On Consciousness

[Leumas]

It's not a matter of conjecture. It's a matter of computability theory. If the task of interpreting these images was not a computable function it would theoretically impossible for anyone or anything to interpret them. Since humans can interpret them, then this must be a computable function, and therefore it is theoretically possible for a computer to do this.

Whether or not it would be practical to program a computer to do this is a different question entirely. Possibly it might take an absurd amount of processing power or time (which amounts to the same thing, as double the amount of time is equivalent to double the amount of processing power), or it might require programming in vast amounts of contextual knowledge and understanding that humans unconsciously acquire in childhood.

But the fact that the USA Military is involved in this is completely irrelevant to the discussion.

​

 

[Mr. Scott]

My Growing Collection of Magic Beans

I've decided to collect magic beans. Aside from the master magic bean of consciousness, we've heard of:

1) The Magic Bean of Smartness.
2) ...of Image Recognition (i.e.enemy hideouts in satellite pics).
3) ...of Creating Fine Art.
4) ...of Understanding (i.e. the Chinese Room experiment)

On #1, it depends on how smartness is defined, but a quick google yields "Characterized by sharp quick thought; bright." That's how I'd describe Watson, though one could argue it doesn't have the Magic Bean of Thought.

On #2, I've worked on many image recognition projects and knew some masters of the craft. Some of the hardest image recognition problems are regularly cracked with neural network simulations. They work!

On #3, I once proposed a program that generated animated cartoons in real time. That's exactly what computer games are doing now, and getting better at it every day. Kids sit and watch self-play mode for hours.

It will be a long time before computers have a shot at achieving the finest of fine art. I don't know how one could argue that they NEVER could (except by invoking the magic bean argument).

On #4, I killed the Chinese Room Experiment in that layman thread and would gladly repost here if anyone's interested.

 

[Leumas]

This is about god? Elaborate.

Your obsession with attributing magic beans (as evinced by your post below) to anyone and everything that is beyond your realm of understanding is a very religionist characteristic you must have retained from your woo days.

Just as when you used to be woo you could not fathom how anyone does not have woo and accordingly you fallaciously attributed to them woo albeit of a type you spurned.

Just like an ex-alcoholic remains forever obsessed with alcohol and struggles with rejecting the temptation on daily basis.

I've decided to collect magic beans.

Are you sure this is wise? I think an ex-woo is better off not tempting himself with collecting woo even if it is woo only in his obsessed imagination .

 

[punshhh]

Unless you are The One

The fifth element?

 

[keyfeatures]

On #1, it depends on how smartness is defined, but a quick google yields "Characterized by sharp quick thought; bright." That's how I'd describe Watson, though one could argue it doesn't have the Magic Bean of Thought.

I suspect anyone going on a date with Watson based on the description of being sharp, quick thinking and bright might be sorely disappointed...

Perhaps my standards are too high?

 

[Leumas]

I suspect anyone going on a date with Watson based on the description of being sharp, quick thinking and bright might be sorely disappointed...

Perhaps my standards are too high?

 

[Mr. Scott]

Your obsession with attributing magic beans (as evinced by your post below) to anyone and everything that is beyond your realm of understanding is a very religionist characteristic you must have retained from your woo days.

Just as when you used to be woo you could not fathom how anyone does not have woo and accordingly you fallaciously attributed to them woo albeit of a type you spurned.

Just like an ex-alcoholic remains forever obsessed with alcohol and struggles with rejecting the temptation on daily basis.




Are you sure this is wise? I think an ex-woo is better off not tempting himself with collecting woo even if it is woo only in his obsessed imagination .

Cute, clever, creative, completely delusional.

The debate reminds me of the year-long one I had with a creationist here, Kleinman.

Like creationists, the anti-computationalists have no evidence for what they believe in and can only throw stones at their opposition.

There's a huge body of evidence supporting the view that the brain is a data processing machine, and no evidence otherwise, AFAIK. As always, you're invited to prove me wrong.

That AI has not yet achieved the touchdown of creating a human-like mind in a machine is not evidence they never will. I've been around long enough to see anti-computationalists run down the field carrying the goal posts away from the AI team again and again. I'm sure that will never end. When AI makes its touchdown, I don't doubt the opposition will only acknowledge the achievement of a Philosopher's Zombie, cognitive dissonance being what it is. At that point, it will no longer matter.

Leumas, if you have positive evidence for whatever it is you believe in regarding the uncomputability of the mind, then share it with us, will you?

And, if you don't like the phrase "magic bean" maybe you can suggest another that appeals to you so we can both use the same name for whatever it is that the brain purportedly has that no machine has without stirring emotions. How's "metaphysical attribute?"

 

[Roboramma]

​

What do you find contradictory in what he wrote?

The "might"s in that post are related to the specific ways that the hardware and software of intelligence may have to be designed, not general point (which he said is not conjecture) that it could be done by a turing machine.

