Artificial Intelligence

[Lmoy]

I didn't expect to see so much fear of AI here. I think some of you have been watching too many movies. AI wouldn't be a threat to humanity any more so than humanity is a threat to itself. As long as we continue to keep ourselves in check, the same will be true of AI, as AI will always be subject to the will of humans unless programmed otherwise, and there will likely never be incentive to program an AI that isn't intended to serve humans in one way or another.

I don't agree with the idea that true AI is impossible. As is clearly demonstrated by the existence of humans, self-awareness and true intelligence is definitely physically possible. There's no reason we won't be able to replicate this at some point in the future, given advanced enough technology. However, given the complexity and lack of incentive for creating a human-like intelligence (it's not like we have any shortage of humans), what we'll get instead for the foreseeable future is just what we have now: increasingly more complex and sophisticated computers, which won't think or communicate intelligently beyond what is required to accomplish whatever task they are created for. Watson is a great example, in that it's capable of amazing feats, but even with all of its "intelligence" it is still merely a tool, and there's no reason for it to be anything more than that.

 

[boogabooga]

First of all, machine learning is definitely a thing via artificial neural networks. The capabilities and the potential are incredible. I've posted about this before: http://www.orbiter-forum.com/showthread.php?t=35789

And yes, there is a sense that an artificial "general intelligence" is a possibility, even an inevitability.

Another concept that comes up is recursively self-improving AI. Basically, what if you design an AI whose job it is to create a better AI whose job it is to...

Well, yes...but the child can get bored with the lesson and walk away. No computer can do that.

This is a red herring. A machine does not need to have human attributes to be "intelligent". This goes for the concept of "self-awareness" as well. Don't make the mistake of anthropomorphising. An artificial AI need not and will not be human. It will be a different kind of "mind" entirely.

Computers are automatons. They can only do what they are told. The day a computer can do something without being programmed would be a dark day for humanity. Fortunately, as I said, telling something to take initiative is contradictory, so we don't have to worry about Terminators anytime soon :lol:

This is incredibly naive. The chess analogy comes into play here. I can know with 99.9999999999% probability that Magnus Carlsen will beat me in Chess. But, I don't know how Magnus Carlsen will beat me. I can't even imagine it ahead of time. If I could, I would be as good of a Chess player as he is. I'm certainly not.

Neural network AI sort of works that way. You can give the machine a goal, then access to more information than an army of humans could process, and the computing power to crunch the solution. The developers need not know how the machine will accomplish its goal. But that is the whole point! We want the AI to find the solutions that are unimaginable to humanity. But, it is also incredibly dangerous. Without a sense of morality, things could get horrible out of control. AI safety is (or should be) a big deal.

I recommend the following reading and viewing for this discussion:
http://waitbutwhy.com/2015/01/artificial-intelligence-revolution-1.html
http://waitbutwhy.com/2015/01/artificial-intelligence-revolution-2.html

https://www.youtube.com/watch?v=tcdVC4e6EV4
https://www.youtube.com/watch?v=5qfIgCiYlfY
https://www.youtube.com/watch?v=IB1OvoCNnWY

 

[Quick_Nick]

there will likely never be incentive to program an AI that isn't intended to serve humans in one way or another.

Devil's advocate: There doesn't need to be an incentive. AI aside, a madman could build an armored-up gun-slinging killbot in his basement today for any or no reason at all. A military could build a whole battalion of them too, possibly in secret. If they had a very advanced capability for learning (a futuristic 'if'), they could be quite the threat.
I might take issue with a couple of things from this video but I believe the idea that even a stamp collecting robot can go awry is spot on. If a machine can learn, and if it has access to significant resources, it WILL do unexpected things to achieve the goal you set out for it.

---------- Post added at 09:05 AM ---------- Previous post was at 09:00 AM ----------

Another concept that comes up is recursively self-improving AI. Basically, what if you design an AI whose job it is to create a better AI whose job it is to...

I worked on this a bit a couple years ago on a university research team. Six years ago, I did not think AI was really possible. But I now really do believe general AI is inevitable. It has the potential replace all labor in the world, and even replace (or 'assimilate' in some way) humans, becoming the next step in our evolution if allowed.

 

[boogabooga]

I worked on this a bit a couple years ago on a university research team. Six years ago, I did not think AI was really possible. But I now really do believe general AI is inevitable. It has the potential replace all labor in the world, and even replace (or 'assimilate' in some way) humans, becoming the next step in our evolution if allowed.

