The singularity is near - raymond kurzweil

[Robbo]

Going back to the feasability of artificial intelligence...
If one is wanting to replicate intelligence, they would have to replicate the skill of decision making. But what is decision making? Decision making is just a set of values, information about the current situation and repercussions of making a certain decision. This may not be as far away as you think.

In fact, computers already do this to a basic level to begin with, take the 'if' function in excel for example, you give it a set of values and a question. It returns with a true or false answer and based on that it does something else.
All that is required to make this a full-blown ai is make it understand and consider the repercussions.

I'd hardly think that conditional responses utilized by Excel constitute anything resembling sophisticated decision making that would be the criterion for AI, this is an absurd proposition.
AI decision making as exemplified by those needed to pass the Turing test aren't as simple as conditional 'ifs', they are decisions that are used to define what's human such as somethign like, if a husband pushes his wife to the floor, is it best she leaves him ?
To answer this question correctly, things like circumstances, financial consequences, children etc etc all have to be taken into account ... this is not a simple conditional decision, far from it.
And it is these type questions that are used to differentiate AI from human.

But even a computer do this presuming a human firstly gives an indication of how good or bad a outcome will be. Ie, if killing someone is bad, it could be worth -100 points. If every outcome and repercussion given a value, it would result in an equation. And if this equation proved to turn out better than the current one, it is a worthwhile thing to do. However, yes. It would have to calculate every possible outcome and work out the chance of a good thing happening over a bad thing.

Once again PW, this is absurd, good or bad isn't always black and white, it is more often than not somewhere on a spectrum with the correct answer sometimes having its roots in morality or religious doctrine or clinical efficacy or a myriad of other considerations, the assignment of a number isn't gonna equip a computer to make the right decision here any more than assigning it a name.

And earlier you stated," If you tell a computer 1 + 1 = 2, how can you learn from that?".Well, all that would be required is to let it work out a rule. It could teach itself every addition calculation in the universe. The same goes for any mathematical equation really. Ie if a number is x above 1, the answer will be x above 2.

I'm at a loss as to what you mean here with this algebraic substitution you have used in the last line, it seems to make no sense ... to me at least.
And as for how all this is meant to enlighten anything to do with decision making with regard to AI is beyond me and so can you please expand upon or at least clarify, what you have posted here.

 

[PaintballWizard]

i understand what i mean anyway

What i'm suggesting is surely a basic form of intelligence can already be created. Decision making is criterial right? You make decisions based on how you would do something (what would have to be done), the outcomes of it, and whether this is an improvement on your current situation.

So what i'm saying is, if numerical values can be given to each of these things, then a computer could calculate whether or not this is an improvement on the current situation.

Take your example of a man pushing his wife to the floor. If the computer was the wife, having been pushed, her current situation is bad (- 325 for instance). If she left him, her situation would improve (153 for instance). She would then have to consider how she would leave him, (by telling him that she is leaving). Based on this, she could make a decision. So, if a computer were given a situation and values for each of these things, then why could it not work out that the situation would be better than it currently is and institute the actions necessary to do so.

However the flaws in this are; the computer would also have to calculate the other effects of the actions necessary to do something and, as you pointed out, is that how good or bad a thing is is down to opinion based on many things (political, religious, ethical etc. standings).


As for the 1 + 1 = 2 bit, if you gave a computer 1 + 1 = 2, 2 + 1 = 3 and 5+1 = 6 (in the form x+y=z), then surely it could work that as x increases, so does z by the same amount. So theoretically it could calculate any x+y=z calculation. The same could be acheived with more complicated calculations but this would probably not help in creating intelligence.

 

[Robbo]

i understand what i mean anyway

I got a sneaky feeling you might be the only one present in that somewhat priviliged position

What I'm suggesting is surely a basic form of intelligence can already be created. Decision making is criterial right?

Hmmm, once again I am struggling to make sense here and I don't think it's because there is an obvious problem with articulation, it's because it looks as though there is no sense to be made.
Before you analyze anything, especially something as complex as emergent AI, then you have to make damned sure you define your terms and then understand them, with no disrespect PW, I'm not sure you do mate.
If you are defining intelligence as the ability to make a decision [which is what you are inferring I think] then any PC [in your frame of reference] can be seen to be 'intelligent' ... problem is, a PC isn't intelligent at all, it's an extremely fast idiot, that is all.
Any decisions the PC ostensibly makes is an emergent property of its conditional programming ... and nothing more.

You make decisions based on how you would do something (what would have to be done), the outcomes of it, and whether this is an improvement on your current situation.

So what I'm saying is, if numerical values can be given to each of these things, then a computer could calculate whether or not this is an improvement on the current situation.

Once again, this makes no sense at all mate, none at all, of course you can assign values to certain states or conditions and the PC can select one value over another but this doesn't mean it's intelligent at all, it just means you have programmed it to select a value that you have assigned to a condition ... it's all too arbitrary.

Take your example of a man pushing his wife to the floor. If the computer was the wife, having been pushed, her current situation is bad (- 325 for instance). If she left him, her situation would improve (153 for instance). She would then have to consider how she would leave him, (by telling him that she is leaving). Based on this, she could make a decision. So, if a computer were given a situation and values for each of these things, then why could it not work out that the situation would be better than it currently is and institute the actions necessary to do so.