He basically said: we don't know specifically how it will be implemented, but however it is implemented it will be a turing machine. What's contradictory there?

(and if you respond, could you avoid yelling? I have sensitive ears)

 

[Roboramma]

The question to me isn't even if what the brain does can be done by a sufficiently advanced computer. I suspect it can, but let's assume for the sake of argument that there's something else involved. That penrose is right, for instance, and some sort of quantum mechanical properties are involved in the functionality.

So, one day we figure this out and understand how the brain works. Can we build a machine with different parts that works by the same principles? If not, why not?

And if we can, then we can build a robot, or a computer that happens to work a little differently than the ones we currently have, that works as a human brain does.

Clearly it doesn't have to be an exact replica of a human brain: every brain is slightly different from every other, but we still consider each other to be conscious.

 

[quarky]

Can we make an artificial atom?
Or do we need to use real atom parts to do it?

 

[Brian-M]

You are conveniently EQUIVOCATING the word simulation now.... this is disingenuous to the extreme.....you are relying on the evasiveness so far by computationalists to accept the DISTINCTION between simulation and emulation. I think this has been a DELIBERATE tactic and a very dishonest one.

Clearly you don't understand what these words actually mean. Let me correct that for you...

Emulation: Running a program or other software designed for a different system.
Simulation: Something which simulates a system or environment in order to predict actual behaviour.
Simulate: To model, replicate, duplicate the behavior, appearance or properties of.

Do you think there are computerized replicas (emulations) of cells?

Yes, although precise whole-cell molecular model is still being developed.

It's not a matter of conjecture. It's a matter of computability theory. If the task of interpreting these images was not a computable function it would theoretically impossible for anyone or anything to interpret them. Since humans can interpret them, then this must be a computable function, and therefore it is theoretically possible for a computer to do this.

Whether or not it would be practical to program a computer to do this is a different question entirely. Possibly it might take an absurd amount of processing power or time (which amounts to the same thing, as double the amount of time is equivalent to double the amount of processing power), or it might require programming in vast amounts of contextual knowledge and understanding that humans unconsciously acquire in childhood.

​

Presumably you believe the words you've bolded and highlighted are in contradiction with each other (which would be strange because my clarification that the second paragraph addresses "a different question entirely" is part of the highlighted text).

If so, this would be further evidence of your inadequate reading comprehension skills. I see no reason to continue arguing with someone either unable or unwilling to comprehend plainly worded posts.

 

[epepke]

That isn't a full representation of the industry, though.

In fact every place I have worked there is a constant argument between AI programmers on whether we should concentrate on making the AI scriptable, so designers can get the exact presentation they want ( cinematic games like Modern Warfare ) or making the AI very intelligent and organic, so players can experiment with it and get a more real feeling from it ( open world games like Grand Theft Auto 4, sim games like The Sims, and even some FPS games like Half Life 2 or Crysis ).

It usually ends up being a compromise, with us putting in pretty intelligent behavior in some aspects that end up being entirely scripted anyway due to the designers having the final say in the matter. But that doesn't mean the AI doesn't get more and more advanced each time.

You're right. It doesn't mean that's the AI doesn't get more and more advanced each time.

It doesn't mean that it does get more and more advanced each time. If you want to support the idea that it does, I'm all ears.

Now, funny thing is that I myself was thinking about GTA 4. I actually don't find the AI impressive, and it doesn't seem that open-ended to me. Some of the old fantasy RPGs on the Apple II seemed to me more open-ended and in many ways smarter.

 

[rocketdodger]

You're right. It doesn't mean that's the AI doesn't get more and more advanced each time.

It doesn't mean that it does get more and more advanced each time. If you want to support the idea that it does, I'm all ears.

I am one of the people programming it ! Of course it gets more advanced! Just maybe not in ways you would want, but that is out of my control. In games about combat, that is what gets the brunt of the work. And most players don't even notice, unfortunately.

I spent about a month getting the "green despoiled" character in Wolfenstein to make use of splash damage. He scans the environment around a target and evaluates whether there are any spots nearby where the explosion from his fireballs would at least cause damage, and it works really well. You can be crouching behind cover and he will aim at the wall behind you. You can be hiding behind a corner and he will aim at the ground he can see. You can be running along in a hallway and he will shoot through windows where he can, leading your movement. You can be ducking on a rooftop and he will shoot up at objects above and behind you -- anything he can see that will work. Basically, exactly as a good human player would.

He keeps track of how and when he has hurt you, so it is literally impossible to cheese him. If you are hiding in a corner, and he just can't get at you from where he is, he will try something else, like moving to another spot. And if that doesn't work, he actually runs away and waits for *you* to make the next move. On the other hand if he has been doing well, he keeps on doing the same thing.

But do any players notice all this? I don't think so. They just switch to a strong weapon and kill him as fast as they can, rather than dancing with him like I intended. Oh well. But -- he is still very smart, and a far better combatant than any AI in any game prior. Thus, game AI definitely is improving.

