World in conflict unlimited reinforcement points

#WORLD IN CONFLICT UNLIMITED REINFORCEMENT POINTS HOW TO#

Get this episode by subscribing to our podcast on the worldвЂ™s most pressing problems and how to solve them: type 80,000 Hours into your podcasting app.

What are the most urgent research questions in this area?.

Should we develop AIs to have their own ethical agenda?.

What are the arguments against being concerned about AI?.

In todayвЂ™s conversation we cover, among many other things: Without having to work just to survive, people might flourish in ways they never have before. If cheap thinking machines can one day do most of the work people do now, it could dramatically raise everyoneвЂ™s standard of living, like a second industrial revolution. When each of us can have an AI of our own - one smarter than any person - how do we resolve conflicts between people and their AI agents? How considerate of other peopleвЂ™s interests do we expect AIs to be? How do we avoid them being used in malicious or anti-social ways?ДЄnd if AIs end up doing most work that people do today, how can humans avoid becoming enfeebled, like lazy children tended to by machines, but not intellectually developed enough to know what they really want?Дespite all these problems, the rewards of success could be enormous. The harder problems may end up being social and political. Stuart thinks all of these problems are surmountable, if we put in the work.

And they need to guess when weвЂ™ve rejected an option because weвЂ™ve considered it and decided itвЂ™s a bad idea, and when we simply havenвЂ™t thought about it at all. Among other things, the resulting AIs need to be able to interpret what people really mean to say based on the context of a situation. WeвЂ™ve made progress toward putting these principles into practice, but the remaining engineering problems are substantial. These principles lend themselves towards machines that are modest and cautious, and check in when they arenвЂ™t confident theyвЂ™re truly achieving what we want. As Stuart says, вЂњyou canвЂ™t fetch the coffee if youвЂ™re dead.вЂќ

Stuart thinks this design architecture, if implemented, would be a big step forward towards reliably beneficial AI.ДЇor instance, a machine built on these principles would be happy to be turned off if thatвЂ™s what its owner thought was best, while one built on the standard model should resist being turned off because being deactivated prevents it from achieving its goal. And it figures out what we want by observing our behaviour.

But the system isnвЂ™t sure what we want.The AI systemвЂ™s objective is to achieve what humans want.According to him we need to redesign AI around 3 principles: Stuart isnвЂ™t just dissatisfied with the current model though, he has a specific solution. If weвЂ™re ever to hand over much of the economy to thinking machines, we canвЂ™t count on ourselves correctly saying exactly what we want the AI to do every time. This вЂalignmentвЂ™ problem will get more and more severe as machine learning is embedded in more and more places: recommending us news, operating power grids, deciding prison sentences, doing surgery, and fighting wars. Like King Midas, who asked to be able to turn everything into gold but ended up unable to eat, we get too much of what weвЂ™ve asked for. This isnвЂ™t something we wanted, but it helped achieve the algorithmвЂ™s objective: maximise viewing time. Stuart points to the example of YouTubeвЂ™s recommender algorithm, which reportedly nudged users towards extreme political views because that made it easier to keep them on the site. We already see AIs misbehaving for this reason. ThatвЂ™s true even if the goal isnвЂ™t what we really want, or the methods itвЂ™s choosing are ones we would never accept. AI today lacks common sense, and simply does whatever weвЂ™ve asked it to. Unfortunately thereвЂ™s a big problem with this approach: itвЂ™s incredibly hard to say exactly what you want. This is so obvious it almost doesnвЂ™t even seem like a design choice, but it is. In his new book, Human Compatible, he outlines the вЂstandard modelвЂ™ of AI development, in which intelligence is measured as the ability to achieve some definite, completely-known objective that weвЂ™ve stated explicitly. Stuart Russell, Professor at UC Berkeley and co-author of the most popular AI textbook, thinks the way we approach machine learning today is fundamentally flawed. So itвЂ™s kind of like saying youвЂ™ve come across a terrible accident and you say, вЂњWell, no one should call an ambulance because someoneвЂ™s going to call an ambulanceвЂќ. вЂ¦if no oneвЂ™s allowed to talk about the problems, then no one is going to fix them.

World in conflict unlimited reinforcement points