I am not talking about growing knowledge from a domain base. I am talking about growing knowledge from interactions with human users who will provide the system with facts and relationships usually expressed in a natural (but austere) form, which the program would then have to integrate and explore. This idea is not radically new but the metaphor of 'growing knowledge' makes me think of the problem in new ways. All out radical thinking is not going to work unless it can be used to resolve some complexity. So if you have done some work and developed a novel idea that can work on resolving complexity to some new degree, then that would be worthwhile to examine more closely. The idea that some guy in France tried what I am talking about is not an assertion that can be made without a more careful comparison of the ideas. Looking as quickly as I can at Pitrat's blog, I found this: "For the realization of a General Problem Solving system, a number of meta-problems arise. The main idea of bootstrapping is that these meta-problems will be solved by the system itself, in the same way as it solves the problems for which it was designed." That is the opposite of what I had said. OK, I now see where Nanograte misunderstood me. I had used the word "seeding" and he thought that I meant that you seed the program with domain knowledge at the start. I can see how a non-gardener, non-farmer urbanite might come to that conclusion. I used to grow some stuff in a garden. You have to seed every year, and then you have to nurture the garden. You have to harvest. You have to do something to refurbish the soil. You have to come up with watering systems, you have to watch the weather, you have to cover your plants to survive a first frost, you have to figure out new ways to get better results and ways to grow stuff during the winter. I was able to grow basil all year round by keeping it near the furnace in the winter and putting it under florescent lamps that were on a timer to approximate 12 hour days. In other words my idea is that if I could write a program that would do a lot of the heavy lifting itself, I might be able to get it to function successfully by relying on the principle that human beings can guide it by giving it all sorts of facts and opinions and guesses. In my plan I would have to supply facts, but I would also have to try to help it to think. But the program would have to be able to use this guidance effectively. A person cannot teach the program to learn more effectively using contemporary AI theories. But if complexity were resolved to some extent then it would be much easier to discover how to improve the AI programs. Talking alone isn't going to get it done.
Jim Bromer On Fri, Sep 14, 2018 at 2:36 AM Nanograte Knowledge Technologies via AGI <agi@agi.topicbox.com> wrote: > > A good and fair challenge. I'll just have to find my source. > > ________________________________ > From: Basile Starynkevitch <bas...@starynkevitch.net> > Sent: Friday, 14 September 2018 5:55 AM > To: AGI > Subject: Re: [agi] Growing Knowledge > > > On Wed, 12 Sep 2018 at 15:26, Nanograte Knowledge Technologies via AGI > <agi@agi.topicbox.com> wrote: > > Evn with all its bureaucracy, France has not (officially) and never had a > single head of AI research. > > > However, regarding bootstrapping AI, I guess you are refering to Jacques > Pitrat. He is a pionneer of French AI, is a retired academics (he probably > was born just before WW2) and was a top-level Directeur de recherches at CNRS > and is still working on bootstrapping his CAIA system. He is describing on > his blog, http://bootstrappingartificialintelligence.fr/WordPress3/ his views > on AI and his system. > > > Cheers. > > -- > > Basile STARYNKEVITCH == http://starynkevitch.net/Basile > opinions are mine only - les opinions sont seulement miennes > Bourg La Reine, France > > Artificial General Intelligence List / AGI / see discussions + participants + > delivery options Permalink ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T032c6a46f393dbd9-M05c9efd4ac0af5e77cbeb629 Delivery options: https://agi.topicbox.com/groups/agi/subscription
