I guess the approach of extracting all that information out ahead of time reminds me of the old approach to making robots walk. The robot would take a step and then spend hours performing calculus operations on a model of the robot's leg to determine how the next step should proceed. The newer approach takes a simpler approach of letting the leg itself be its own model, and making real-time corrections to its movement as the robot recognizes it going the wrong direction. It's this on-the-fly correction combined with letting the system be its own model that I see as a good way to deal with impossible-seeming computational problems like determining exactly what everyone wants ahead of time. Besides, what if someone changes their mind? I'm sure that, since we're dealing with 7 billion dynamical systems whose state is perpetually changing, most of us won't even know for ourselves what we want in advance.
On Fri, Aug 24, 2012 at 4:31 PM, Aaron Hosford <[email protected]> wrote: > I meant literally, "Make people happy," in those words, maybe with an, > "Ask people what makes them happy," tacked on or hardcoded in. If the > system truly understands natural language and is even quasi-intelligent, > then we can use the system's own intelligence to derive the complicated > messy details as to what makes people happy by having the system ask people > for their preferences and actually listen to and think about what they tell > it -- especially if they take the time to correct it because it messed up > and did the wrong thing. When you raise a child, you teach the child what's > expected of it. Children don't come with that built in. They have to be > taught. The difference is, children have value in and of themselves, > and they have many goals in addition to making Mom or Dad happy. This > system would only have value derived from its service to humanity (meaning > it doesn't itself deserve moral consideration, since it's just a tool) and > it's one and only goal would be to make its creators happy, in whatever way > they define that. > > > > On Fri, Aug 24, 2012 at 4:21 PM, Matt Mahoney <[email protected]>wrote: > >> On Fri, Aug 24, 2012 at 2:24 PM, Aaron Hosford <[email protected]> >> wrote: >> > >> > So why wouldn't we design a system that attempts to attain a nice simple >> > goal like "make people happy" and build in the awareness that in order >> to >> > define that goal in all its complexity, it needs to *ask* us what we >> want. >> >> Because that's not a shortcut. The goal "make people happy" is not >> nice and simple. It is 10^17 bits, unless you mean make people happy >> by giving them drugs or inserting an electrode into the nucleus >> accumbens. >> >> > Then the system iteratively refines that goal as new information comes >> in at >> > the measly rate of "1 to 5 bits per second through >> > speech, writing, or typing", as time is available and the need arises, >> > making do with a less individualized but still highly effective >> definition >> > of the general goal in the meantime. People recognize the value of >> > information vs. the time it takes to communicate it, and will point out >> the >> > most inconvenient misunderstandings first, so the system can rely on the >> > users to selectively identify and convey the information it needs to >> know in >> > order to meet their needs. In other words, if you want the system to be >> > individualized to your preferences, you pay the cost of gathering & >> > transmitting a description of your preferences. This is the current >> model >> > for all those apps you mention: you go to the preferences page and >> check the >> > boxes according to what you prefer. In the future, it will be >> communicated >> > via natural language, but it will be the same principle at work. >> >> I thought we were already doing that. But yes, the cost of >> communicating our preferences will be the most expensive part of AGI >> once Moore's Law makes the hardware cheap enough. (Right now, it would >> cost $1 quintillion if you could buy it. In 15 years the same >> computing power should cost $1 quadrillion, low enough to make it cost >> effective to replace most human labor). Natural language is better >> than filling out an online survey. Observing your behavior is better >> still. Guessing based on the preferences of other people with similar >> behavior is better still. We already do all of these things because it >> is so expensive. >> >> >> -- Matt Mahoney, [email protected] >> >> >> ------------------------------------------- >> AGI >> Archives: https://www.listbox.com/member/archive/303/=now >> RSS Feed: >> https://www.listbox.com/member/archive/rss/303/23050605-bcb45fb4 >> Modify Your Subscription: >> https://www.listbox.com/member/?&; >> Powered by Listbox: http://www.listbox.com >> > > ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-c97d2393 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-2484a968 Powered by Listbox: http://www.listbox.com
