AI Teaser Topic: How would you define Intelligence?

If we want to study and understand Artificial Intelligence, we first need to understand what "Intelligence" is. Its not easy to define intelligence, and there are probably as many definitions available as there are people who have ever thought about it. Most of the times, people end up giving a backward definition, i.e. they first decide what things are intelligent, and then,based on their first assumption, try to define what intelligence is.

So what/who according to you are intelligent? No-one would argue that human beings are intelligent beings. But is everyone intelligent? Even if we accept that generally human beings are intelligent, can all human actions be defined as intelligent? What about irrational actions? What is rationality? Are rationality and intelligence the same thing?

What about animals? Are animals intelligent? Or would you classify some animals as intelligent and others as not? What about plants? They have life, but are they intelligent? So is intelligence dependent on life? If not, then can we say that machines are intelligent? Is your computer intelligent? What about your mobile phone?

Ok, so lets suppose your computer is intelligent. But can it think? Does it possess cognition? You may argue that a machine can work, but cannot think. But how can you prove that? You can't see the process of cognition, you can only see and judge actions. For example, how do you know that I am not a robot? You've never cut me open to see whats inside. You've never seen me bleed. You've seen me eat, but the food may just go to a repository for temporary storage, but my actual energy comes from batteries. Isn't it theoretically possible that I could be a robot but you don't know that? You think I am a human because I act like a human and speak and walk like a human. So, do we verify intelligence through actions, or cognition? And if an entity acts intelligently, is it possible that it may also have a cognition process?

Keeping all these things in mind, I want to hear your version of the definition of "Intelligence". Remember, there's no right or wrong answer. So give it your best shot. The best answer will get n advanced bonus points with a probability of p

Bonus food for your thought: Are you going to answer this question? If you answer, what would you gain? (Recognition? Self-respect? n*p bonus points?) What would you lose? (Time? Energy?) What are the factors that determine whether you choose to reply to this post or not? Would you reply if I gave you x taka for a good reply? Would it depend on the values of n, p and x? What if p is 1? What if n is negative? Theres the first lessons of Utility theory for you

In order to study AI, you need to study intelligent beings and intelligent processes. And you have one perfect guinea pig right with you all the time: yourself. Think about your thoughts. This is called metacognition (cognition about cognition).

Now, this is the kind of topic I love. It is written after a lot of thinking, and it is making me think a lot as well.

About defining intelligent beings - in my opinion there are several characteristics. As far as I know, human mind(note that I didn't say brain) works in a linear fashion (in a very fast pace). If we consider that all human are intelligent, then the following characteristics are not required to be fulfilled all at the same time. But because some of them can work as "background processes", multiple of them can be be fulfilled at the same time.

1. Remembering.
An intelligent being is able to remember what it has experienced (going through the experience or viewing something else go through the experience).

2. Learning
In other words, learning from mistakes. It does things, and when a mistake occurs - it attempts in a bit different approach, ergo avoiding the mistake.

3. Recognizing Variables/factors
An intelligent being is able to recognize different factors associated with events, and is able to establish relation within these (not necessarily a mesh relation of course) or will attempt to establish relations among these. (this point will have better uses in later points)

4. Emulation
Lets consider that - the intelligent being we are talking about is a robot actually. It has a emulator inside, where it can emulate the previous events it has experienced. But it is also able to emulate those events in a different way by changing the "values" of the factors it was able to recognize (at point 4, i.e. emulating the events that actually never occurred! For humans, we generally call this - imagination.

5. Prioritization
An intelligent being sets priority on the tasks it is going to perform or is assigned to, based on the results. Prioritization depends on the needs of the being.



Well, these are the things I came up with when I read this. I am sure there are many more characteristics which are directly/indirectly related with intelligence/displaying of intelligence.

BTW, I decided to answer this question right when I read the topic title. So n*p, recognition, or self respect was not my target. now that I think about it -

6. Introspection
An intelligent being tries to define intelligence, or find out what is intelligence - and find reasons behind its own - seemingly natural actions.

Thank you Agent Amit

That was a great post, and you raised some interesting points. Especially No. 6: Introspection was a deft touch. I'll wait for some more posts before further discussions. Its a pity 100 people viewed this topic, but no one else posted their definition

Just want to raise a question from your post....