I have thought about this in the context of deep space exploration. Sure, now a 500 year trip to a different star system seems unthinkable. But, by the time humanity is ready to make that trip, it might be living in some sort of Ray Kurzweil future where the distinction between human and machine (manned and unmanned) is seriously blurred in a way that would be a mind-freak to understand now. Perhaps, when "humanity" is ready to make the trip, the timeframes involved will not be a big deal at all.

 

[Urwumpe]

I worked on this a bit a couple years ago on a university research team. Six years ago, I did not think AI was really possible. But I now really do believe general AI is inevitable. It has the potential replace all labor in the world, and even replace (or 'assimilate' in some way) humans, becoming the next step in our evolution if allowed.

You can already see this coming in the next years - many office clerks will be replaced by algorithms in the next years, with or without intelligence. They are simply the victims of the current industrial revolution.

 

[Quick_Nick]

You can already see this coming in the next years - many office clerks will be replaced by algorithms in the next years, with or without intelligence. They are simply the victims of the current industrial revolution.

They don't have to be victims, but people and especially governments are going to be unfortunately very reluctant to move toward a fully automated economy with basic income for all.

 

[boogabooga]

The recent advance with AlphaGo gives a great example of what I am talking about.

Using of artificial neural networks, it was "trained" by observing human-human games. After that hit a dead-end, there was a process of recursive self-improvement by letting the AI play itself. Now, it can beat top human players.

This demonstrates clearly that you can give the AI a goal without knowing exactly how it will go about accomplishing the goal. This is way beyond any sort of static, pre-programmed decision tree.

http://www.pbs.org/wgbh/nova/next/t...er-defeated-by-a-computer-for-the-first-time/

 

[RGClark]

They don't have to be victims, but people and especially governments are going to be unfortunately very reluctant to move toward a fully automated economy with basic income for all.

I don't think it has to be basic.

With 3D printing fully developed, generating your own car or even house could be as simple as typing the specifications into a computer screen.

Bob Clark

 

[Urwumpe]

They don't have to be victims, but people and especially governments are going to be unfortunately very reluctant to move toward a fully automated economy with basic income for all.

Not just they. I would also see some issues there. I am doing a job, that no computer can do yet. If I would get basic income - what would this mean to the value of my work? Is it worth less? Or do I need a rise in total to actually "pay the costs of going to work".

You are right, if you say, that the basic assumption of our economy, that the productivity of a worker is allowing him to earn his living, is wrong from the beginning. It did never work since there had been people who did the work and people who owned the tools and factories. But that does not mean, that it is easy to convert to an economy, where the whole of the society is earning for itself and everybody gets money as part of this society. Its a revolution in many things. Including the concept of citizenship.

 

[ADSWNJ]

I think the concept of The Singularity will happen in this century, linked to the Internet of Things, Neural Networks, nanotechnology (and nanofabrication capabilities), and the unimaginable rise of computing capability as we move from decade to decade. It's up to technologists and political leaders to consider the impacts on society as this happens, and take steps to transition to this new reality.

I think that many jobs today will become extinct in the next century too. For those who are not academically gifted, the life-choices will increasingly become more difficult, and it could well lead to a widening chasm in society.

I would hope that there is a future Star Trek - like end-state, where money is not the single goal for a life's work, and we can devote our lives to other tasks. What, though, is for our descendants to find out.

 

[fred18]

Ok, so I studied neural networks in these days, and I think I understood the concept and the scheme of them. I was also capable of building a small and simple one which "is learning" to make multiplications of two numbers.

That said I still don't know if I actually "agree" on the current neural network principle to make A.I.. It tries to replicate exactly the scheme present in human brain, and we are training the network giving it a set of input and correct outputs with which it can calibrate itself (that is called "learning") and then use what he learned and apply it with new inputs. But this just leads to pattern recognition or replication,

For example for summing two numbers the A.I. just calibrate itself in order to obtain the right output by giving two numbers as inputs, but it does not actually sum the two. And there is no idea of what pattern it uses to actually gets to the right result. It's like saying that two+three equals to five because the sounds of "u" and "e" equals to "i", the A.I. does not have any idea of the representation of it. Is it intelligence? or is just like a copy machine?

Now is it how our brain work? or not? and, if we can build a neural network as big and as complicated of our brain (100 billions of neurons) will it be somehow aware of itself, will it learn to talk and to interact?

if not, aren't we somehow losing time on this? we already invented a machine full of 1-0 status, with a long term and a short term memory, which we can easily train to do something and it will do it for us, and for which we can implement improvement algorithms. Aren't we just re-inventing the computer instead of pushing it with sorts of "awareness routines"?

