The Consensus Machine

Recently a financial institution asked verification for my corporate identity. I filled form, given every bit information they ask. Literally spilled from head to toe, even they know color of my underwear now.

Still, unable to pass the verification at the second attempt and i reached to support desk. Had a nice chat with guy and noticed lack of legal knowledge in his questions. He's asking for a very specific document which i don't possess due to citizenship status but i already sent another document that legally covers the information he asked.

I explained him how things work in delaware and meanwhile pasted same sentence to chatgpt in another window to validate my statement. He's didn't convinced. I responded with screenshot from chatgpt also had to give exact same prompt i typed to confirm on his side then went to sleep. When i wake up received 4 e-mails from them, last one was saying

Please disregard previous messages, your account is now ready to use

Similar situations happened in my entire life, chatgpt wasn't there back then. In a software contract, client literally asked update to his operating system. I was giving wikipedia link to convince him that isn't possible. Another client was asking a complete rational and reasonable request yet not happy with the terms of delivery date. So i had to give another wikipedia link to prove that it's a long term project.

We're all grown with encyclopedia britannica which existed shelves of every house at 90's. Every country had a their own version, britannica had an imperial bias to shiny out how big the british empire is, soviet translation was like only russians exists in the planet. German editions stripped away last 100 years like never happened. 99% of wikipedia content written by regular people, based on knowledge that learned from those books. Large language models including chatgpt rely on that data while composing their responses hence it's biggest dataset available on internet.

It doesn't matter that given source of knowledge credible or not. There is an evolutionary bug in that human brain, it always takes shortcuts as a survival mechanism. Smoke? something red? fire! run! If we had to calculate everything in the jungle we're eradicated already. A mouse can take sharp turns while encounter with a big predator due to it's small and low-latency nervous system. Humans can't. We subconsciously look for signs instead, symmetrical face? big breasts? marry then. Cool name? street cred? give your money.

Nowadays chatgpt exactly put itself into that position. It have definitive truth because people using it, no matter the warning under the input box says. Ruptured tendons? ask chatgpt. Is your wife cheating? ask chatgpt. Is your team refactoring entire codebase to rust? because chatgpt said so. Same situation happened with bitcoin at 10 years ago, coins literally made up from thin air just by wasting electricity to solve a cryptographic puzzle created by an anonymous developer and became extremely valuable just because people interested in it.

There is two kind of information, the one that fetched and that one that acquired. The traditional education systems encouraging people to fetch information. Read many books as you can, always appeal to authority. Stop and wait on red light, no matter if the road empty or not. Chatgpt, ai, llm whatever you call it just byproduct of that mindset in this era. Ten years later maybe another ai product came in that predicts the future and people would say

ooh we should stay at home, machine says a traffic accident going to happen

and some other guy write blog entry about our fate is not in the hands of machines.