collection of information, but its generation was still roughly constrained by human creators and human activity. Generative AI can now manufacture information at an enormous scale. This is different than one person or company with a huge microphone misinforming or disinforming people.

Now, with generative AI, there's an increasing supply of information - that we use every day - that has slop in it. And people who want to misinform, or who even just want to produce it, can do so at a blistering, industrial scale. We read it all the time - LinkedIn, Instagram, websites, and even when it's used at our volunteer meetings or at work. It's not always intended to be disinformation, but it's not "clean" and we are exposed to it everywhere.

If our ability to make decisions depends on information, that information is getting more and more polluted - just like processed foods, dirty water, or filthy air.

It matters if our information is polluted, because decisions matter, both big ones and small ones. What if our information environment was polluted enough for estimates of GDP growth to be off by a fraction of a percent, either because the inputs are sloppy, the analysis was sloppy, or because a government can now manipulate how the data gets interpreted more easily with propaganda. Even a small deviation in those sorts of estimates would distort how companies budget and allocate capital by billions of dollars.

At a smaller scale, there's the decision about where to go on a date with my wife. I would use ChatGPT to research new restaurants and plays we could go to. But what if restaurant reviews are now synthetic or bots, what if more and more blogs are AI-generated? Now, the information I'm using to pick a place is distorted, both because of false information but also because of quantity favoring one restaurant over another. This little example is a smaller distortion, but the scale is huge, billions of people across the world probably make a decision like this every week.

And that's the point, just like air, water, and food - we are bringing information into our minds and bodies all the time. And everyone on the planet is making decisions all the time, maybe thousands a day. Even if the information stream we use is a little contaminated, it has huge effects.

Which now means we have to think about managing information pollution. It could be more like a park cleanup, where volunteers on social media already flag information as false. Or it could be more like a superfund cleanup, like in the math community who may end up taking years to verify work on the frontier of math that OpenAI dropped this week, work that has a lot of groundbreaking material but may also have a lot that is unverified, not cited, or even just slop.

Fortunately, we have approaches we learned when managing pollution in the natural environment. What if we had ways to test the quality of information like we did water? What if there was a clean data act, like there was the clean water act? What if we had a superfund program, for cleanup of huge contamination events? What if instead of reduce, reuse, recycle, a campaign was verify, modify, revise?

Whether we like it or not, the problem of information pollution is here. And we have to manage it.