## What are the criteria for a system to be considered intelligent?

9

3

For example, could you provide reasons why a sundial is not "intelligent"? A sundial senses its environment and acts rationally. It outputs the time. It also stores percepts. (The numbers the engineer wrote on it.)

What properties of a self driving car would make it "intelligent"?

Where is the line between non intelligent matter and an intelligent system?

Question was closed 2020-08-18T23:48:50.490

After reading control theory wikipedia: The sundial only carries out two of the four functions of a control system: 1:measure 2:compare, 3:compute 4:correct

– Conor Cosnett – 2016-08-17T08:56:30.590

4

Typically, I think of intelligence in terms of the control of perception. [1] A related, but different, definition of intelligence is the (at least partial) restriction of possible future states. For example, an intelligent Chess player is one whose future rarely includes 'lost at chess to a weaker opponent' states; they're able to make changes that move those states to 'won at chess' states.

These are both broad and continuous definitions of intelligence, where we can talk about differences of degree. A sundial doesn't exert any control over its environment; it passively casts a shadow, and so doesn't have intelligence worth speaking of. A thermostat attached to a heating or cooling system, on the other hand, does exert control over its environment, trying to keep the temperature of its sensor within some preferred range. So a thermostat does have intelligence, but not very much.

Self-driving cars obviously fit those definitions of intelligence.

[1] Control is meant in the context of control theory, a branch of engineering that deals with dynamical systems that perceive some fact about the external world and also have a way by which they change that fact. When perception is explicitly contrasted to observations, it typically refers to an abstract feature of observations (you observe the intensity of light from individual pixels, you perceive the apple that they represent) but here I mean it as a superset that includes observation. The thermostat is a dynamical system that perceives temperature and acts to exert pressure on the temperature it perceives.

(There's a philosophical point here that the thermostat cares directly about its sensor reading, not whatever the temperature "actually" is. I think that's not something that should be included in intelligence, and should deserve a name of its own, because understanding the difference between perception and reality and seeking to make sure one's perceptions are accurate to reality is another thing that seems partially independent of intelligence.)

The thermostat is a great example of a minimal intelligent system. Could you give me some example of what you mean by "control of perception". – Conor Cosnett – 2016-08-16T16:22:51.303

1@JohnConorCosnett added. The other definition comes from writing by Eliezer Yudkowsky, but I can't find a good reference at the moment. – Matthew Graves – 2016-08-16T19:28:47.600

2

To ask what makes a system intelligent almost begs the question 'in this context what do we mean by artificially intelligent?' which I think this what this question is really gearing towards.

From my studies, I've come to see that 'Artificial Intelligence' is a catchy term to use but perhaps misleading, and it conjures up images of these self-driving cars and robots that will take over the earth.

What I've found AI, and 'intelligent' systems moreso represent is an aid or a support that works for us, rather than one that works because of us... hear me out:

What makes the jump to an intelligent system for me is the step where the system begins to 'adapt / learn' or otherwise do things I didn't directly tell it to do. With the sundial, I measured and cut every inch of it by hand, and put it in a specific way to do a specific thing.

When a programmer gets into a car he automated, it may do some things he didn't directly program or maybe couldn't even expect (just one example: querying some database to see lots of people are driving somewhere, discovering a concert is going on there, and asking if the driver wants directions / tickets)

--

In conclusion, an intelligent system to me is one that we build in such a way that it educates and supports us, rather than a system we ourselves 'educate' to do a specific task. Supportive systems that elucidate and adapt and act 'rationally' even when we didn't tell it what 'rational' behaviour was.

Its hard to define "do things I didn't directly tell it to do". Any system of sufficient complexity will do things you didn't directly specify and even surprise humans (humans can only predict the behaviour of simple systems). The Eliza chat bot surprised people. How about a chaotic pendulum?) But I agree if my car perceived patterns and discovered a concert and bought tickets even my granny would consider it intelligent.

– Conor Cosnett – 2016-08-17T08:30:17.043

1Yes I do agree I might've been a bit off on that point. There's many words up there but I was just trying to say that AI is something I've come to view differently, in the sense that to me it's not something that even could lead to sentient evil robots; instead it's supportive services and systems like Watson that educate and assist us. Thus an 'intelligent' system has to be analyzed in that framework. Not a rote medium, but an adapting system that can learn and teach us things we didn't know we were looking for. – Avik Mohan – 2016-08-17T15:21:47.800

-2

Intelligence is the efficiency of an action in serving some purpose.

Both sundials and self-driving cars are intelligent systems.

Anything that serves some purpose exhibits intelligence.

One thing is more intelligent than another thing if it achieves some purpose in less steps.