To make the example thinking of orbiter: if we simulate the combustion of an engine in all its details, for each of the fuel molecole we will have a complete simulation of how the engine develops it's power, but will it be better than applying a thrust force as we are doing now? since we know how to apply the force, isn't it just to make it beautiful and more detailed, but without any significant improvement to what we can do with orbiter?

 

[Quick_Nick]

Ok, so I studied neural networks in these days, and I think I understood the concept and the scheme of them. I was also capable of building a small and simple one which "is learning" to make multiplications of two numbers.

That said I still don't know if I actually "agree" on the current neural network principle to make A.I.. It tries to replicate exactly the scheme present in human brain, and we are training the network giving it a set of input and correct outputs with which it can calibrate itself (that is called "learning") and then use what he learned and apply it with new inputs. But this just leads to pattern recognition or replication,

Be aware that you're only playing with the simplest version of things, conceived of when computers were barely starting to exist (50s-60s). There are many variations on neutral networks and other methods of learning.
Your questions are rather philosophical so I'll have to wait until I have more time to answer them coherently with my own opinions.

 

[boogabooga]

That said I still don't know if I actually "agree" on the current neural network principle to make A.I.. It tries to replicate exactly the scheme present in human brain, and we are training the network giving it a set of input and correct outputs with which it can calibrate itself (that is called "learning") and then use what he learned and apply it with new inputs. But this just leads to pattern recognition or replication...the A.I. does not have any idea of the representation of it. Is it intelligence? or is just like a copy machine?

I'm sorry, what did you think intelligence was, exactly?

If we can build a neural network as big and as complicated of our brain (100 billions of neurons) will it be somehow aware of itself, will it learn to talk and to interact?

Again with the self-awareness thing, you are anthropomorphising. "Self-awareness" is a concept applicable to humans, and perhaps some other animals. It is irrelevant to AI. The AI ill be an entirely different kind of "mind." Still intelligent.

Let me give you another example. Spiders arguably demonstrate intelligence by spinning a web to catch food. Are spiders "self-aware"?

To make the example thinking of orbiter: if we simulate the combustion of an engine in all its details, for each of the fuel molecole we will have a complete simulation of how the engine develops it's power, but will it be better than applying a thrust force as we are doing now? since we know how to apply the force, isn't it just to make it beautiful and more detailed, but without any significant improvement to what we can do with orbiter?

Yes, if you end up with a general model of combustion. You could then use Orbiter to calculate how many BTU furnace you would need to heat your house, for example. This is intelligence, using the experience and "skills" in order to solve new/different problems.

This is exactly like IBM Watson starting out by playing Jeopardy and now being used in medical diagnostics.

 

[TMac3000]

I'm sorry, what did you think intelligence was, exactly?

I agree--this is the central question of this whole discussion, and frankly a very difficult one. It has distinctly philosophical overtones.

I would start with saying what intelligence is not.

It is not a store of knowledge, or a filing cabinet would be intelligent.

It is not the ability to disseminate knowledge, or an answering machine would be intelligent.

It is not both together, or your computer would be intelligent.

Now then...I would say that what intelligence is is the ability of an entity (organic or mechanical) to both store knowledge and do something with it that no outside entity instructed it to do.

The idea of abstraction of methods (you know what it's doing, but you can't know how and/or why) falls short, because if we don't know how or why a machine is doing something, then we don't know whether or not it was programmed to do it, and thus we don't know whether the machine is taking initiative.

It really is a very blurry line between intelligence and self-awareness. As has been said already, one does not require the other.

 

[fred18]

Thanks for the answers guys! :thumbup:

Let me give you another example. Spiders arguably demonstrate intelligence by spinning a web to catch food. Are spiders "self-aware"?

I think they are somehow aware of themselves: if they don't eat they die, so they catch food in order to survive, if they want to live it's a sort of self awareness.

Now then...I would say that what intelligence is is the ability of an entity (organic or mechanical) to both store knowledge and do something with it that no outside entity instructed it to do.

Exactly! You got quite my point, I wasn't clear enough, let me explain my doubt a bit further:

let's assume that I want to let a machine do the sum of two numbers correctly. I could build a neural network and give him some training with couple of numbers and their sums and in the end it will give me the proper results when I query it for a sum of two numbers.

