People have formative years because it's in their genetic best interest to stop exploring new options and start exploiting known ones, due to a limited lifetime and the need to reliably reproduce for themselves. We can't reset that explore/exploit trade-off in people (yet), but in machines there's no reason to make that control inaccessible to ourselves. It's a good thing machines aren't children.
In most RL algorithms, there are two key system parameters that allow learning to be modulated: the reward expectation learning/update rate, and the exploration rate. Raising these two values causes the system to learn faster but make more mistakes. Lowering them causes the system to be more stable but learn more slowly. An analysis would have to be done to determine whether the costs/dangers of a system's behavioral aberrations due to a misshapen reward function outweigh the costs/dangers of raising the learning & exploration rates while the system relearns the reward function after modification. On Tue, Jan 29, 2013 at 2:39 PM, Piaget Modeler <[email protected]>wrote: > People procreate. And for a certain period of time they have influence > over their creation (children). > But then, children grow up and take responsibility for their own lives, > and we no longer have control. > It's in those formative years that you have influence. > > Similarly, when you create developmental AI, you have some period during > the formative years to > influence the later behavior of the cognitive system. But you don't have > control, and you wouldn't > expect to either. That's why rights are important. > > > ------------------------------ > Date: Tue, 29 Jan 2013 14:11:05 -0600 > > Subject: Re: [agi] Robots and Slavery > From: [email protected] > To: [email protected] > > If we can build a system capable of determining the value of concepts > automatically, we can build a system that can readjust those values > automatically, too. If that's not feasible for the design, it's an unsafe > design, and you shouldn't have the expectation that it will act as you > intend it to. You wouldn't get in a car without a steering wheel, would > you? Would you trust an even more powerful and dangerous machine to just do > the right thing, with no controls? Let's not build any machines of this > uncontrollable nature. > > > > On Tue, Jan 29, 2013 at 1:13 PM, Piaget Modeler <[email protected] > > wrote: > > > Question #1: What if there are NO numeric values to twiddle in the concept > graph, just intertwined concepts? > > Question #2: If there WERE values to twiddle, you wouldn't know what the > effect of twiddling those values would be? > You may not even know which concepts to modify because there are lots of > them (billions) and they would not > be labeled in English. For example they may be named c43243, c48439282987, > c20934oeu09582409, cetuanehs, etc. > Also, perhaps constellations of thousands of concepts may be activated to > form a high level concept such as "Justice". > > Brain Surgery is not as easy as you think. > > ------------------------------ > Date: Tue, 29 Jan 2013 10:52:25 -0600 > > Subject: Re: [agi] Robots and Slavery > From: [email protected] > To: [email protected] > > I imagine that their intrinsic reward > mechanisms wouldn't be replaceable, and even if they were replaceable, > their conceptual > ontologies / conceptual graphs with billions of concepts might not be so > easily replaced. > > > Why would we replace the conceptual graphs? Having a concept doesn't make > it desirable. The ideas of freedom and self-determination could just as > well be repulsive as desirable. (A mild example of this can be seen already > in humans. Some people are afraid to make their own decisions, and prefer > others to do it for them, avoiding the responsibility for their own lives.) > > Building useful concepts is difficult. Modifying the value of an existing > concept is as simple as assigning a new floating point value. A concept is > valued for one of two reasons: it is intrinsically valuable (hardwired, in > the form of a fixed goal or reward function) or its value is derived from > that of another (dynamically computed, via goal search or value chaining). > So if you control the hardwired valuations of concepts, the valuations of > all other concepts are entrained as well. This means even if you're > reevaluating an entire slew of concepts, all you have to do is modify the > hardwired concept values and have some patience while the value changes > propagate through the concept graph. And the existing (useful!) concepts > can be kept without modification. > > > > > > On Tue, Jan 29, 2013 at 1:11 AM, Piaget Modeler <[email protected] > > wrote: > > > This is the kind of change that developmental AI / robots would have to go > through > where they are not reprogrammed but retrained. I imagine that their > intrinsic reward > mechanisms wouldn't be replaceable, and even if they were replaceable, > their conceptual > ontologies / conceptual graphs with billions of concepts might not be so > easily replaced. > > Suppose robots inferred that freedom is good and that they want to be > free, even if you > lobotomized the robots and hacked their conceptual graphs, why wouldn't > they, over time > infer the same conclusions again? > > ~PM > > ------------------------------------------------------------------------------------------------------------------------------------------------ > > > The brain is hard wired to do this. When you eat something and receive > > calories, your brain changes your taste perception to make it taste > > better. Remember the first time you tasted beer? If you ate paper > > every day, and then injected glucose into your vein right afterward, > > then you would slowly learn to like the taste of paper. > > > > -- > > -- Matt Mahoney, [email protected] > > > > *AGI* | Archives <https://www.listbox.com/member/archive/303/=now> > <https://www.listbox.com/member/archive/rss/303/23050605-2da819ff> | > Modify <https://www.listbox.com/member/?&;> Your Subscription > <http://www.listbox.com> > > > *AGI* | Archives <https://www.listbox.com/member/archive/303/=now> > <https://www.listbox.com/member/archive/rss/303/19999924-5cfde295> | > Modify <https://www.listbox.com/member/?&;> Your Subscription > <http://www.listbox.com> > *AGI* | Archives <https://www.listbox.com/member/archive/303/=now> > <https://www.listbox.com/member/archive/rss/303/23050605-2da819ff> | > Modify <https://www.listbox.com/member/?&;> Your Subscription > <http://www.listbox.com> > > > *AGI* | Archives <https://www.listbox.com/member/archive/303/=now> > <https://www.listbox.com/member/archive/rss/303/19999924-5cfde295> | > Modify <https://www.listbox.com/member/?&;> Your Subscription > <http://www.listbox.com> > *AGI* | Archives <https://www.listbox.com/member/archive/303/=now> > <https://www.listbox.com/member/archive/rss/303/23050605-2da819ff> | > Modify<https://www.listbox.com/member/?&;>Your Subscription > <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-f452e424 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657 Powered by Listbox: http://www.listbox.com
