Author
|
Topic: Intelligence (re)defined (Read 807 times) |
|
David Lucifer
Archon
Posts: 2642
Reputation: 8.53
Rate David Lucifer
Enlighten me.
  
|
 |
Intelligence (re)defined
« on: 2005-07-30 18:22:58 » |
 |
Allow me to propose a new (as far as I know) definition of intelligence:
Quote:| A system is intelligent to the extent that it can generate high quality choices and decisions. |
It is always risky to attempt to redefine a widely-used term but I think it can be worthwhile if it leads to new insights while still capturing the connotations of previous definitions. Intelligence has been notoriously hard to define (it is an ongoing joke within the field of AI). But how can you hope to understand an intelligent system (let alone build one) if you can't even define it? I hope this new definition sheds some new light on some old questions: What kinds of entities are intelligent? How does one recognize or test for intelligence? Are there different kinds and domains of intelligence? Does it matter how something is intelligent, i.e. can something act intelligently but not actually be intelligent? How does consciousness figure in (if at all)? Can intelligence arise from non-intelligence? Is it an emergent behaviour? Are animals intelligent? Can a computer (program) be intelligent?
What do you think the answers are given the above definition?
|
|
|
|
|
Dorian
Initiate
Gender: 
Posts: 3
Reputation: 5.04
Rate Dorian
|
|
Re:Intelligence (re)defined
« Reply #1 on: 2005-08-11 08:56:01 » |
|
Quote:| What kinds of entities are intelligent? |
Under this definiton, it seems as if the majority of the animal and plant kingdoms are firmly grandfathered in, especially if we judge the quality of the decisions made by these entities by the survival of their respective species, as a whole. Even to do that, however, I have to modify your definition:
Quote:| A system is intelligent to the extent that it can routinely generate high quality choices and decisions. |
In the short term, it seems foolhardy to judge the intelligence of a system based on isolated incidents. Perhaps the observer lacks enough of a context to perceive the quality of the decision, etc. Also, otherwise "intelligent" (here, meaning educated) individuals have been known to make poor choices in isolated incidents.
So what is left? Consider the following program:
Code:
#include <stdio.h>
int main() {
while (1) {
char buf[80];
scanf("%s", buf);
printf("Yes.");
}
return 0;
}
|
Now you see why I modified the definition. If we asked this program "Is Fermat's Last Theorem true?" we might be willing to judge it intelligent, when in fact further questioning would prove it to say the same of everything, even questions that cannot be answered with yes.
If we were to put this (and other programs like it, in general: finite state programs that return a finite number of responses for a finite number of categories of inputs) to a version of the Turing Test which, instead of testing for humanity, tested the program's ability to make choices, I feel a finite state method would fail. There are more scenarios that can be imagined than are dreamt of in a programmer's philosophy, to paraphrase.
What I think may be sufficient, then, is for a being to be able to communicate a context. When a meme is transmitted through a medium from one being to another, the context of the meme is the collection of enivronmental and psychological factors which allow one being (A) to 'understand' an interpretation of a meme which the other being (B) can derive A's meaning from. If A and B have different contexts, B may misinterpret A's meme, and be infected with a different meme. Further conversation between A and B may reveal this cognitive dissonance, allowing for A to reinterpret its conception of B's context, and vice versa.
The feedback between A and B is the important part. Therefore, I speculate that a sort of cybernetic feedback loop within a computer program to guage the effects of its responses is necessary before declaring the program to be intelligent in the sense you have given.
Feedback loops like this can't happen in machines that are strictly Turing-complete (that is, are minimally-complete machines without other, user-friendly additions.) The feedback comes when there is an open channel of communication between an intelligent judge and a being in question.
I'm curious, however, if a judge could accurately judge a being that it could not communicate with. What do you think?
|
Mathematics, Philosophy, and Physics. This is all I require.
|
|
|
David Lucifer
Archon
Posts: 2642
Reputation: 8.53
Rate David Lucifer
Enlighten me.
|
|
Re:Intelligence (re)defined
« Reply #2 on: 2005-09-02 00:23:33 » |
|
Quote from: Dorian on 2005-08-11 08:56:01
Now you see why I modified the definition. If we asked this program "Is Fermat's Last Theorem true?" we might be willing to judge it intelligent, when in fact further questioning would prove it to say the same of everything, even questions that cannot be answered with yes.
|
I see what you're saying and I agree. I thought "routinely" was implied by the plurality of "choices and decisions" but if it isn't obvious then your clarification is justified.