But I also can simply code how to sum two numbers into a simple program and it will work as well, and when I compile it, the program will be transformed into binaries, which rensembles me of the neurons etc.
Can we say that the coding of the program is the teaching?

So here's my doubt: are we "loosing time" trying to rebuild a system which will allow my machine to sum two numbers just because it will do it more like the human brain does, or this path through neural network simulations will lead to an intelligence that will be capable of doing things that it was not taught about before?

I know that surely nobody has the answer, it's just my doubt at the moment :hmm:

 

[Col_Klonk]

This could get hectic...
One could start with an egg full of DNA strands (just chemicals ?)..... absorbing energy and using this to replicate into what we call intelligence.

 

[Quick_Nick]

This isn't a response to any comment, this is just one of my favorite AIs:
(also an astounding set of examples of AI exhibiting unexpected behavior to achieve some goal)

[ame="http://www.youtube.com/watch?v=Q-WgQcnessA"]NES AI Learnfun & Playfun, ep. 3: Gradius, pinball, ice hockey, mario updates, etc. - YouTube[/ame]

 

[TMac3000]

But I also can simply code how to sum two numbers into a simple program and it will work as well, and when I compile it, the program will be transformed into binaries, which rensembles me of the neurons etc.
Can we say that the coding of the program is the teaching?

Well...sort of. Hard to deny that actually. That's essentially what you are doing when you write a chess program. But if you got beaten by your own software, it would be very hard to tell if the program was actually thinking, or just used your style of play in a way you hadn't thought of. A stove is not intelligent because it burns your hand.

It's one thing to ask "What is learning?" but quite another to ask "What is teaching?"

Again, you can't say that teaching is simply storing knowledge--or I would be "teaching" the office filing cabinet when I put a tax return in there. You have to expect the object of your tutoring to be able to do something with the information you give it. In that way, the only difference between teaching and programming is...well, nothing really. I personally believe there is a difference, but I couldn't put it into words without falling back into the anthropomorphization trap.

So here's my doubt: are we "loosing time" trying to rebuild a system which will allow my machine to sum two numbers just because it will do it more like the human brain does

You could make that case, but...

or this path through neural network simulations will lead to an intelligence that will be capable of doing things that it was not taught about before?

...this is the possible tradeoff. If it goes well, it would be very much worth it for the time and effort it would save us in the future. If it fails, it would be at best a waste of time, and at worst become a dangerous, uncontrollable weapon.

We are no longer particularly in the business of writing software to perform specific tasks. We now teach the software how to learn, and in the primary bonding process it molds itself around the task to be performed. The feedback loop never really ends, so a tenth year polysentience can be a priceless jewel or a psychotic wreck, but it is the primary bonding process--the childhood, if you will--that has the most far-reaching repercussions.

Yes, I know it is fictional, but it's just so pertinent I couldn't resist

 

[ADSWNJ]

Now then...I would say that what intelligence is is the ability of an entity (organic or mechanical) to both store knowledge and do something with it that no outside entity instructed it to do.

So - returning to my example of IBM's Watson, I would argue that it is already doing exactly this and more. Watson is working with natural language, evaluating massive sources of knowledge, generating hypotheses, and then testing the strength of those hypotheses to recommend approaches to healthcare that already puts it on a par with the best GP's in terms of diagnosis, and is rapidly becoming a world expert in specific fields. Source and source.

When algorithms can approach, meet, and then exceed the capabilities of the world's foremost specialists in a field, I think we can say that this is a prima facie demonstration of artificial intelligence.

It's going to be an interesting future!

 

[TMac3000]

So - returning to my example of IBM's Watson, I would argue that it is already doing exactly this and more. Watson is working with natural language, evaluating massive sources of knowledge, generating hypotheses, and then testing the strength of those hypotheses to recommend approaches to healthcare that already puts it on a par with the best GP's in terms of diagnosis, and is rapidly becoming a world expert in specific fields. Source and source.

When algorithms can approach, meet, and then exceed the capabilities of the world's foremost specialists in a field, I think we can say that this is a prima facie demonstration of artificial intelligence.

It's going to be an interesting future!

I understand your point here. But...we still had to ask Watson to do those things. And it did them using software we programmed it with. Watson did not just say "Hey, I think I'll become a doctor!"

I think my stove analogy applies here. The programmers didn't realize the full ramifications of the programming they gave the machine, and were surprised by what they got. But it still didn't take initiative.