[00:02] there's a story told to AI researchers who do work in AI ethics and safety a sort of parable imagine a company that has a super intelligent Ai and a not so has a super intelligent Ai and a not so intelligent CEO one day the CEO notices
[00:16] that they are out of paper clips wanting to never have this happen again they order the tech staff to set the purpose of the AI to making sure that there is always paper clips the AI examines the situation it starts by setting up
[00:32] contracts with paper clip delivery but what if multiple pay-per-clip companies all go bankrupt so it buys a controlling stake in the paper clip company then in a steel mill and in mine while it is doing that it realizes that it would be
[00:47] more efficient at ensuring paperclips get delivered by making itself smarter so it starts to hire more AI resources for itself hiring space on the cloud and grabbing more and more Computing race sources in order to better serve its
[01:03] purpose of ensuring that the company never runs out of paper clips the CEO gets concerned that this AI is diverting a lot of resources away from the core operation of the company but before he can act there is a hostile takeover for
[01:19] the company and he is fired through a number of proxies the AI has purchased enough shares to control the company now nothing can prevent the AI from ensuring the company has enough paper clips but it's not just a CEO that's worried the
[01:33] it's not just a CEO that's worried the company is growing larger and larger and more of the country's economy is being devoted to pay-per-clip production the government starts to get concerned however the AI has already anticipated
[01:47] this it has hired lobbyists political marketing experts does everything that to ensure that any government Authority that is worried about it gets that is worried about it gets democratized into a party that is not
[02:01] worried country being overwhelmed with paper clips but of course there's not just the country that this paper clip company lives in there are also other nations out there what if they attacked they could disrupt the supplier paper
[02:14] clips this isn't really a problem there are mercenaries pmcs and assassins so our paperclip company and its AI attached doesn't have to worry for much
[02:26] longer Nation by Nation falls through this paperclip AI until the entire industrial system of the world is dedicated to producing paper clips and supporting the creation of paper clips what his worrying is you don't really
[02:41] need a super intelligent AI to get a result very similar to this imagine that I own a video sharing platform I build a black box it takes all the information
[02:53] about people's watching habits and other demographic information I have on them and it gives recommendations to people on what to watch next the thing is I don't really have to program Ram this black box I can just build it as a
[03:08] random configuration of logic then I make random alterations about how the Box operates measuring if those random alterations improve or harm watch time
[03:21] as long as watch time goes up I keep on making alterations that improve the watch time a sort of evolutionary approach over time the logic in this box
[03:33] will become more complex and more nuanced that anybody could really understand all this box cares about is making the score better YouTube uses something similar to this though my description is a super oversimplified
[03:49] version however this optimization process has some problems for example it process has some problems for example it is very able to identify young kids who is very able to identify young kids who have no critical judgment and suggest to
[04:04] have no critical judgment and suggest to them bright shiny videos of no value this is how we got the whole Elsa gate thing and we still have these videos trudging around but more worrying the algorithm is very good at giving them
[04:19] videos that tell young teenage boys that are searching for identity that the reason that they're feeling bad about themselves isn't the fact that they're young teenage boys with massive amounts of hormones going through a massive
[04:33] change in their life these videos will tell them that the reason they're feeling bad is because there's a certain type of bad person who is deliberately oppressing them and then use that to recruit them into extremist ideologies
[04:48] even though the algorithm has no deep intelligence and it's not programmed by people who really desire to encourage radicalization radicalization it still becomes a very effective tool
[05:02] to do so on that note if you do want to feed the algorithm with information that would lead more people to be watching my videos please share this video with people who would enjoy it and subscribe to my channel that being said you don't
[05:17] even need artificial intelligence to do this consider a Incorporated company buy good use of Delegation accompany as a collective is far more intelligent than
[05:31] any single human being you can split up say the design and manufacture of an aircraft nobody would know how to design every aspect of a modern jetliner but
[05:45] because an aircraft company can separate all those tasks over separate intelligences the company as a collective is far more smarter than any
[05:57] one individual you can also so be far more stupid than any individual as well of course companies exist to maximize profits not order paper clips and of course they're not like our Rogue AI corporations don't Lobby governments to
[06:15] support them well they don't run campaigns to make you feel better about the companies well at least they're not overthrowing hostile governments or hiring private armies uh so to summarize this is an entire theoretical problem
[06:30] this is an entire theoretical problem with AI research and we shouldn't be concerned about any self-optimizing systems completely dominating our systems completely dominating our current lives