Science Thoughts on Artificial Intelligence

[cjp]

Or do it like good ol' Asimov...

Problem is, many ethics principles, including Asimov's laws, are too vague to be translated into reliable software. The program might do things that are, according to its own interpretation of the laws, perfectly acceptable, but that still aren't really what we human beings want.

When it comes to human beings, we want them to have freedom, because we are human beings, and we want freedom for ourselves. So we just have to trust each other to a certain degree. But there is no reason to give an AI the same amount of freedom. So, for now, a 'jail' is the right place for such a machine. For as far as its owner does give it freedom, responsibility for the consequences is fully on the owner.

I think Artlav made a nice description of the actual problems. I do not wish to create a 'neuron simulator', for performace reasons. So I get stuck with the 'algorithm writer' model. At first, I just want to write an 'algorithm translator', without any innovative power. A bit like an advanced compiler. The actual 'inventing' will probably involve a lot of heuristics (I think this is also how human beings do it). With this included, it will really be a Heuristically programmed ALgorithmic computer.

 

[Moonwalker]

As long as we don't understand the human brain and how consciousness works, we won't be able to create a machine or software that has similar capabilities.

 

[Andy44]

As long as we don't understand the human brain and how consciousness works, we won't be able to create a machine or software that has similar capabilities.

True, if you're trying to build an artificial human mind. But if you just want an intelligent machine, you just need to build the basics and let it learn or adapt to get the rest on its own.

 

[Max Pain]

Just something that doesn't get out of my mind when reading this thread:

How can someone proof that a machine is really consciouss? I mean you can just program the machine so it behaves like a consciuss being.
Would that be the same?

 

[Artlav]

How can someone proof that a machine is really consciouss? I mean you can just program the machine so it behaves like a consciuss being.
Would that be the same?

What is consciousness?
How does a conscious being behave like?

No one around can answer these questions, so it's a case of making that - don't know what.

Most likely they'll know it when they get there.

 

[Tschachim]

How can someone proof that a machine is really consciouss? I mean you can just program the machine so it behaves like a consciuss being. Would that be the same?

Yes: http://en.wikipedia.org/wiki/Turing_test

 

[Quick_Nick]

Yes: http://en.wikipedia.org/wiki/Turing_test

Indeed. By that test, it seems that I do simulate conciousness well enough to appear human.

 

[RisingFury]

As long as we don't understand the human brain and how consciousness works, we won't be able to create a machine or software that has similar capabilities.

To some degree we already have intelligent programs. I think one of the best examples of these are FPS games.

I remember during the UT3 hype, one of the AI coders did an interview and reviled that once the coding was complete and testing began, the bots appeared to cheat. So they started observing the bots in game and it turned out that they've learned to exploit map bugs.

After that, they had to dumb the bots down a little bit.

They designed the AI to learn. It keeps track of pickups respawning, it avoids areas where it's been killed often before and it plans ambushes in areas that it scored many kills.
The AI is able to do that on any map.

 

[Urwumpe]

But is that already intelligence - can the AI do, what the programmer didn't think about?

 

[garyw]

Point 1:
What about the safety of humans. If the robot is "intelligent" like humans, it may think it's better than humans and attempt to take over the nation and kill anyone who tries to interfere. We can't have an army of these robots deciding they are better than humans because they are intelligent and taking over a nation or the planet. They would have to have some protocol on how to act. Like not causing a human to die, or just sitting there while a human is dying. Rule number one of robotics clearly states that a robot cannot cause a human to die. The next rule clearly states that a robot must do what it can to prevent a death.

Only in the world of Issac Assimov.

 

[Jarvitä]

Consciousness, HAL-9000 style, as depicted in 2001, is probably not possible, unless HAL was just emulating human characteristics.

Human consciousness, the human mind, is integrated with the physical body. A disembodied spirit like HAL has no real body, etc

No.

The human brain is, in essence, a massively parallel computer with a terribly inefficient power supply. If you have sufficient computing power, you can easily emulate it. The rest of the hardware you were talking about can be emulated even easier. Once you have that in place, there's no reason an emulated human couldn't grow mentally in the same way we do.

We already have plenty of software that was referred to as "artificial intelligence" before it was actually written - but once its inner works are figured out (sometimes through human brain observation), those advances are relegated to less-cool sounding designations such as speech recognition and natural language parsing. What we understand as consciousness is just another area of the mind that will sooner or later be mapped out and replicated in software form, either before or shortly after we have the capability to emulate the entire brain.

Kurzweil et al. estimate, based on Moore's laws and the exponential growth observed in other areas of technology, that we should reach that point of progress sometime in the middle of the 2020s.

 

[Moonwalker]

Yes, we have intelligent machines and programs. But if I understand it right, this thread is about computer programs that could become similar to a human being. This is not possible if we don't understand the human brain and consciousness. And we are still at the very beginning of trying to do so.

Compared to human beings, all our machines and software just is rudimentary toys. I'm not impressed of computers and robots. I'm impressed of what the nature does, and even more how and why...

