It sounds like your design is a slightly finer-grained decomposition than my own, with the goal-based portion consisting of the observers and reflectors taken together and the understanding-based portion consisting of the coordinators together with the model itself.
On Wed, Mar 12, 2014 at 11:08 AM, Piaget Modeler <[email protected]>wrote: > Perhaps I misunderstood what you meant by physical separation. > > I took separation to mean 'disconnected' in the sense that at times > there would be no communication between the two components. > > In PAM-P2 (diagrams attached) we have three categories of agents: > observers (that interact with the outside world), coordinators (that > build the model through inferencing), and reflectors (which attend > to goal creation, goal satisfaction, learning, and equilibration). > > We've created a language called Premise in which the agents are > written and through which they perform their respective tasks. All > the agents share a global memory. This is one way to decompose > a cognitive architecture. Whether there is any intelligence in it remains > to be seen, but we are hopeful. > > Cheers, > > ~PM > > > ------------------------------ > Date: Wed, 12 Mar 2014 08:55:56 -0500 > Subject: Re: [agi] Decomposition of Intelligence > From: [email protected] > To: [email protected] > > > That is interesting, and I would take it as validation of the sanity > (meta-application!) of at least the fundamental concept of decomposition of > intelligence into modeling and goal-seeking. However, I don't come to the > same conclusion as yourself. > > Physical separation does not imply functional independence; the modeling > system would necessarily validate its sanity with information received from > the goal-seeking agents. Part of its algorithm could well be to slow or > even stop its activities in the case of periods of reduced validation > opportunities. Also, with separation comes the ability to have multiple > agents sharing the same model but having different goals, including some > whose sole purpose might be to ask questions and otherwise seek information > to validate the model's conclusions. > > Additionally, with the separation of goals, it would be possible to assign > some agents to the sole task of improving the system's design, while all > others are designed to prefer being strictly hands-off with respect to the > system's design -- and the agents that are tasked with improving it. This > would simplify many of the complicated feedback loops that could produce > ill-defined behavior as discussed in other threads relating to the friendly > AI problem. > > > > On Wed, Mar 12, 2014 at 12:45 AM, Piaget Modeler < > [email protected]> wrote: > > Heard an interesting story on this some months ago. It turns out that > people form a mental model of their world. And they continually > validate their model by asking others questions. This model validation > helps them to maintain "sanity" and not start thinking thoughts that > are completely out of touch with reality. Prisoners in solitary > confinement do not have the luxury of asking questions to validate > their mental models and consequently their thinking may quickly > lose correlation with "reality". > > The conclusion is that your AGI needs to have a means of continually > checking whether its hypotheses are correct. Separating sensory > input from action or behavior (particularly when one of those behaviors > is question asking) then may not be prudent. > > ~PM > > -------------- > > > From: [email protected] > > Date: Wed, 12 Mar 2014 00:37:38 -0500 > > Subject: [agi] Decomposition of Intelligence > > To: [email protected] > > > > > The concept of the "dispassionate observer" got me wondering, can > > intelligence be decomposed? I think a lot of confusion with regards to > > the definition of intelligence comes down to confusion between > > intelligent thought/understanding versus intelligent action/behavior. > > > > The understanding-based definition of intelligence: Intelligence is > > the accurate modeling of the environment through information attained > > by the senses, irrespective of any behavior taken based on that > > understanding. > > > > The behavior-based definition of intelligence: Intelligence is > > effective goal-seeking behavior within the environment, irrespective > > of any model of the environment used to determine that behavior. > > > > It seems to me that if we were to build a system capable of > > constructing an accurate model of the environment, goal-seeking > > behavior would be relatively trivial to implement on top of this. This > > suggests a possible solution to the "friendly AI" problem: Keep the > > modeling system physically separate from the goal-seeking system. In > > the event the goal seeking system goes awry, throw a kill switch that > > prevents it from accessing the modeling system. Without the capability > > for understanding, it ceases to behave intelligently, and is > > effectively contained. > > > > > > > > > > Aaron Hosford > > > > > > ------------------------------------------- > > AGI > > Archives: https://www.listbox.com/member/archive/303/=now > > RSS Feed: > https://www.listbox.com/member/archive/rss/303/19999924-4a978ccc > > 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> > <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/303/19999924-4a978ccc> | > 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
