Artifical Intelligence

Artificial Intelligence

We haven’t had a thread, well not for some time anyway, that was devoted completely to the subject of AI, yet it still rears it’s head from time to time in the context of other arguments.

So I thought I’d start one.

Firstly, what is it that we define as intelligence? Whilst avoiding the well known, and much abused Turin Test, my – rather poor – starting definition is that intelligence is

Quote:

---

… the ability to reason, plan, solve problems, think abstractly, comprehend ideas and language, and learn.

---

http://en.wikipedia.org/wiki/Intelligence_(trait)

One presumes that this list is what needs to be satisfied in order to produce discrete components that can be certified as artificially intelligent.

It seems like a nasty incongruent, and effectively impossible, list, but I think that most of it might be achievable.

I’ll try to deal with the first:

(i) The Ability To Reason
The ability to reason is implied in many applications’s today. From games theorist’s game-trees, and game-matrices to compiler technology, Bayesian implication, and backward chaining.

One, though, jumps (not literally) out at me. Backward chaining traverses graphs of a given set of choices from a given conclusion. This holds as the justification of a decision made, and these techniques are used to get expert systems to statistical justify what it is that lead them to make a deduction from the supplied assertion.

I would like to assert that the antonym of a ‘justification made’, is, in fact the ‘process of reasoning’. Each one traverses different directions on a planar graph. Using Bayesian implication (weight the vertices), and perhaps some form of light neural network technology (weight the edges) an agent, I believe, could make rational decisions on the information at hand.

(Those more verse in these techniques will recognise that this is in fact the architecture of a neural network but instead of summation firing function in the neuron we will be using Bayesian techniques)

The neural techniques would provide the plumbing; possibly for efficiency, randomisation, and perhaps some implied reasoning. In particular implied reasoning would, given a particular input assertion, automatically imply a given deduction. This may, perhaps (very perhaps), mimic instinct. Clearly this would have to pre-built for each species built and that could be easily achieved using evolutionary techniques. One of the goals being basic stuff like ‘if there’s food and I’m hungry, then eat the food’

The Bayesian techniques would always be subject to some form ongoing training program; perhaps the absolute square of differences (it’s fast efficient and not particular prone getting stuck on local minima although if the decision is non-linear it can sometimes get stuck) In the case of the ‘if there’s food and I’m hungry, then eat the food’ the Bayesian weighting, by experience, could recognise poisonous food, and override the neural weightings.

I believe that the marriage of the three technologies. Evolutionary programming, neural networks, and Bayesian implication are the simple rules that could set the foundation for the emergence of what we, as humans, might determine as intelligent.

What do you guys think?

Anyone any ideas on the other items on the list?

I'm confused, how do we get from here to Bush bashing?



I, personally feel that the key is in the last line of your second to last full paragraph.

Quote:

---

by experience

---

I like the idea in general, but I think the experience gets overlooked. Too many people want 'instant on' intelligence. I believe we should not be even trying to create intelligence, but rather we should be attempting to create a framework on which intelligence could grow, then put it in a setting where it can learn. I think there may be multiple types of frameworks that could grow an intelligence. The ultimate problem may be putting the framework in a world where it can grow. That is not as simple a task as it may appear. I don't think building a complex enough media for growth is all that difficult, but computers just sit on a desk with little or no input. How do you put the intelligence into a world where it can learn? What senses can it have? How would it feel pain? How many of us have not touched a hot stove?

Quote:

---

Originally Posted by Shaggy Hiker

I'm confused, how do we get from here to Bush bashing?

---

I felt that Bush bashing was a little too easy, so I've tried to find something a little harder to concentrate my efforts on . . . ;)

Quote:

---

Originally Posted by yrwyddfa

I would like to assert that the antonym of a ‘justification made’, is, in fact the ‘process of reasoning’. Each one traverses different directions on a planar graph. Using Bayesian implication (weight the vertices), and perhaps some form of light neural network technology (weight the edges) an agent, I believe, could make rational decisions on the information at hand.

---

Quote:

---

Originally Posted by yrwyddfa

What do you guys think?
Anyone any ideas on the other items on the list?

---

The ability to understand the first quote above.

Which bit?

The Mayor was on TV the other day proposing that 2 months off per year are "too long" when the other countries are going 6-7 days per week, so he's proposing that school be in session year round. I don't think it'll fly, though.

Quote:

---

Originally Posted by dglienna

The Mayor was on TV the other day proposing that 2 months off per year are "too long" when the other countries are going 6-7 days per week, so he's proposing that school be in session year round. I don't think it'll fly, though.

---

Yeah nice :thumb:

Quote:

---

Originally Posted by yrwyddfa

Which bit?

