Will computers ever be smarter than humans?

[someoneis777]

Obviously the topic isn't new...but...

What do you think? In my opinion computers have the potential to be much smarter than us and that is - the reason for this not happening is not the lack of hardware...but simply: perhaps we don't know yet how to program them. To quote Bill Gates: "I've never doubtd in the magic of software". I am a firm believer that AI has to do mostly with the way the program is programmed and not in the number of transistors - actually the whole transistor argument is kind of laghuable, now that i have thought about it for...days and weeks. Hence, the whole question in my opinion is not: Will they become - but: is it wise to create them so smart? What are the real implications of computers destroying Earth once we make them so smart? I think the chance for a dramatic collision between humans and AI is relatively big. It's not the safest bet...but...we won't be humans if we don't try.

 

[JaredJames]

Depends what you consider "smart".

If it's "amount of stored knowledge" then computers are already there. If it's intelligence and ability to apply the knowledge creatively then it's humans.

AI is a completely different beast and not solely limited by hardware. It's being able to program creativity into a computer that makes it difficult.

How do we program a computer to come up with an original thought?

But the issue of processing power is certainly there.

 

[someoneis777]

Depends what you consider "smart".

If it's "amount of stored knowledge" then computers are already there. If it's intelligence and ability to apply the knowledge creatively then it's humans.

AI is a completely different beast and not solely limited by hardware. It's being able to program creativity into a computer that makes it difficult.

How do we program a computer to come up with an original thought?

But the issue of processing power is certainly there.

Appearantly creativity is what makes us humans, or...as i like to joke: what makes us humans is...the orbitofrontal cortex. That is...an area in the brain that is most likely responsible for decision making, reward and many cognitive and "high" thought processes. Nevertheless the very cortex is poorly understood due to its proximity to the eyes and hence the danger of detailed infrared study - but that's another story...
Over the years though, many AI scientists have come with some various ideas, that while chaotic and non leading directly to AI can be the milestone for such a thing:

1. Make it learn from the internet.
2. The human brain changes all the time - the AI should be able to change its own code all the time - a feature known as..."self modifying code" and currently being available only in hard and low level programming languages like assembly. Other languages have functions to do so, but they aren't exactly doing this, rather they have a programming property known as "reflection", which can be viewed as a very limited case of self modifying code. The SMC can be "imitated" very easily though: Just make the program write the new code into a text file and compile the text file - then boom...you have 2 programs instead of one...yes, it's primitive...but it works.
3. Imitate the neural network. Neurons are now understood relatively well and many attempts have been made to simulate their behaviour, "neural networks" - while most of these attempts can be described as failures...it's still a hot topic in AI research these days. In the very best though they are nothing but a very primitive form of AI, nothing like a human level intelligence.


In my opinion an AI should definitely take advantage of the 3 things mentioned above - internet is a good source of information and quite comfortable for a computr, the program should be able to change its own code all the time so something like a "super reflection" or...self mod. code is a must and finally while the way neurons work might never be simulated 100% it's a guideline on how complex logic might be nothing but a sum of simple logic and nothing but a large network.

 

[Evo]

The internet is also the world's largest depository of misinformation. Not a good place to learn from.

 

[JaredJames]

Just a note, the internet is a poor source of information.

Yes it contains a lot of good, but it has a heck of a lot of bad stuff too (if not significantly more).

 

[someoneis777]

The internet is also the world's largest depository of misinformation. Not a good place to learn from.

Totally disagree...
Actually, this is what makes us humans again - the ability to distinguish good information from bad information. Once the AI reaches an "adult" stage, it will most likely have the ability to differ good information from bad information - or have an idea about that. I will not call a place which has so many published papers of Harvard, MIT, Oxford, Tokyo University a "bad source of information". How is the outside world a good source? You have bronx...but you have cambridge too. Is the world outside a bad source following this line of reasoning? Of course...a smart AI will be smart enogh to pick cambridge over bronx just like it will pick www.harvard.edu[/url] over [PLAIN]...censored... . :). I guess, it was my mistake...because i assumed a priori that most people here will be aware that when you program a...program to extract data from the so called "internet" - you don't give it "internet" as a destination...you give it a website. So has the first internet webspider started and so work every example online which shows how to extract data from a webpage. It's like...you decide and can lead your child to study either in bronx or in Oxford. In my case...if I were to create an Ai, I will program the AI to start in www.oxford.edu or any othr big and good website source of information. Once it gains more and more knowledge it will decide on its own where to go and what to do...in the internet(its first world...).

