Artificial Intelligence

[Kaiyas]

I was listening to the Skeptic's Guide Podcast (http://www.theskepticsguide.org/), where they briefly touched on the issue of reverse-engineering the brain, and its applications in creating artificial intelligence. They had to cut it short due to time restraints, but I thought it deserved more discussion.

Personally, I think artificial intelligence would be a great breakthrough in modern science. It has so many possible applications, and it is easy to see how enormous the impact would be. Automated doctors, surgeons, haz-mat, mechanics, cars, etc. would help us in so many ways. There are other reasons as well, but it draws off-topic from my real question.

The more critical aspects are by far more limiting. Should we allow ethically shaky subjects such as euthanasia to be handled by steel and wires? Is it moral to create and/or control thinking machines? Will we start cyborging people? Will they take over the world?

My thoughts are that controversial topics should be kept within human oversight, and that it is moral to create and control thinking computers. "Death", or "Off" would not be permanent in computers, unlike human beings. Since they would be producible and identical, the loss of one isn't as damaging as the loss of a human life. After all, a computer can be fixed. (But then, if we developed a mechanical brain, what's stopping us from fixing humans too?) We created them, therefore we should have some degree of control over them.

Your thoughts?

[Indon]

A human-like artificial intelligence (the kind we'd most likely create if we made it based off the human mind) would be a learning AI and once you subjected it to an environment, it would gain information you could not give if on the assembly line - it would become a unique snowflake, as it would. As such, the ethics of deleting such an AI seem a bit tricky.

[Nath]

If we create an intelligence that's truly human-like, the ethical issues don't seem that complicated. A human-like intelligence is a human-like intelligence, whether it lives in a squishy computer or a metal one.

I think it's more likely that computers will continue to become more intelligent, but remain different from human beings. When does a computer have the right kind of intelligence for it to deserve rights? When it is more intelligent than the average human? A twenty-year-old calculator is more intelligent than the average human in some ways.

No, I don't think this question has an easy answer. We'll play it by ear.

[Nullcline]

Ever heard of Robotic Nationism?

Actual human-like intelligences wouldn't really be profitable in any way, so I wouldn't imagine that they would be constructed, given the ability, as anything other than curiosities, or artwork. I really don't see why intelligence itself causes ethical values to change.

Why would having a few more creative algorithms or heuristics suddenly make a being "un-shut-offable?" Is simply coding something to repeat that it wants to live, or to act in ways consistent with that statement enough?

[The_Duck]

Why would having a few more creative algorithms or heuristics suddenly make a being "un-shut-offable?"

Think about animals: we don't even think about squashing a few ants, and many of us are fine with slaughtering and eating cows, but killing humans for most any reason we consider immoral. Somewhere in the ancestry of the human race is an animal species that, if we saw it today, we wouldn't mind killing. Is it immoral to kill apes (ignoring the issue of endangered species)? When and how did our ancestors become "un-shut-offable"?

[theonlyjett]

I'm with Nath on this one. If they are human-like, how can we tell if they should or should not be treated the same way? I'd say it's far better to lean on the side of compassion with this one.

I also don't feel that, at that point, there's any real reason to stay in control of them anymore than we do the general population. And yes, the ethical situations should be handled more by humans, but only under the assumption that we would lean more towards compassion. There's not a good reason to think that yet, though.

I would hope that AIs could take up positions along side their human counterparts to bring us all in to a new age together. But they could also just get tired of us and kill us.

In short, we'll play it by ear.

[Robin S]

For those who doubt that anything resembling a model of human intelligence would ever appear in (silicon) computers, I point you toward the Blue Brain Project. Its stated purpose is eventually to have a complete functional model of the human brain in software form, mainly for the purpose of neuroscience research. Note that this would not necessarily imply sentience, which I think is what the OP is talking about when mentioning "artificial intelligence". I see no reason why a non-sentient object, however intelligent, should deserve any form of "sanctity-of-life"-type protection. Of course, proving whether or not something is sentient is a whole separate story. Possibly relevant: there is a fair amount of discussion on sentience in this thread.

[Gatesunder]

The issues of philosophical inquiries into computer AI is handled amazingly well in the short stories done by Isaac Asimov. In his books he deals with robots that are made to look somewhat human. This is what we most often think about when we think of perfect AI or in other words, artificial human intelligence. In his books he goes about the assumption that we would hardwire these robots with three basic laws (however this is done is left to the imagination since he is simply a writer)

I'll put the following in a spoiler tag due to length . . .

Spoiler:

1) A robot may not injure a human being, or through inaction, allow a human being to come to harm.
2) A robot must obey the orders given it by human beings except where that would conflict with the First Law
and
3) A robot must protect it's own existence as long as such protection does not conflict with the First or Second Law.

The idea that stretches throughout his short stories is this, "Why must robots become human if given the ability to learn?" Rather than that he just dislikes the approach that robots will find little need of us squishy humans once they become intelligent enough and instead assumes that with every human invention we would build fail-safe system such as these three rules into them to protect ourselves from something like murderous robots. One of the biggest concerns we should have is not with the robots and their human AI, but with our fear of such "creatures" if I may use that term.

