I'm not exactly clear what your criticism of the idea is based on. What exactly do you mean by "inner direction", and why is it that RL can't produce it together with appropriate concepts? (I would include episodic memory in the toolset, too, btw. But we've been using "concept" in a very generic way, to include concept activation, too, so maybe that's already a given.)
This is the same problem as insisting that Bayesian Reasoning along with > some priors are all that is necessary to explain human intelligence. I don't see the comparison. RL is a means for selecting appropriate responses to a situation. It is not a means for understanding that situation. That depends on concept formation, or in Bayesian terms, identification of the correct priors, which is a non-trivial task. So I'm not claiming that RL produces intelligence by itself. In fact, what I said was that without RL or some other goal-directed behavior system intentionally built into it, a program may be intelligent, but won't have desires or motivations. So, for clarity: - A system may be intelligent with or without motivations, desires, or goal-directed behavior. This is strictly observational intelligence, and requires a technique for concept formation, which is not implemented by RL or other goal-directed behavioral mechanisms. - In order for an intelligent system be more than merely observational, behaving in a goal-directed manner, it must have RL or some other goal-directed behavioral mechanism. This is not sufficient for intelligence, but rather is complementary to it and necessary for intelligent *behavior*. So... This is interesting. Having written this, I look back and realize the above explanation contains an implicit definition of intelligence: *Intelligence is the ability to accurately represent important aspects of the environment in terms of useful concepts* (whether the aforementioned importance or usefulness is due to the ability to predict outcomes, recognize/classify situations according to similarity, accurately describe the environment, provide the conceptual information necessary to seek a goal, etc.). *Intelligent behavior, on the other hand, is the effective utilization of useful concepts towards achieving goals.* As defined here, intelligence includes understanding but excludes the will, whereas intelligent behavior includes the will but excludes understanding and intelligence itself. It is important to point out that goals may only be expressible in terms of sufficiently useful concepts, in which case a sufficiently high level of intelligence is necessary before the will can be fully determined -- so in some cases the will is undefined without intelligence, but not vice versa. It should also be clear that in order for intelligence as defined above to be measured (in terms of the usefulness of the concepts it generates), there must be some sort of external optimization process (i.e. goal seeking or reward maximization) or a priori value function through which concepts can be assigned a level of utility. On Tue, Jan 29, 2013 at 7:54 AM, Jim Bromer <[email protected]> wrote: > On Mon, Jan 28, 2013 at 6:21 PM, Aaron Hosford <[email protected]>wrote: > >> In regards to the idea that intrinsic rewards are somehow different from >> extrinsic ones, a reward signal can just as easily be modulated by internal >> events (thoughts) as external ones (percepts). Furthermore, if you read up >> on RL, you'll see that in all effective multi-step RL-style algorithms, >> there is a backward chaining of reward, so that previous behaviors or other >> early triggers for a behavior are rewarded, not just the immediate actions. >> All actions, whether extrinsically or intrinsically rewarding, derive their >> value from either immediate or indirect/backward-chained reward signals, >> which means we can modulate behavior arbitrarily to any level of complexity >> with relatively minimal difficulty by taking advantage of this backward >> chaining. >> > Well the fact that backwards chaining of the actions leading up to a > rewarded behavior is an interesting point. And while anyone with a little > imagination could come up with a creative means to develop a way to use RL > to reinforce complex behaviors based on parts of a behavior string that is > reinforced this is not explained by the backward-chained reward signals > that you mentioned. > But looking beyond that the claim that any internal motivation could be > explained by external reinforcement is unnecessarily complicated because it > is dependent on external rewards which would demand that things like the > massive levels of complexity of infinitesimal past rewards could explain > inner direction. This is the same problem as insisting that Bayesian > Reasoning along with some priors are all that is necessary to explain human > intelligence. Sorry but it just does not work - unless you change the > presumptions of what is meant by Reinforcement Learning or Bayesian > Reasoning. (Which is ok, I am just saying...) > Jim Bromer > *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