 

[JaredJames]

Totally disagree...

You can disagree all you like but I can Google "tarot cards" and find plenty of information claiming to be factual when it's all a load of nonsense. That applies to many areas such as homeopathy etc.

Actually, this is what makes us humans again - the ability to distinguish good information from bad information.

Actually, humans can't without prior knowledge. We're surprisingly bad at separating good from bad. If you weren't aware of the placebo effect you'd think most herbal cures where miracles in a bottle. People believe things far too easily.

Once the AI reaches an "adult" stage, it will most likely have the ability to differ good information from bad information - or have an idea about that. I will not call a place which has so many published papers of Harvard, MIT, Oxford, Tokyo University a "bad source of information". How is the outside world a good source? You have bronx...but you have cambridge too. Is the world outside a bad source following this line of reasoning? Of course...a smart AI will be smart enogh to pick cambridge over bronx just like it will pick www.harvard.edu[/url] over [url]www.seeboobs.com[/url]. :). I guess, it was my mistake...because i assumed a priori that most people here will be aware that when you program a...program to extract data from the so called "internet" - you don't give it "internet" as a destination...you give it a website. So has the first internet webspider started and so work every example online which shows how to extract data from a webpage. It's like...you decide and can lead your child to study either in bronx or in Oxford. In my case...if I were to create an Ai, I will program the AI to start in [url]www.oxford.edu[/URL] or any othr big and good website source of information. Once it gains more and more knowledge it will decide on its own where to go and what to do...in the internet(its first world...).[/QUOTE]

Noone said the outside world was any good. If anything, it comes in a close second to the internet.

Again, we are not good at distinguishing good from bad information. If I provided you with ample evidence of my ability to read minds, would you believe it? Of course you would. It's only once you've gained initial knowledge of cold reading and other techniques you can see the 'bad'.

Please remove the link to an adult site.

 

[someoneis777]

This "source of knowledge" argument is somewhat circular...again I can counter and say that you can go to any library and see 99% of the tarrot related books describing tarrot as something very useful. PLus, I gave harvard.edu as an example...whereas I am almost sure that tarrot aren't as glofiried in such websites. And frankly, my best candidate for a "starting point" would be a website like wikipedia.org which has relatively trustworthy information...in what a quantity.
Again...I am surprised that the source of knowledhe turns to be a great problem here, if that was the real problem - all the Japanese robot designers would simply create a robot, put it in the middle of street and start taking pictures (like..japanese tourists...) while hopign that the robot will reach a human like intelligence simply because it's exposed to fresh air...untill it gets hit by a car of course...(it's in the middle of street... :). Seriously, the hardest task is and will always be: perception. How does the computer interpret information in a human like way or in a way superior to us.Even a single sentence such as: "Humans are mammals" whether true or not should be analyzed in a way much different than simply entering this sentence into Oracle software database.

 

[JaredJames]

This "source of knowledge" argument is somewhat circular...again I can counter and say that you can go to any library and see 99% of the tarrot related books describing tarrot as something very useful.

Yep, as I noted in my post. We need to be taught.

PLus, I gave harvard.edu as an example...

And I can give you 20 sites that dispute what is on the Harvard site. It means nothing. In the case of the internet, quantity is outweighing quality badly.

whereas I am almost sure that tarrot aren't as glofiried in such websites.

Glorified, maybe not. But given the number of people who believe tarot I have no doubt the tarrot sites have a larger user base.

And frankly, my best candidate for a "starting point" would be a website like wikipedia.org which has relatively trustworthy information...in what a quantity.

So you'd point the device to specific websites. Ok, but then how does it learn? Learning =/= copying.