EDIT -- I just noticed that this is the guy in my avatar, lolz.

Now, funny thing is that I myself was thinking about GTA 4. I actually don't find the AI impressive, and it doesn't seem that open-ended to me. Some of the old fantasy RPGs on the Apple II seemed to me more open-ended and in many ways smarter.

Well you have to evaluate it in terms of the areas that the devs put work into. In the case of GTA, that is obviously violence and general mayhem.

If you experiment with harassing the AI, they have some pretty deep animation and behavior. You can shoot or injure them in any part of their body and they react appropriately. When you are in combat they gang up on you and flank you really well. If you bump into them repeatedly they fight with you. If you carjack them they fight with you, trying to pull you out of their car. And when you do that, you can repeatedly stop and go, dragging them as they hold onto the door ( admittedly not smart, but deep at least ).

I could reference Skyrim as well, which has pretty good open ended AI.

And what you need to remember is that all this AI is going on while competing with CPU cycles in a measly 3 core chip against the graphics, streaming, physics, and everything else. In games like GTA4 and Skyrim, it is literally a marvel of engineering to see what comes out of those old xbox 360s. If we could devote nearly the whole CPU to a single AI it would be pretty amazing results I think.

 

[Leumas]

Edited by Gaspode: 

Removed breach of rules 0&12.

 

[Mr. Scott]

#5 The Magic Bean of Experience, found in another thread!

 

[Mr. Scott]

That isn't a full representation of the industry, though.

In fact every place I have worked there is a constant argument between AI programmers on whether we should concentrate on making the AI scriptable, so designers can get the exact presentation they want ( cinematic games like Modern Warfare ) or making the AI very intelligent and organic, so players can experiment with it and get a more real feeling from it ( open world games like Grand Theft Auto 4, sim games like The Sims, and even some FPS games like Half Life 2 or Crysis ).

It usually ends up being a compromise, with us putting in pretty intelligent behavior in some aspects that end up being entirely scripted anyway due to the designers having the final say in the matter. But that doesn't mean the AI doesn't get more and more advanced each time.

Mr. Scott "likes"

Perhaps it's about job preservation.

Advanced games will be able to anticipate and manipulate the emotions of players, creating graphics, sounds, music, models, and stories as they execute, reducing or eliminating the need for designers who aren't able to encapsulate their design strategies in the game software.

I actually did work on software that composed and arranged music in real time as a game was played. Work of this type has been going on since the 50s.

 

[keyfeatures]

I spent about a month getting the "green despoiled" character in Wolfenstein to make use of splash damage. He scans the environment around a target and evaluates whether there are any spots nearby where the explosion from his fireballs would at least cause damage, and it works really well. You can be crouching behind cover and he will aim at the wall behind you. You can be hiding behind a corner and he will aim at the ground he can see. You can be running along in a hallway and he will shoot through windows where he can, leading your movement. You can be ducking on a rooftop and he will shoot up at objects above and behind you -- anything he can see that will work. Basically, exactly as a good human player would.

Wow, he sounds dreamy. Good with his hands 'n' stuff. What happens if you invite him in for a coffee? Or ask him if he considers there to be such a thing as a number without awareness? Or whether he really wanted to be a splash damage specialist when he grew up?

 

[keyfeatures]

So, one day we figure this out and understand how the brain works. Can we build a machine with different parts that works by the same principles? If not, why not?

Plato's cave.

 

[Mr. Scott]

One more question, Leumas. You brought up god in a thread about consciousness. Is the idea that a machine could be conscious in conflict with your spiritual beliefs?

 

[annnnoid]

Wow, he sounds dreamy. Good with his hands 'n' stuff. What happens if you invite him in for a coffee? Or ask him if he considers there to be such a thing as a number without awareness? Or whether he really wanted to be a splash damage specialist when he grew up?

There is this phrase in England….’taking the piss’ I think it is. But just the same, I thought I’d take a stab at your questions.

Q: What happens if you invite him in for a coffee?
A: He will fireball you to death.

Q: Or ask him if he considers there to be such a thing as a number without awareness?
A: He will fireball you to death.

Q: Or whether he really wanted to be a splash damage specialist when he grew up?
A: He will fireball you to death.

Dreamy indeed! I guess we’ll just have to take it on faith that those in the AI community are actually aware of the shortcomings of their creations.

Here’s an academic question: If AI is modeled on human consciousness / behavior (presumably), would not an appreciation of AI shortcomings be directly related to an ability to appreciate the equivalent shortcomings / dysfunctions in the human version.

I mean…should we assume that Chomsky was right in his characterization of scientists as being some of the last people you should rely on to adjudicate the verities of human nature.

So can we expect AI ‘personalities’ that reflect the various neurosis of their creators… either individually or collectively? Or can we expect AI personalities that achieve God-like insight, wisdom, maturity, and compassion (the question of the degree to which such things can be instantiated we’ll leave for another day)? Hmmmmmmmmmmmmmmmm…..

Just a question.