Post

Hey y'all! Sr. DevOps Eng / SRE here, let's talk about the Russian bot issue! You're probably hearing a lot of conflicting narratives on this, and I'm here to tell you they're not mutually exclusive! Both search crawlers and influence bots have been destroyed and degraded!

Let's start with the basics: Yandex is essentially the Google + AWS in Russia. Both a search engine and a public cloud provider in one company. Ukraine blew up several of their datacenters (we call them "availability zones") which dramatically disrupted a myriad of different industries using them.

Typically, when we architect anything important (think: banking, medical records, safety sensitive systems) we design our application to run in multiple regions at once. A region is typically comprised of at least 3 availability zones.

A common pattern for an international company, operating across regulatory jurisdictions, is 2 regions in each country. Theoretically, this would take the destruction of 12 datacenters across multiple countries, to fully knock out.

But in practice, it often doesn't work like that. Regulatory rules like GDPR increasingly force companies to story user data in the jurisdiction they reside in. In Russia's case, few hosting providers at the scale of Yandex/AWS will willingly allow them to host botnets outside the country.

It's typically against the terms of service of major hosting providers to do so, and you'll be kicked off the platform with enough complaints. So what about the "it's all just search crawlers" claim Yandex is making?

This is, IMO, a total red herring by Russia to muddy the waters, and the only reason it's getting traction is because of a lack of technical understanding by the American media.

You can take out 1,000,000 search crawlers for every one political bot. If that same strike takes out 100 political influence bots, it's still an effective strike. 100 instances of an LLM-based chatbot can post thousands of times per hour, exerting significant influence over online sentiment.

The absolute number may be less impressive than the initial headlines by raw numbers, but the effect is the same. I don't think we should write off the effects of these kinetic strikes on public sentiment.

But how do we measure it? Social media companies stand to loose a lot of (fake, but still profitable) engagement on their platforms, and will likely be very tight lipped about the metrics they've collected around this incident. Getting an accurate big-picture sense won't be easy.