Again...I am surprised that the source of knowledhe turns to be a great problem here, if that was the real problem - all the Japanese robot designers would simply create a robot, put it in the middle of street and start taking pictures (like..japanese tourists...) while hopign that the robot will reach a human like intelligence simply because it's exposed to fresh air...untill it gets hit by a car of course...(it's in the middle of street... :).

Robots can't evolve - at least not in the same manner as biological creations. So let's not confuse the two.

Seriously, the hardest task is and will always be: perception. How does the computer interpret information in a human like way or in a way superior to us.Even a single sentence such as: "Humans are mammals" whether true or not should be analyzed in a way much different than simply entering this sentence into Oracle software database.

Yes it should, but being able to 'read' input and provide an answer isn't major problem. We already have systems that can do that in a basic sense.

The key is being able to come up with creative ideas and generate thoughts. Get a computer to ask what the purpose of its life is without ever putting the 'idea' into its 'head' - that's what humans did.

 

[Chronos]

The internet is also the world's largest depository of misinformation. Not a good place to learn from.

Discerning signal from noise is the essence of human intelligience, evo.

 

[JaredJames]

Discerning signal from noise is the essence of human intelligience, evo.

But that's just it, it's a catch 22 situation.

You need to be taught the good and only once you have the background knowledge can you work through the internet.

I can spot pseudoscience a mile off, but that's only because I have a base knowledge in physics, chemistry and biology.

The problem you face is when you have a claim that doesn't involve the above, it's purely spiritual - how does the computer read that and realize it's BS?

I think I'd be more impressed with a computer that would see something at the beginning of its learning curve, such as "healing angels", believe it if it had no reason not to and then as it learned other things it would adjust its beliefs - hopefully to get rid of it - but at the same time, I'd say a computer that takes on beliefs such as those would be extremely close to human.

Naturally it's something we'd have to program out because we don't want the non-sense being kept.

 

[Futurama]

Probably the best recent example is IBM's 'Watson' that appeared on Jeopardy. It stored all this stored information on its 16TB of RAM and had lots of processors to sort through that information. What the most impressive thing about this was the software developed for it was able to discern human humor, and slight cues and actually beat other humans. Does this mean that it is more intelligent than a person? No. But it does mean that one day, an A.I. will be able to understand -- and replicate humor and emotions.

 

[Chronos]

Computers are unable [at present] to self actualize, nor have their 'brains' been forged by the trials of natural selection. They have no sense of self or when to turn on the afterburners in their decision making process. We humans will lie, cheat, steal, run, hide, fight, or even die heroicly to ensure our and ours survival. It is unclear how you would program a computer to mimic this behavior - and probably not a great idea. I would be somewhat alarmed if I hit the 'turn off computer' button and the message "I'm sorry Dave, I'm afraid I can't do that." appeared.

 

[Futurama]

Computers are unable [at present] to self actualize, nor have their 'brains' been forged by the trials of natural selection. They have no sense of self or when to turn on the afterburners in their decision making process. We humans will lie, cheat, steal, run, hide, fight, or even die heroicly to ensure our and ours survival. It is unclear how you would program a computer to mimic this behavior - and probably not a great idea. I would be somewhat alarmed if I hit the 'turn off computer' button and the message "I'm sorry Dave, I'm afraid I can't do that." appeared.

Love the '2001' reference. It is possible to simulate this and have computers show decision making processes - but only at a simple layer.

The only way I can see this becoming possible is perhaps integrating organic life with computers. Scientists have already made microbes act as hard drives and store information.

 

[Chronos]

Computers have already been used to restore at least some functionality to paralyzed limbs. Implanting computers to facilitate or enhance human performance in general is already being done on a limited basis [e.g., pacemakers], and is an active area of research. Computer assisted human intelligience already exists, we just have not yet implanted it [as far as we know]. 'Resistance is futile' looks like an increasingly plausible evolutionary path.

 

[Futurama]

Computers have already been used to restore at least some functionality to paralyzed limbs. Implanting computers to facilitate or enhance human performance in general is already being done on a limited basis [e.g., pacemakers], and is an active area of research. Computer assisted human intelligience already exists, we just have not yet implanted it [as far as we know]. 'Resistance is futile' looks like an increasingly plausible evolutionary path.

This will certainly make the future more interesting...