---------- Post added at 11:59 PM ---------- Previous post was at 11:50 PM ----------

What we understand as consciousness is just another area of the mind that will sooner or later be mapped out and replicated in software form, either before or shortly after we have the capability to emulate the entire brain.

Can you explain the another area of the mind and how it is going to be mapped out?

The brain ties informations, learns and changes. There is not a lot that could be emulated, especially not software. There is no software within the brain.

 

[Andy44]

Just something that doesn't get out of my mind when reading this thread:

How can someone proof that a machine is really consciouss? I mean you can just program the machine so it behaves like a consciuss being.
Would that be the same?

Like I said before, from my point of view, the only thing I am certain is conscious is me.

I don't know if you're conscious, even though you are clearly human and walk and talk, how do I know that you really anything more than a meat machine which is programmed to think its conscious?

What you're asking is a philosophical question. It can really only be speculated on, but if the machine can talk it can at least claim it's conscious.

Your cat isn't capable of speech, nor does it have the memory or cognition of a human, but you are fairly certain it is conscious and self-aware, so why would you treat a machine any different?

 

[Hielor]

Point 2:
Are these robots going to replace jobs that can be done efficiently by humans. Humans are actually more efficient than humans, and humans need to make a living. If you replace humans with robots, you'll kill the economy (or what's left of it) and poverty will skyrocket (not that it hasn't already happened) because it's "cheaper" to use robots. Plus, there are some moral values that a robot would never be able to process. Like some things MacGyver did to catch criminals and/or save lives. AI, and even some cases of non-AI machines are a bad idea.

I agree! We should do everything we can to prevent robots from taking human jobs, like in car factories, where humans....oh, wait. That's already done largely by robots. This hasn't killed the global economy, it's made cars safer and more readily available.

Moreover--MacGyver? Really? I would suggest you stop learning about how the world works from fictional movies and tv shows.

 

[Moonwalker]

Your cat isn't capable of speech, nor does it have the memory or cognition of a human, but you are fairly certain it is conscious and self-aware, so why would you treat a machine any different?

A machine does not feel. It has no sense organs and does not react to the environment the way life does. Machines just do what its software and plain sensors are supposted to do. Software and machines won't ever have consciousness because metall, plastics, rubber, silicon, ventilators, laser and electricity can't feel, think and, combined or not, won't become alive.

How sad humans have to be be to think machines could be something like humans. Why not start with a plain dog or fish? Even this doesn't work. There'll always be a difference between machines and life. Except that machines at least can be used to pretend something. Machines are just tools. No more, no less.

 

[Zatnikitelman]

Actually the Mythbusters pretty much busted all of MacGyver's stuff. The sodium required was too much, the plane couldn't make a powered cliff-takeoff (though I think they messed something up). The other stuff was simple little stuff like signals, photographic development and lockpicking that weren't seen first on MacGyver.

 

[Jarvitä]

The brain ties informations, learns and changes. There is not a lot that could be emulated, especially not software. There is no software within the brain.

Wrong, we just don't have sufficient brain-scanning technology to tell the differences between "software" and "hardware".

And, any computational hardware can, by definition, be emulated if it's understood sufficiently. And once it's understood sufficiently, it can also be optimised to run the same processes using up less computing power - a step we mostly skipped during the natural evolution of our brain. I've no doubt that an optimised emulation of the human brain would reduce the amount of computing power needed by at least a couple orders of magnitude compared to a simulation of each individual neuron.

Also, regarding conscience -consider this hypothetical scenario:

You have a mechanical computing machine (cogs and levers) that's turing complete. Then you stack enough of them in a parallel configuration that they have sufficient power to run the brain emulation (granted, the volume of such a setup would likely exceed earth's). Now you feed it input from the "senses", in form of binary punch cards it interprets as sight, hearing, smell, etc. Then, it runs the stimuli over it's "consciousnes" algorithm and produces output in a similar format. Is this machine conscious and humanlike? If not, what makes a few billion neurons more conscious than a few trillion cogs and levers that compute the same algorithms?

 

[Moonwalker]

As I said: metall, plastics, rubber, silicon, ventilators, laser, electricity and so on... (like your cogs and levers as well) can't love, can't feel, can't think and can't develope consciousness. Computers are no organisms. They are no creatures. They are indeed pretty death and absolutely non-amazing compared to what the nature offers. No matter how perfectly you simulate a human brain, it will remain death and do nothing more than just convey the impression of being similar to a real human brain.

 

[Hielor]

As I said: metall, plastics, rubber, silicon, ventilators, laser, electricity and so on... (like your cogs and levers as well) can't love, can't feel, can't think and can't develope consciousness. Computers are no organisms. They are no creatures. They are indeed pretty death and absolutely non-amazing compared to what the nature offers. No matter how perfectly you simulate a human brain, it will remain death and do nothing more than just convey the impression of being similar to a real human brain.

If metal, plastics, rubber, and silicon cannot love or feel, how then do carbon chains do it? Without invoking magic.

 

[Moonwalker]

If metal, plastics, rubber, and silicon cannot love or feel, how then do carbon chains do it? Without invoking magic.

Do carbon chains feel?