Natural Intelligence in Artificial Systems

NATURAL INTELLIGENCE IN ARTIFICIAL SYSTEMS

3/10/99 11:46:19 AM

Hi Joan,

I ran North up Mad River Beach yesterday. Just a little bit out, noticed large waves travelling smoothly and slowly upstream. A phenomenon not noticed before. Realized why - being pre-informed by your note I inferred that the source was the new mouth. Quite amazing. The ocean was angry and the strip of beach seemed fragile. Felt vulnerable to the elements

For about a month I have been working on the CKB/EB project = design of Conceptually formulated Knowledge Bases using Encyclopedia Britannica as the example. When you were here last, about one third of the work had been done. Now two thirds to three fourths. These estimates are r/t the present phase which is investigative so that I can have a handle on the design [re: automation and generality] and have one good, if manual and specific, example in hand. Numerous amazing things have happened. My programming skills are serving me as a powerful tool. Becoming aware of the power locked up in Word 97 and other Office 97 programs; how to use this. The most amazing thing happened this morning. I began the project because I wanted a list of ways of “Exploration”...and one thing led to another so the beginning of the project was very manual and full of lots of context dependent information - not very good for automation. But the detail used the formatting capabilities of Word to distinguish the different kinds of conceptual levels and contexts. Then the project itself became important [as you know its something I’ve wanted to do and is part of the EDA and B/E program] and I began to automate - you saw some of that. The chore was the change in midstream from manual to automated work and translation from manual to automated manipulation owing to not wanting to lose the manual labor. The question arose: why not just start over with the automation, would not that be more efficient? Partly because of the challenge, partly because I wanted to retain the context based coding, and finally because this kind of thing happens a lot [we understand projects better as we learn so we have to go back to start] I wanted to use the manual work as the basis of automation. There has to be manual entry and so on, but in large automated systems the manual entry is carefully designed. Anyway, this has been a challenge and I have learned a lot including how to design for automation later. But the amazing thing that happened this morning was that the automation that I implemented is way more intelligent than I planned or anticipated. Not hoped because I just was not counting on the result. The automation has retained the context dependent information in a way better than I would have done manually. That is, it is doing a better job even where not explicitly designed to and a better job than I did laboriously by hand

This makes me think about the problems of software design. A major problem is the reliability of mission critical programs. Mistakes are costly. My automation makes mistakes: not in the area [so far as I know] that it is designed for, but in the context dependent codes that it was not designed for. But while it makes, say 5% mistakes, it is 90% correct and 5% brilliant. That’s a little like life and more like life than mission critical stuff which enables us to control vast populations, territories, technologies but is prone by its nature to failure. Life is prone to failure but in its own terms there is no such thing as failure; risk-fearing humans cradled in the warmth of an indulgent civilization: we project the idea of failure. So. As far as research and using a CKB system: perhaps a little mistake here and there is not a bad price to pay. If I tweak my knee on the way up to the summit, or if some hikers die...what does that mean? Life grows upon death. Certitude itself is uncertain; we like to maintain the delusion. It is, of course, an absolute delusion in the realm of the proximate... Data/processing systems incorporate “fault-tolerant” computing. That means that correct results are obtained [with some degree of confidence; the hope is that this is acceptable] despite hardware and software failures. Alternatively [but not exclusively], perhaps there is a large future in which computing is evolutionary and “faults” are part of the results. This mimics life. It is more robust and stands to gain more

...and reminds me of the point made in the essay Dynamic Uses of Computers. The title would have referred to “Electronic Information and Data Processing Systems” rather than computers if I had wanted to sound sophisticated. Anyway, the point. It was that, regarding Artificial Intelligence, the intelligence [and consciousness if that is at all possible] is to be found in the interaction between the human and the tool, and the origin will be in the evolution of that interaction. The process includes designs but takes on its own life and is larger than human designs

Anil