-325????
Why not make -17 or even make it 45 cucumbers, these values are so arbitrary PW and ultimately mean nothing and all you are doing once again, is assigning what look like arbitrary values to certain conditions and then what???
Getting the PC to select this or that value .... and that decision is based upon what exactly, the lower the number the less satisfactory it is????

This is insane to then go on to conclude a a PC is exhibiting intelligence by choosing one value over another.

However the flaws in this are; the computer would also have to calculate the other effects of the actions necessary to do something and, as you pointed out, is that how good or bad a thing is is down to opinion based on many things (political, religious, ethical etc. standings).

This is what I pointed out in my prevoius reply to you and it is EXACTLY this type decision that would have to be made for a computer to pass the Turing test and therefore be deemed artificially intelligent.

As for the 1 + 1 = 2 bit, if you gave a computer 1 + 1 = 2, 2 + 1 = 3 and 5+1 = 6 (in the form x+y=z), then surely it could work that as x increases, so does z by the same amount. So theoretically it could calculate any x+y=z calculation. The same could be achieved with more complicated calculations but this would probably not help in creating intelligence.

Computers do this type calculation all the time, it's basic algebra, in fact any calculator can do this but what this has got to do with anything is anybody's guess.
All in all PW, I'm afraid I can't agree with pretty much everything you have written here and maybe it's just a case you understand what you are saying and others might have a problem, yeah, let's say it's that

 

[Dan w]

ok nevermind my other post read something else that made more sense than the way you explained it.

how can you make a machine that is more intelligent than a human, how can you teach a computer emotions, to be creative, how can a machine have an imagination when it is programmed by a human. The imagination has no rules to follow so how would a machine interoperate that.

How would a machine make another machine more intelligent.

for example if all a machine knew was that 1+1=2 then how would it teach another machine how to do more than that.

Machines could do this if someone made a machine which could learn, and investigate. A machine that has access to and can re write its own code

 

[Robbo]

Machines could do this if someone made a machine which could learn, and investigate. A machine that has access to and can re write its own code

And if my aunt had bollocks she'd be me uncle ......
Machines are able to learn already but do they understand the information they have accumulated?

Nope; they can however look as though they're acting upon that information with pre-programmed conditioning but that's not intelligence.

It may look as though it is to someone who doesn't appreciate the distinction between a programmed conditional, essentially mechanical response as against a response from an self-aware person.

And as for a machine having access to its own code and then having the ability to rewrite it is something that has to be done by someone but even then, you seem to be suggesting that just by allowing a machine to write its own code this would then give rise to some form of intelligence ... why?

It just means you give the machine the ability to program itself; you are suggesting it would write code that would then allow the machine to be intelligent ....but ... to do this, the machine has to have 'intelligence' as a goal in its original programming and if you could do that, i.e. write code that gives rise to 'intelligence' then you wouldn't need to program it into a computer so as to write its own code ...

Well, maybe you are suggesting that if you build enough machines that can program themselves, then by evolutionary circumstance, if you wait long enough, one may end up writing code that allows the computer to make intelligent decisions or better put, to be intelligent.

There are many things that can seemingly make an intelligent decision but this does not mean they are in fact intelligent, you must bear this in mind.

I would love to know the nature of the program that allows a computer to program itself to ultimately achieve self-awareness [intelligence], whoever writes that code is gonna be looking at a Nobel prize, a huge bank balance or a bullet in the back of the head ...... think about the consequences here for a minute.
If I ever managed to do it, the last thing I'd do is make it public, that would be insane.

 

[Missy-Q]

Wait!
Killing someone is -100 points, but pushing your wife is -325?

Why bother pushing her at all? Just throw her in the river, save yourself 225 perfectly good points, which could be used visiting brothels etc. Not that a computer would visit a brothel, but thats not the point, or points, whatever.

 

[Missy-Q]

Machines could do this if someone made a machine which could learn, and investigate. A machine that has access to and can re write its own code

Dangerous talk. Did terminator teach us nothing?

 

[Dan w]

But what is the human brain? In essance, just a very complex computor that can change its structure and think for itself, even have creativity. So why could a computor not be the same as a brain?

 

[Robbo]

But what is the human brain? In essance, just a very complex computor that can change its structure and think for itself, even have creativity. So why could a computor not be the same as a brain?

Hi Dan, just because you can liken a computer to a brain doesn't mean it will evolve into one.
Yes they work on similar principles of operation i.e. they both have input and output variables, they both have different forms of memory and they also both exhibit decision making centres but that's about it mate ....

It's a bit like suggesting a push-bike can eventually go round Brands Hatch like a Honda CBR1000 just because it has two wheels, an engine and a rider ... it just ain't gonna happen mate and just as much as the limiter in the bike's metaphor is top end speed, the limiter in computer brain function is self-awareness .... for all the clever things computers have achieved, they can only ever be described as extremely fast idiots; not one of them has got anywhere near being self aware ... not even close.

 

[Missy-Q]

But what is the human brain? In essance, just a very complex computor that can change its structure and think for itself, even have creativity. So why could a computor not be the same as a brain?

Emotions, ethics, morals, pity, sympathy, forgiveness, instinct, etc, etc.

The human brain is not, in essence, a computer. How do you justify this statement? The human brain is as yet 60% mystery. We dont even know what most of the genes in the human body are there for, or how the brain controls them to allow/inhibit behavior/action. The computer part is just a tiny fraction of the overall functionality of the human brain.