I lost my cool with some comments Steve wrote on a message thread that I started recently. He started by reporting on some problems with a program that he wrote 20 or 30 years ago. That was cool because it gave me chance to use his experimental results to look at my own ideas. I gained something from his effort. I was able to sharpen my reply in response to his comments. My argument is reasonable and I will state it below. But then, instead of thinking about my response his next response was to declare that my ideas to be like Arthur Miller's ideas and Arthur got a lot of criticism in this group. Well I don't know how Arthur's ideas have progressed but I don't ever remember Arthur talking about what I am talking about and Steve did not provide any substance to demonstrate that my ideas (especially the idea that I was proposing in the message thread) were just like Arthur's. So I lost my cool and pointed out that Steve's dismissal was totally lacking in substance. His comments on his own experiences did contain substance but his dismissal that implied that I was talking about the same thing that Arthur had talked about was shallow.
It is difficult to say how a substantial argument might be made. You really need to be able to support your argument with some way to validate that it is indeed substantial. So it seems that the only way to prove that an argument contains something of substance is to show that substantial arguments can be made to support that it did. My argument, that I developed in the thread thanks to the more substantial comments that I got was this. Text-based would-be-bases-for-AGI programs have not gotten much traction. I believe that complexity is a major problem but it should be able to show that the program can get enough traction to prove the concept beyond the capabilities of contemporary AI programs. A major problem is that the program would need some way to process direction from the user-teacher in order to teach it to look for anaphoric-like connections and to disambiguate language. However, if hard-edged language (or other hard-edged input) is used to designate these directions then the program will not need to be creative enough to do the substantial learning that it will need to do on its own. On the other hand if the program is creative enough so that it can try lots of possibilities then this creativity will tend to overwhelm the program right away. So my solution is to give the program the underlying possibility to recognize that some parts of conversation might effectively contain directions on how it can interpret other statements, and then if the user starts off by repeatedly using very simple sentential forms to input this directive information for the program it should be possible for it to gain enough traction to produce some genuine learning. The program will need to learn the rudiments of natural language but by programming it with the underlying possibility to interpret some statements as directives it should be able to start off. Of course there is more to it but this argument of mine is substantial, reasonable and I believe it is probably novel. 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
