I don't see how I am going to make a serous contribution to XAI at this time. However, I am interested and I have some insights that could help. And I would be happy to do some coding if I could be part of a group.
One of my recent insights is on what I am calling the computational advantage. There are two types of computational advantages, one is computational mathematics (arithmetic) and the other is precise memory of reasonably finite data. It is my conclusion that if you want to create a competitive product in a challenging field you are going to need to use these two forms of computational advantage. But, I do not think that the past efforts in this field are adequate at this time. For example, the attempt to use semantic vectors looks really awful to me. (I am not completely dismissive of that approach, I just do not think that it will work in the near future.) And although deep learning has made some significant advances I am a little skeptical of methods which build approximations on approximations. I do not doubt that we use approximations in our thinking, but I just have an intuitive sense that a computer program should use distinct memories to build and act as representatives of generalizations and abstractions. I am thinking that if an XAI program is going to build generalizations of some sort it has to be able to explore the distinct memories of the objects (or of representatives of the objects) which were foundation to those generalizations. That way, the use of a speculative generalization or abstraction could be accompanied by particulars which could be used to examine the use of the abstraction from other conceptual vantage points. From this basis I would say that we should develop a mathematical system to index these kinds of relations so that the effort to find relevant characteristics of a situation could be done using the computational advantage of arithmetic thus avoiding many of the deep search problems. Jim Bromer ------------------------------------------- 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