Quote:
If we were to put this (and other programs like it, in general: finite state programs that return a finite number of responses for a finite number of categories of inputs) to a version of the Turing Test which, instead of testing for humanity, tested the program's ability to make choices, I feel a finite state method would fail. There are more scenarios that can be imagined than are dreamt of in a programmer's philosophy, to paraphrase.
|
Unless you have an infinite number of neurons then you too are an FSM so where does that leave you?
Quote:
Feedback loops like this can't happen in machines that are strictly Turing-complete (that is, are minimally-complete machines without other, user-friendly additions.) The feedback comes when there is an open channel of communication between an intelligent judge and a being in question.
|
Is your definition of "Turing-complete" somehow different from that of the universal Turing machine? I ask because UTMs can compute anything that is theoretically computable and humans aren't even that computationally powerful.
|
|
|
|
|
|
|
Dorian
Initiate
Gender:
Posts: 3
Reputation: 5.04
Rate Dorian
|
|
Re:Intelligence (re)defined
« Reply #4 on: 2005-09-14 22:49:56 » |
|
I deny that humans are finite state machines.
1.) Finite state machines are not self-modifable. Humans are.
2.) Given two identical inputs, a finite state machine will yield the same result. A human may not.
Your typical personal computer is not a universal turing machine. In fact, true UTMs cannot exist in this universe.
|
Mathematics, Philosophy, and Physics. This is all I require.
|
|
|
|
|
rhinoceros
Archon
Gender:
Posts: 1318
Reputation: 8.02
Rate rhinoceros
My point is ...
|
|
Re:Intelligence (re)defined
« Reply #6 on: 2005-09-18 17:33:16 » |
|
Quote from: Dorian on 2005-09-14 22:49:56
I deny that humans are finite state machines.
1.) Finite state machines are not self-modifable. Humans are.
|
Could you clarify in what sense finite state machines aren't self-modifiable (and in what sense humans are). Do you mean it in a way that punching symbols on a tape does not cut it?
I can easily imagine a Universal Turing Machine spawning modified copies of itself.
Quote:
2.) Given two identical inputs, a finite state machine will yield the same result. A human may not.
|
I can neither agree nor disagree with the second sentence. How can it be ever shown that under exactly the same conditions a human may respond differently? How can either this or the opposite be ever disproved? It's on faith.
I'd say that it is in the spirit of science to assume and try to establish cause(s) and effect. On the other hand, the task does seem impossible. We start to go to a movie, we meet an old friend from school... we end up at a cafe. We have to account for the whole Universe... So, perhaps it is true that the Turing Machine model is not productive in this case.
Quote:
Your typical personal computer is not a universal turing machine. In fact, true UTMs cannot exist in this universe.
|
This is only true in the same sense that a true straight line cannot exist in this universe. The Universal Turing Machine is a model. A model applies to the real world under some assumptions, as long as it serves a purpose. The UMT model assumes infinite memory and infinite time, but it is routinely used for personal computers in cases where lack of memory or time do not come into play, that is, for handling questions of what is logically possible.
By the way, for the fellow-virians who are not up-to-date regarding Turing Machines, I found this article prety good.
http://plato.stanford.edu/entries/turing-machine/
|
|
|
|
|
David Lucifer
Archon
Posts: 2642
Reputation: 8.53
Rate David Lucifer
Enlighten me.
|
|
Re:Intelligence (re)defined
« Reply #7 on: 2005-09-18 23:21:25 » |
|
Further to rhino's reply...
Quote from: Dorian on 2005-09-14 22:49:56
I deny that humans are finite state machines.
1.) Finite state machines are not self-modifable. Humans are.
|
For every finite being (including humans) there is an equivalent FSM (though it is probably too large to exist in one universe).
Quote:
2.) Given two identical inputs, a finite state machine will yield the same result. A human may not.
Your typical personal computer is not a universal turing machine. In fact, true UTMs cannot exist in this universe.
|
The first sentence is simply false. It is trivial to demonstrate a 2-state FSM that yields 2 different results to one identical input.
|
|
|
|
|
|
Logged
