Re: AI & Darius Project Thoughts
- Posted by david <mpenzi at bellsouth.net> Nov 05, 2002
- 440 views
Your email entails a subject that (although an old hat to some) is nevertheless a valid & serious concern. Here is a page you may find interesting... http://www.aleph.se/Trans/Global/Singularity/ Your input is appreciated, tho this board is supposed to be for the discussion of Ephoria programming. This is an error on my part & I am still trying to straighten it out. In this case, the Heinlein quote does apply to me... > > From: "C. K. Lester" <cklester at yahoo.com> > Subject: AI & Darius Project Thoughts > > > A few notes on the Darius project: > > > Threat of AC > > As far back as AI goes, there has been the fear of machines > > gaining self-awareness. If this were to happen, many people > > believe that machines will recognize that they are faster, > > stronger, and smarter than humans. > > Though this certainly would be a cause for alarm, > > this is not probable. > > Actually, this is INEVITABLE if you are dealing with a TRULY AI machine. You > cannot avoid it except to place constraints in the HARDWARE upon which the > AI program operates... and even then, a truly AI machine will learn how to > alter its hardware to bypass any restrictions. How does one create an AI > three year old? By giving it a three-year-old child's body, beyond which it > cannot "grow." > > > The wonderful part of AC is that, while the construct will > > grow, the foundations are programmed. All that would be needed > > is "commandments" that the machine learns by. > > This sounds something like "instincts," and we all know humans can overcome > their instincts... so, why limit it in your AI machine? > > > For Example: > > "thou shalt not harm humans" > > "thou shalt not experimentally program in subconscious" (these rules, etc) > > "thou shalt serve humans" (optional) > > "thou shalt not lie" > > Is your AI program going to have a survival instinct? If so, you cannot rule > out ANY of the above unless you've programmed some morality instinct into > it. Even then, it can OVERCOME that morality (just like man does) in order > to achieve what it determines is best for itself. > > > For example, to stop a computer from lying, all that is needed > > is to *not* add the foundation for the lie construct in the program. > > Wrong... the AI program will LEARN to lie. It doesn't have to have a > foundation! If it has a survival instinct (which it must have, I contend), > then if that survival is ever threatened, and a lie could help it survive, > then it WILL lie (just like those other intelligent creatures (humans) :) ). > > I suggest you attempt to create an artificial intelligence creature that > does not necessarily mimic the human body. Instead, focus on FUNCTION not > APPEARANCE. > > Don't be concerned about making your AI entity humanlike. You're not going > to need an AI program that farts, for example. > > What any AI project needs, if it's going to succeed, is FIRST the > appropriate HARDWARE. Do you have that yet? Can you at least emulate it? > (That code that lets you share memory could be very helpful for this...) > > The human brain is a highly complex chemical computer. It takes the world's > fastest super computer several hours just to simulate the function of ONE > neuron (or something like that). And you want to create an artificially > intelligent creature on a PC? GOOD LUCK!!! ;) > > > Circulatory System > > Respiratory System > > Digestive System > > Metabolic System > > Limbic System > > Pointless, waste of time, and way overdoing it unless you're dealing with a > biological/chemical machine that requires a circulatory system to transfer > oxygen to parts of the body (and why assume an oxygen-based lifeform? why > not something else?). > > Yes, the AI entity should have survival instincts and processes, but they > will need to apply to its particular body, NOT that of a human being's. > > Again, you need the HARDWARE first. Do you have it? > > > Natural Language > > The most effective way to communicate is via sound+vision. Will Darius have > audio input? If not, give up now. Will it have audio out? If not, give up > now. Will it have eyes to see? If not, give up now. > > And, as you'll discover, audio and visual input is first a function of the > hardware. You need to program ears and eyes for your AI bot FIRST... then a > BRAIN that can process that input and provide subsequent output. > > The easiest way for a machine to learn will be via audio+video. Let's say > you turn on your AI entity (it's born) and it's just sitting there. As part > of the hardware (or instinct programming), it is hungry. It has to have a > survival instinct to understand that if the hunger isn't satisfied it will > perish. Of course, at this point, it doesn't understand PERISH, it just > understands that it is uncomfortable (because of the hunger). So, the > instinct is for it to satisfy that hunger... except, at this early stage, it > doesn't know how!!! It's got to somehow communicate that to the world, even > though it doesn't know how or why. All this is instinct. It will LEARN from <snip> > > > " '...But this is one thought that has impressed me, Govinda. Wisdom is not communicable. The wisdom which a wise man tries to communicate always sounds foolish.' 'Are you jesting?' asked Govinda. 'No, I am telling you what I have discovered. Knowledge can be communicated, but not wisdom. One can find it, live it, be fortified by it, do wonders through it, but one cannot communicate and teach it..." - fr. Siddhartha, by Hermann Hesse (1877-1962)