I think goal selection is an important issue for AGI. Human decisions are often influenced by coincidental occurrences, but I think that the stronger goals and goals that are more feasible are going to be less influenced by coincidental occurrences. The lessons acquired through coincidence are going to tend to be more relevant to a problem when a person is working on a problem that he is able to solve and that he is motivated to work on then when he is working on something that he does not understand well or when he is not highly motivated to work on it. But we have to be free to explore alternative paths as well. When we become aware of something unexpected we may become more curious about the new awareness so we may become motivated to explore that area more carefully after learning something new about it. Many goal selections can be made according to motivation but the motivation for a simple AGI program would tend to be based on knowledge that it had already learned and on choosing actions that the program could take given some knowledge that it had. An AGI program has to be working on many different levels at once, even though it might not be focusing on all the levels at the same time. Two basic tasks for a simple AGI program are to learn and to express what it think it knows. However, these broad tasks (or goals) can be broken down into different kinds of ways and means to achieve the goal. For a program that is capable of learning and then build subsequent knowledge on what it had previously learned, those ways and means are more than just sub-goals. They are actions that can be combined from which abstractions and generalizations can be derived. Since these strings (or possibly fields) of basic actions can be associated with kinds of knowledge they can become differentiated. Once you think about it is obvious that we have to have different kinds of ways to work with different subject matters. So this association between combinations of behaviors (or procedures) and subject matter must be made in order to derive working strategies to deal with different situations. So here goal selection comes from knowledge of effective strategies that are associated with particular kinds of situations or subjects. Jim Bromer From: [email protected] To: [email protected] Subject: [agi] RE: Goal Selection Date: Mon, 13 May 2013 14:04:29 -0700
I've been thinking about Goal Selection, preferences and priorities along these lines: http://www.ilovephilosophy.com/viewtopic.php?f=9&t=143153 My burning philosophical question of the day is why do we choose the goals we do at any particular moment? I can account for action selection. That's easy. We have a goal and we need to achieve the goal. So we pickactions that we believe help us to achieve the goal. But in the myriad of possible goals to choose, how do we hone in on the goals we do? My personal distinction is that Knowledge is action selection, whereas Wisdom is goal selection. Also, How do we represent goal rejection? Given an agenda, and a collection of current situations, how do we know when to reject certain goals, like theft as a means for attaining an object? Your thoughts? ~PM AGI | Archives | Modify Your Subscription ------------------------------------------- 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
