The questions are whether some independent intelligence is at all possible for a computer program and whether a simple program can show some of the higher orders of independent general intelligence? I think that computers already have done enough smart stuff to answer the first question. The second question is the one I cannot answer yet. I feel that emergence is something that we could see in simple AI programs if they are designed smartly enough. With a narrow form of reasoning you will see productions that are completely fixed based on the input. For general reasoning you will need to see something that is a little different. You will first need to see that the program has a general grasp of the subject matter so that it can not only produce a good answer to a narrow question but it can also have some insight about what it did. The simplest example of this is for the program to act as if it was aware of what it did. As this pseudo-awareness becomes more insightful, as it gains more knowledge about what it did, it will become aware. This isn't the sci-fi "aware" (like Skynet became *aware* in the more powerful sense) but a more fundamental kind of awareness. Like what I am doing right now is writing about a possible route toward AGI in an online discussion group. This awareness makes me able to respond to other kinds of statements about what I am doing so not only am I able to produce some written thoughts about AGI but I can also make some statements about this. I am not only able to produce some output but I have some simple awareness about the fact what I am doing.
So, by creating a network of "ideas" which are able to fit these ideas into many different kinds of situations the computer program could hypothetically become somewhat aware (in this basic sense) of what it is doing. However, this model is very difficult. For example, suppose that it had some "ideas" that it felt that it was using correctly according to how the user has reacted to it in the past. If a new idea was input and if the program decided that the new idea concerned something that it had previously learned about then how is this new idea going to be integrated into the previously learned ideas. Does it supplement some ideas that had been previously learned with this new idea, or is the new idea meant to correct and replace an idea that was wrong? Was the old idea wrong or was it complete nonsense? A new idea might even present another theory about the subject matter in which case the old ideas are not being proclaimed wrong but the new idea is intended as an alternative. These kinds of possibilities can be reduced to simple relations just like the "rule" in rule-based AI that Stan was suggesting. In fact a rule is one possibility. Another possible relation is a relation between idea-objects, which is like a rule in some ways but not completely the same. A new statement may be intended to represent an alternative selection for a rule or relation. Or it may be intended as a correction. A correction may be hard (completely wrong) or soft (some points were wrong but some points were right.) Of course the teacher may make a mistake but if the program is able to learn to make some inferences based on a greater awareness then it might one day be able to find a possible conflict in what the teacher has presented. Obviously, it is not going to work. However, by putting an abstract of its internal reasoning on the screen along with the text based IO that I want to use, if it ever comes close to seeing relations in the Input that I want it to see, I can emphasize those relations by using the mouse to select phrases or snippets of its reasoning (or the material that I presented to it) that I want it to emphasize. Although this is not going to be true AGI I will be able to see how my if my ideas make any sense in a working model. Then I would hope that I could find ways to emphasize relations without the pointing and clicking while still using the abstract of its internal reasoning. I am making sense. Jim Bromer. ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-c97d2393 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-2484a968 Powered by Listbox: http://www.listbox.com