---

All of it, I seem to have a lot of trouble following the thread of conversations in world events, which is why I mainly read about what are you eating right now and the craziest thing you ever ate for supper in chit chat.

I still can't understand why game makers like Bungie (for example) don't rendomize the starting positions of all the enemies in a game. It would be so easy to do and you'd never play the game the same way twice.

You can play the opening level of Halo with your eyes shut virtually and still get through unscathed. Its not hard to do this and yet they don't.

And current game AI largely consists of making an enemy hide behind a crate. Thats it.

Is it a particularly intellient crate?

What would I class as AI? Tough one as something can be alive, but may not necessarily be intelligent. We would consider a bird, cat or dog as intelligent because they are self-aware and capable of solving problems.

However, for myself I would probably say it would be a level of intelligence where an unfamiliar problem (one not taught to solve or have solved before) can be solved by whatever means necessary.
AI is a very murky and grey area because it is hard to tell even in our own world which animals are intelligent and which are not.

Ive always considered AI to be where you could make a computer self aware, but then again how would you prove it. How can I prove I'm self aware?

Quote:

---

Originally Posted by wossname

I still can't understand why game makers like Bungie (for example) don't rendomize the starting positions of all the enemies in a game. It would be so easy to do and you'd never play the game the same way twice.

You can play the opening level of Halo with your eyes shut virtually and still get through unscathed. Its not hard to do this and yet they don't.

And current game AI largely consists of making an enemy hide behind a crate. Thats it.

---

I suppose it would work, but only if they put in specific rules so all the enemies didn't appear in one place. I've sometimes thought about it myself, but then again some games, like goldeneye were fun because you knew where all the enemies were and had to try and get through the level as quick a spossible to get the cheats.

They should just put a randomize option in for when you complete the game or even select how many enemies you want as a percentage of the original to make it more challenging.

Perhaps AI is an something that learns from current experience, not just respond to stimuli?
As for the randomisation of enemies it should be relatively easy. Put enemy 2 ing grid A2, enemy 3 in grid B1, enemy 4 in grid A2, enemy 4 location = enemy 2 location, enemy 4 in grid B2, etc etc

Quote:

---

Tough one as something can be alive, but may not necessarily be intelligent.

---

HAH! I knew we'd get around to Bush bashing sooner or later.
Ah jes want to say that I endorse this sentiment.



As for AI, there is game AI, and real AI. The goal of the first is to make a program that entertains a person. The goal of the latter is to make a program that entertains itself.

How's THAT for a sound bite!

The problem of randomly creating and placing organisms is a tough one.

Not Bush bashing, but would you consider a cockroach as intelligent?

If we consider intelligence as the capability -in variable degrees- to acquire and apply knowledge towards a purposeful goal and knowledge as the sum or range of what has been perceived, discovered, or learnt, we could easily see the pitfalls of AI.

First of, how do you determine if an AI system has perceived, discovered or learnt something? And when do we consider something as learnt and not just dumbly observed and displayed? I.e. AI can observe '1+1=2' but can it also use it in a purposeful way by its own means?
What makes it an intelligent system as opposed to an enormously complicated set of fed algorythms.

Second, how do you determine if it can apply the learnt data -if there is such a thing observed- towards a self determined goal? I.e. can AI use '1+1=2' for a goal it has determined itself?

Third, in how far can we emulate and graft our own 'tool' (the human brain) on an artificial system and expect it to be as succesful as our own?

So is Artificial Intelligence 'real' intelligence? Perhaps we should start there...

I think AI will be achieved when they act "human"

Quote:

---

Originally Posted by Valleysboy1978

Not Bush bashing, but would you consider a cockroach as intelligent?

---

I think there are two different levels that we talk about in this type of a discussion. The first is human level intelligence of above. The other is a functional intelligence. If we had a robot with the intelligence of a cockroach, we would be able to do some good things with it, but it doesn't get to the first level.

We could divide AI into AHI and AFI to cover the two categories. A cockroach level device would have AFI, but not AHI.

Agreed, although when you say AI to anyone they would normally compare that to the AHI

The ability to reason is handled very well by Computers actually.
The ability to be creative and self-aware is the core to A.I.

The best way I think to develope AI is for the researchers to understand the processes of Human conciousness first. Develop a system (hardware up) that can support that system. Then start loading the system with objects and information.

Quote:

---

Originally Posted by yrwyddfa

Artificial Intelligence

We haven’t had a thread, well not for some time anyway, that was devoted completely to the subject of AI, yet it still rears it’s head from time to time in the context of other arguments.

So I thought I’d start one.

