When Bots Talk to Bots: The Strange Experiment That Exposed Social Media’s Dark Side

A Familiar Mess

Let’s be honestsocial media isn’t exactly the shining “digital town square” we were once promised. At this point, calling it toxic feels almost redundant. Anyone who’s spent ten minutes scrolling through Twitter (sorry, “X”), or dipped into a Facebook comment thread about vaccines, knows the drill. Outrage spreads faster than facts. Misinformation gets juiced by algorithms. And somewhere along the way, the idea of these platforms as places for thoughtful debate died a quiet, miserable death.

But what if we could rewind the clock? What if we tried to build social media from scratchonly instead of actual people, every user was a bot? Would the same problems appear, or would an artificial society behave differently?

That’s exactly what a team of researchers in Amsterdam set out to test. And the results, well… they weren’t exactly encouraging.

A World of Bots

The project, led by assistant professor Petter Törnberg and research assistant Maik Larooij at the University of Amsterdam, was simple in concept but unsettling in its execution. They created an entire social network populated exclusively by AI users. No humans. No influencers. No trolls sitting in their basements. Just bots powered by GPT4o, OpenAI’s latest large language model.

Then they ran the simulation, watching how these digital citizens behaved, what content spread, and whether common “fixes” for toxic social media could steer the platform toward something healthier.

You can probably guess where this is going.

Trying to Fix the Machine

The researchers tested six wellknown intervention strategiesideas that social media reformers have been tossing around for years. Things like:

• Switching to chronological feeds instead of algorithmic ones.

• Boosting diverse viewpoints to break echo chambers.

• Hiding follower counts and “likes” to reduce popularity contests.

• Removing user bios to cut back on tribal identity signaling.

In theory, each of these should make online spaces less polarizing. But when the bots tried them out? The outcomes were underwhelming at best and counterproductive at worst.

For example, chronological feeds did reduce attention inequality, meaning fewer posts got buried by the algorithm. That sounds good, right? But there was a catch: extreme content floated to the top more easily. So instead of endless outrage buried under “trending” algorithms, you had outrage plastered right in front of everyone.

The Same Old Patterns

The depressing takeaway was that, with or without interventions, the bot society drifted toward the same dynamics we see in real platforms: polarization, toxic clusters, and a massive inequality of attention. A tiny minority of posts dominated the network, while most content went unseen.

In other words, the experiment confirmed what many users already sensesocial media doesn’t just happen to get toxic; the system itself bends in that direction. And tinkering with surfacelevel fixes like chronological feeds or removing follower counts isn’t enough to reverse it.

Echo Chambers by Design

Törnberg explained it well: it’s not only about which individual posts are toxic. The very structure of the networkwho interacts with whom, how groups form, and how feedback loops reinforce existing opinionsshapes the whole environment. Once those structures emerge, the system naturally rewards the most provocative voices.

Think about it like this: if you drop a bunch of bots into a virtual town and ask them to start chatting, they’ll quickly find their tribes, stick to those tribes, and reward the loudest, most extreme members. That, unfortunately, looks an awful lot like what human users do too.

Generative AI Just Makes It Worse

Here’s where things get even darker. Right now, humans are still producing most of the content on social platforms, even if some posts are botamplified. But as generative AI tools get cheaper and better, we’re likely heading toward a flood of machinewritten content designed purely to grab attention.

Törnberg pointed out that this is already happening on X (Twitter): AI accounts pumping out endless streams of polarized takes, conspiracyladen “news,” and manipulative hotbutton posts. Why? Because outrage equals clicks, and clicks equal money. If that’s the business model, then AI is like pouring gasoline on a fire.

He put it bluntly: “I have a hard time seeing the conventional social media models surviving that.”

The Bitter Irony

What’s fascinating, and a little depressing, is that this entire experiment was supposed to explore how we might “fix” social media. Instead, it highlighted how stubborn the underlying problems are. Even a network with no real humansno history, no grudges, no bad faith actorsstill devolved into toxicity.

That suggests the issue might not just be human flaws like greed or anger. It might be baked into the very architecture of online networks themselves. Once you build a system designed to amplify content, reward attention, and keep users hooked, polarization isn’t a glitch. It’s the default outcome.

Can Anything Be Done?

So where does that leave us? If bots act like humans and humans act like bots, what hope is there for building healthier platforms? Törnberg himself doesn’t offer a silver bullet. He acknowledges that AI isn’t a perfect model of societybiases and limitations are still therebut it does reveal how alarmingly plausible these dynamics are.

Some optimists might argue that we just need smarter regulations, or better algorithms, or stronger community moderation. Those things help, sure, but this experiment suggests they may only chip at the edges. The deeper problem is that attention itself is the currency of these platforms, and attention tends to flow toward extremes.

Unless that incentive structure changes, every new fix risks becoming another bandaid.

The Future Looks… Complicated

It’s tempting to shrug and say social media is doomed, but that feels a little too fatalistic. After all, human societies have always had echo chamberscoffeehouses in the 1700s were notorious for gossip and political conspiracies. What’s different now is the speed, scale, and automation of it all. AI doesn’t just mimic human polarization; it accelerates it.

So maybe the challenge isn’t eliminating echo chambers altogether (probably impossible), but figuring out how to make them less destructive. Smaller platforms, niche communities, stricter moderationthese may end up being the lifeboats while the big platforms sink under their own toxicity.

Final Thoughts

The Amsterdam experiment didn’t solve social media’s problems. If anything, it underscored how intractable they are. A world full of bots, it turns out, isn’t that different from a world full of us. And that’s a sobering thought.

Still, there’s value in knowing this. If the structure itself is flawed, then maybe we stop expecting superficial tweaks to save us. Maybe the answer isn’t chronological feeds or hidden follower counts, but a complete rethinking of what online community should even look like.

Until then, we’ll keep logging in, scrolling past the outrage, and wondering if the bots are really all that different from the rest of us.

Open Your Mind !!!

Source: Flipboard