Jim Bootstrapping a computational platform with domain knowledge (seeding with insights), was already done a few years ago by the ex head of AI research in France. I need to find his blogs again, but apparently he had amazing results with regards re-solving classical mathematical problems.
Our question is; would that constitute AGI? I appreciate your comment on how such an approach would not be considered radical at all. However, the claim you make immediately thereafter; that the approach would help to think of the problem in a different way, is refutable. The thinking in terms of relationships suffer the same fate. Not radical, and not thinking in a new or different way. As such, we need to think as radically as we could possibly do. We need to find a few radical approaches and see if they could be focused on a few avenues of pragmatic research. May the best approach win. For example, instead of relationships, thinking free-will (random) associations. This is not a semantic ploy, but a radical departure in terms of AGI architecture. Furthermore, instead of thinking of seeding, rather allowing the computational platform to Find, Frame, Make and Share. This would denote another radical departure in current thinking (I did come across a similar approach recently). Rob ________________________________ From: Jim Bromer via AGI <agi@agi.topicbox.com> Sent: Wednesday, 12 September 2018 2:25 PM To: a...@listbox.com Subject: [agi] Growing Knowledge The idea that an AGI program has to be able to 'grow' knowledge is not conceptually radical but the use of the idea that a program might be seeded with certain kinds of insights does make me think about the problem in a slightly different way. By developing a program to work along principles that are meant to incorporate some way to build on the basis of insights that are provided as the program explores different kinds of subjects I think I might be able to see this theory in the terms of a transition from programming discrete instructions that correspond to a particular sequence of computer operations into programming with instructions that have a potential to grow relationships between the knowledge data. The kinds of relationships do not need to be absolutely pre-determined because the use of basic relationships and references to specific ideas can implicitly develop into more sophisticated relationships that would only need to be recognized. For example, an abstraction of generalization seems pretty fundamental to Old AI. However, I believe that just by using more basic relationships which can refer to other specific ideas and to groups of ideas, the relationships that will effectively refer to a kind of abstraction may develop naturally - in primitive forms. It would be necessary to 'teach' the AGI program to recognize and appreciate these abstractions so that it could then use abstraction more explicitly. Jim Bromer ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T032c6a46f393dbd9-M44ff28f8a47bd56696724c2f Delivery options: https://agi.topicbox.com/groups/agi/subscription