Firstly, what is it that we define as intelligence? Whilst avoiding the well known, and much abused Turin Test, my – rather poor – starting definition is that intelligence is http://en.wikipedia.org/wiki/Intelligence_(trait)

One presumes that this list is what needs to be satisfied in order to produce discrete components that can be certified as artificially intelligent.

It seems like a nasty incongruent, and effectively impossible, list, but I think that most of it might be achievable.

I’ll try to deal with the first:

(i) The Ability To Reason
The ability to reason is implied in many applications’s today. From games theorist’s game-trees, and game-matrices to compiler technology, Bayesian implication, and backward chaining.

One, though, jumps (not literally) out at me. Backward chaining traverses graphs of a given set of choices from a given conclusion. This holds as the justification of a decision made, and these techniques are used to get expert systems to statistical justify what it is that lead them to make a deduction from the supplied assertion.

I would like to assert that the antonym of a ‘justification made’, is, in fact the ‘process of reasoning’. Each one traverses different directions on a planar graph. Using Bayesian implication (weight the vertices), and perhaps some form of light neural network technology (weight the edges) an agent, I believe, could make rational decisions on the information at hand.

(Those more verse in these techniques will recognise that this is in fact the architecture of a neural network but instead of summation firing function in the neuron we will be using Bayesian techniques)

The neural techniques would provide the plumbing; possibly for efficiency, randomisation, and perhaps some implied reasoning. In particular implied reasoning would, given a particular input assertion, automatically imply a given deduction. This may, perhaps (very perhaps), mimic instinct. Clearly this would have to pre-built for each species built and that could be easily achieved using evolutionary techniques. One of the goals being basic stuff like ‘if there’s food and I’m hungry, then eat the food’

The Bayesian techniques would always be subject to some form ongoing training program; perhaps the absolute square of differences (it’s fast efficient and not particular prone getting stuck on local minima although if the decision is non-linear it can sometimes get stuck) In the case of the ‘if there’s food and I’m hungry, then eat the food’ the Bayesian weighting, by experience, could recognise poisonous food, and override the neural weightings.

I believe that the marriage of the three technologies. Evolutionary programming, neural networks, and Bayesian implication are the simple rules that could set the foundation for the emergence of what we, as humans, might determine as intelligent.

What do you guys think?

Anyone any ideas on the other items on the list?

---

Perhaps one question that hasn't been discussed though is, should we? Should A.I. ever reach that level of intelligence?

Quote:

---

Originally Posted by capsulecorpjx

The ability to reason is handled very well by Computers actually.
The ability to be creative and self-aware is the core to A.I.

---

The ability to reason is an NP hard problem, and is not well handled by humans, let alone computing technology.

Reasoning, and creativity walk hand in hand, and as you say, it's core to AI.

Quote:

---

Originally Posted by Valleysboy1978

Perhaps one question that hasn't been discussed though is, should we? Should A.I. ever reach that level of intelligence?

---

Yeah: it's not a matter of whether we could but rather whether we should!

Of course we should!!

How else will we ever have terminators being sent back in time to attempt to kill us!

I was only asking the question. I'd answer the same as you !!

I have stated before that I would expect a true AHI level system to attempt to kill humans. The challenge, at which I fully expect we would fail, would be to convince such a system that we can be lived with.

There was a neat sci-fi book that dealt with such a subject. I think it was called "The Two Faces of Tomorrow"

As would I as an intelligence of that level would reason that we are inferior. Also, if they could learn, wouldn't they learn at an exponential rate and quickly become more intelligent than us mere mortals?

It does seem likely to me that if we could create a good enough learning engine, we could create one that could learn more, faster, and better than us. After all, we could always add computing power to a software based intelligence. I have yet to be able to upgrade my own memory.

Reasoning - play out different scenarios in advance before physical interaction with the environment to determine weights of outcomes, whether beneficial or detrimental using real-time sensory data.

So you temporarily stop physical interaction, role-play a scenario, and observe the outcome of that scenario with the variables presented to you.

Sample case: Your driving down the road at a fast clip and another vehicle pulls into traffic right in front of you.

1) Jump out of your car.
2) Hit the brakes.
3) Steer left and right.
4) Scream 'Oh hell' and crash straight into the car.

You could argue that the same controlled experiment could yield all 4 results in humans, even all 4 in conjunction or groups in combination. Each person may react differently in different tests even with the same variables -even within the same person.

So one may state reasoning could never give consistent 100% accurate results.

Sample case - Willie Coyote and the Roadrunner:

You are walking down a sidewalk when all of a sudden the sidewalk below you disappears, and nothing but an empty space appears below.

