Thread by @chrisnoessel

How powerful could #ASI become? Note that when an AI gets better than a human at something it gets much, much, much better. (Also note the terrifying scenario of an AGI/ASI running 1:1 for office.) @MIRIBerkeley #sketchnotes

A favorite concept from the book is that “we already have something that does everything that humans can do, which is humans.” But the charming _bon mot_ is a disservice to the potential at-scale effects of ANI.

Eventually ANI may be able to do everything that humans can do, but do it faster, tirelessly, and cheaper than humans. We are going to have to get psych, civics, and law ahead of the potential displacement of humans. Some good news is that ANI will always need human partners.

In Chapter 3 of #SmarterThanUs, Armstrong makes the case that a capabilities-measurement of #ASI is more useful than a philosophical one. (Out of order because I reposted with corrected attribution for Marcus Hutter of @anucecs. Hat tip to @robbensinger.)

Chapter 5 speaks of the trappings of communicating human goals to a digital intelligence. #sketchnotes #SmarterThanUs

In Chapter 6 of #SmarterThanUs, Armstrong reminds us that the implicit “reasonableness” of our values are complex, and they are fragile. And as AGI gets smarter, things get worse. #sketchnotes #AI

Filing under “things I’m glad I have gotten to sketchnote.”

With the command, “Get my mom out of that building!” this chapter illustrates the problem of literal interpreters, i.e. the amount of specification needed to ensure reasonable results is unreasonable. (This builds on @ESYudkowsky’s lesswrong.com/posts/4ARaTpNX…) #sketchnotes #ASI #AI

Chapter 8 asks if we have to get everything about #ASI exactly right. Short answer: Yes. Because it will be so much more aware and powerful than us, we need it to be that much more moral than us. #sketchnotes #SmarterThanUs

Chapter 9 of “Smarter Than Us” reminds the readers that philosophers and computer scientists have been struggling with these issues for a LONG time. #sketchnotes #sandwich #philosophy

Chapter 10 is pointers to many folks doing the hard work, getting us to a safe seed #AI. @CSERCambridge @MIRI @FHIOxford (Others off twitter…) (But it’s not the last chapter.) #sketchnotes