You said that n*p, recognition, or self respect wasn't your target. But what was? What did you gain from this post? According to Utility theory, an intelligent being doesn't perform an action unless it gains a net positive utility from it. You obviously lost some utilities: time, physical energy, mental energy, which are all valuable to you. So if you didn't gain anything that was worth your time and energy, you performed an unintelligent action by posting here So what is it that you gained? In AI, you have to quantify the gains and losses in every move that you make.

Waiting for more posts.
Kabir

P.S. In AI, intelligent entities are called "Agents". So, throughout our course, we'll be referring to each others as agents

I never thought I'd read a post this long But now that I've finished reading it, i find it rather interesting.

Well, intelligence is a tough thing to define. But i think Amit have done a good job with that, and I completely agree with his reasons. Humans are intelligent. To a certain extent, animals are intelligent too, because they analyze their environment to maximize their chance of survival. But intelligence and rationality aren't the same thing.

I'm gonna try to explain it in a weird way. Let's assume that you're a robot. Your goal is to teach as many human beings as you can. So what you did is you destroyed all the other facilities and turned this planet into a massive teaching facility for humans. This is called being rational.

But, since you're intelligent, you know when to stop. That means you have just changed your goal. I guess what I'm trying to say is, being rational means achieving some preset goal by making smart decisions. Being intelligent means being able to set a new goal as well. That's how i know you're not a Robot. (unless ofcourse you have a diabolical masterplan to teach us all )

Machines aren't intelligent. They can never be intelligent (We don't want a technological singularity, do we?). They are only able to mimic intelligent behavior that we programmed them to do. A machine will be truly intelligent when it gains the ability to reprogram (and thus improve) itself. Also, an intelligent being is aware of it's existence.

Machines don't posses cognition either. Machines can't do something they're not programmed to do, and thus can't learn something that they're not programmed to learn.

So my version of intelligence is Amit's version plus self-awareness and ability to improve itself.

Now, why am i answering these is an interesting question. I like to think that I'm rational and intelligent. My answer would definitely depend on the values of n, p and x if i weren't intelligent. But since i am, it serves me a different purpose. It's an interesting post and I enjoy arguing about this kind of things. So for my effort, i get to enjoy this argument for a while. I think it's a pretty good deal. (rational, because it's a good tradeoff. Intelligent, beacause I'm able to enjoy)

Excellent post Agent Shakkhor

With your (wacky) robots-teaching-human example, you just pointed out the problems of a Goal-based agent. More intelligent agents are, therefore, Utility-based agents that try to maximize the utility rather than set one big goal. The big goal (if there is any) can be a huge positive utility so that there is scope to make little sacrifices to reach the ultimate goal.

Self-awareness is a great point that you mentioned about intelligence. I agree with you: machines do not possess self-awareness. There are however people who argue that they do (or may). But about machines not being able to improve themselves, I do not agree. Machines can be designed to learn, and learning is self-improvement.

Your analysis about your motivation for answering is superb. However, why do you say that answering based on the values of n, p, x would be unintelligent? Is it not intelligent to look for bonus points or money? Of course you could argue that this is a lower/selfish level of intelligence, but it is intelligent behavior all the same.

Kabir-IIT wrote:But about machines not being able to improve themselves, I do not agree. Machines can be designed to learn, and learning is self-improvement.

I think i mixed up.

What I'm trying to say is, machines won't learn anything they're not programmed to learn. But yes, they can improve themselves. I've seen chess engines capable of recursive self improvement. It gets better by playing against itself. But a chess engine won't learn to play Sodoku, or Poker.

Edit: Learning optimizes the output of a machine, not the machine itself. For example, a human brain in neolithic age was far less capable than a human brain of our age. Both can learn, to do things more efficiently. But the human brain evolved and became more capable. This is not possible for a machine.

Let's consider another scenario. There's a Robot that has the ability to learn how to make Robots. Now if we give it enough resources to make a smarter and better Robot, will it be able to do so? My point is, machines lack creativity. That's why they can't improve themselves, they're simply able to improve their output.

Edit 2: You're right. Stricken off text is not relevant here. Humans can't create smarter and better humans either...yet. It just happens.

Kabir-IIT wrote:why do you say that answering based on the values of n, p, x would be unintelligent?

Actually I'm not saying that. What I'm saying is if I weren't intelligent, I'd still answer/not answer depending on the values of n, p, x, because it's the rational thing to do.

I get your points Agent Shakkhor Btw, I'm very impressed by your knowledge about AI. Get ready for some in-class presentations (same goes for Amit )