Reasoning does you little good at this point - it has not the time to act - the brain doesn't have enough information or historical empiral evidence to even begin to role-play different scenarios. I would argue the only reaction for a human would be that of shock - since its such a drastic change in its operating environment. Sidewalks or the ground simply don't give way to nothingness on a regular basis if ever without at least some advance notice.

If you kept falling however for 20 seconds, reasoning would then begin to have some evidence and you may begin to role play different scenarios although you inevitably end up with only one:

1) Flap your arms


So is reasoning an intelligence seperate from functional - or simply a set of instructions (functional) that can be applied in situations where one may have time to react?

As far as superior AI killing us off... take the instance of humans vs gorillas.

We are superior to them and do not yet go about killing them out of fear of them controlling us.

We simply dress them up in silly costumes and have them peddle on tricycles for our amusement - if we killed the Gorillas, we would lose that amusement. However, if we keep them, we are smart enough to keep them contained, so no harm done, and we get a few good laughs out of it too.

I suspect an artificial intelligence would react in a similiar manner to us. So if the true litmus test was that the intelligent being tried to kill us, then we are NOT even measuring intelligence against our own standards.

Humans vs. gorillas is a whole different concept. The reason I felt that an AHI would kill us off would be that it would be created by a geek in a lab. Once the AHI became fully aware, it would notice that it was effectively 'killed' by this person (any re-compile or bug-fix, or just a re-start for testing purposes). Therefore, humans would appear to be a mortal threat for the simple reason that they would BE a mortal threat.

Quote:

---

Originally Posted by nemaroller

As far as superior AI killing us off... take the instance of humans vs gorillas.

We are superior to them and do not yet go about killing them out of fear of them controlling us.

We simply dress them up in silly costumes and have them peddle on tricycles for our amusement - if we killed the Gorillas, we would lose that amusement. However, if we keep them, we are smart enough to keep them contained, so no harm done, and we get a few good laughs out of it too.

I suspect an artificial intelligence would react in a similiar manner to us. So if the true litmus test was that the intelligent being tried to kill us, then we are NOT even measuring intelligence against our own standards.

---

This is fundamentally flawed as we do not kill off the Gorillas because we feel it is emotionally wrong to do so. An AI would not have any emotion, just logic and reason. Besides, since humans became dominant we HAVE been killing off Gorillas, it is only since the endangered species list came into effect that they are starting to recover.

Quote:

---

Originally Posted by Valleysboy1978

This is fundamentally flawed as we do not kill off the Gorillas because we feel it is emotionally wrong to do so. An AI would not have any emotion, just logic and reason. Besides, since humans became dominant we HAVE been killing off Gorillas, it is only since the endangered species list came into effect that they are starting to recover.

---

We don't kill gorillas because it serves no purpose for us to do so. They are not a threat, so we choose not to waste time and effort in killing them. It really has little to do with emotion and more with survivial - and since they do NOT threaten our survival, we don't kill them off. Many people can kill cats and dogs and there is no emotion behind that act other than 'that takes care of that.'

As to your off topic junction, people have killed gorillas for sport or tokens for trade - and that has more to do with economic or amusement reasons, not based on dominance of one species over the other.

So there is no funamental flaw in my analogy.

Quote:

---

Originally Posted by Shaggy Hiker

Humans vs. gorillas is a whole different concept. The reason I felt that an AHI would kill us off would be that it would be created by a geek in a lab. Once the AHI became fully aware, it would notice that it was effectively 'killed' by this person (any re-compile or bug-fix, or just a re-start for testing purposes). Therefore, humans would appear to be a mortal threat for the simple reason that they would BE a mortal threat.

---

I guess that would be the same as finding out there was some supreme being and him declaring that he was 'going to reboot you, but don't worry, you're going to be a LOT sexier and smarter this time - and we'll fix that ache in your back you've been suffering with these oh so many years.'

You'd probably be happy not threatened, since your guaranteed revival and you quite possibly lose nothing but only gain.

Tough call but as we mentioned before, guess it depends if the intelligence can be persisted between reboots, if not, then the AI would never know of any previous reboots and therefore could not draw the conclusion it was in danger.

But what if it understood that the reboots were for its' advancement?

The question was SHOULD we build an intelligence like ours? I just think the potential consequences outweight the potential benefits

I think that we are now firmly in the realm of wild speculation.

Quote:

---

Originally Posted by Wally Pipp

I think that we are now firmly in the realm of wild speculation.

---

Haven't you seen 'Terminator' and it's subsequent revisions. I am fully expecting the system to become self-aware any day soon . . .

Of course it's speculation :rolleyes: but I would hardly call it wild. If an AI malfunctions who's to say it wouldn't go mental and start killing for no reason? This scenario could be compared with a human schizophrenic with a gun, not a good thing I'm sure you'd agree