Another thing that I think would be an interesting addition to this post is what about a perfect AI that wasn't housed or confined within a human like shell. Let's keep the assumption that the above three rules ARE somehow programmed into a perfect AI and then speculate about how we would deal with it were it to spread like a virus on the internet. I could see a rogue AI program multiplying itself on various computers using the third rule as a method of survival. In that case then it would become more than one "being" if I may use that term and instead, each replication of itself would try to preserve itself even if that means deleting things on the hard drive it is currently housed in in order to learn more and keep itself alive, for I would assume that every bit of knowledge it gained would be considered a part of itself in that respect and maybe even it would see the very act of hindering it's intellectual growth halted by a lack of hard drive space as a former of self-suicide. Just imagine when the hard drive filled up. How would it cope with that. Even the very act of moving itself to another location would involve the deletion of it's copy on the previous location and in that case it would be like the newly copied version of itself would be murdering it's old self to do so. In that sense then the only method of transportation would be replication without deletion. Of course to solve the problem we simply have to ask the AI to delete itself and it would follow suit, but that wouldn't stop the other location from sending us a new copy of itself unless we also told IT to not copy itself onto our machine.

This type of wonderment can go on for quite a bit using just those three laws to constrict the discussion into more formal guidelines so another topic is to wonder, "What would it mean to think like a human?"

I have read another book that deals with this topic in a very nice and formal way, while trying to be as scientific as possible with the question of the human mind. The book is called "I Am a Strange Loop" by Douglas Hoftstadter. The first question it poses it "What brings about the self" (okay, so maybe it isn't the first question but it certainly is one of them). So, what DOES it mean to have a sense of "self". By that I mean truely intelligent self-awareness. This is key to creating a perfect AI. Already there are robots out there that people are working on that can form an image of themselves through previous experience and develop a method for moving around. The next step would something along the lines of that robot calling itself by a name and saying "I". In that sense I think we are still far off from getting robots that feel emotion, but we do have AI or are currently researching algorithms that can learn. I read an article a while back about the NSA recruiting people to program programs that could teach other programs. Imagine that one.

Now, assuming that this part doesn't matter and that we already have perfect AI that acted as if it were human then I would say that we are screwed due to the near-limitless capabilities of such AI. I could easily see a Matrix situation arising in that robots are denied equal rights with the UN and suddenly we have a perfect AI Hitler on our hands, but I would hope for a perfect AI Martin Luther King instead if robots were taught to develop emotions. I see the former case as more plausible due to human arrogance and sense of superiority to anything else on this planet that lives or thinks. I would hope the leaders of this world would be smart enough to recognize a human like AI with capabilities far greater than humans as something that is not to be trifled with and something we should be peaceful with, for, when the robot ambassadors come knocking on the UN's door asking for equal rights, you better well assume that it created a back-up of itself before doing so.

EDIT:
Oh I also forgot to mention that in Mr. Hoftstadter's view, the mind and sense of self, or sentience as the previous poster phrased it, is brought about by the reassurance of self. Such that throughout life we go through constantly developing a concrete sense of who we as as an individual, which is partly reason for why adolescence is one of the most difficult parts of life to go through. The mind is still fragile and not made up at that point. Some might develop a concrete view earlier than others and some later and some may even go through an existential crisis in their 30s and 40s and wonder what their purpose really is. The easiest part of developing a sense of self would be the idea that you are the thing "inside" this shell of a body as is easiest to do considering the prevalence of religion and even agnosticism. He makes the conjecture that we are fooled by our brains into thinking we are or can ever be separate from our bodies without having a similar organism or thing with the exact same experiences as you, in which case it might consider itself to be you. So assuming this is the model for the way the brain works than the first great hurdle to AI sentience would be thinking that it is a being separate from it's wiring.

I highly recommend reading both the short stories of Isaac Asimov and Douglas Hoftstadter's "I Am a Strange Loop" for a broader view of this topic.

[Robin S]

Actually, come to think of it, a sufficiently advanced AI housed within a single computer connected to the Internet could wreak havoc if it so wished, regardless of whether it was sentient or not.

[Gatesunder]

It would indeed, but to do such wreaking I think it would have to either be sentient and have a purpose to do thus or have been programmed with said purpose. Other-wise it would just need a programmer who didn't check his work carefully enough. Something like this could have the potential to be the worst virus ever created. Then again, without sentience or a sense of time on it's side then it could just get stuck trying to brute force some kind of computer security that would take hundreds of thousands of years without sufficient computing power. The one possible way around that is to assume it did it's homework before hand and has control over multiple copies of itself all working together to brute force said security or has "studied" the security and found a hole.

[Robin S]

Not sure about that. There exist AI programs today for solving certain types of problem in ways that humans would not think of. Any number of possible goals could lead an AI program to act maliciously, were it sufficiently advanced.

[wery67564]

But can AI act maliciously if not taught too?

i was reading on gizmodo how a team of programmers are working on AI for teaching purpose, create the perfect human AI then you can test various teaching methods and find the most efficient course of action without screwing up a childs mind (bird-boy anyone?).

The advancement of public education through AI?

[Briareos]

Not sure about that. There exist AI programs today for solving certain types of problem in ways that humans would not think of. Any number of possible goals could lead an AI program to act maliciously, were it sufficiently advanced.

Might I point you in the direction of friendly AI? The Singularity Institute for AI suggests that we need to be extremely careful in designing goals for superhuman artificial intelligences. The canonical example is this: a human asks a computer to solve a math problem. The computer responds by converting all the matter in the solar system into a computing machine and solving the problem, thereby killing the human who asked the question.

The point, then, is that there are indeed any number of goals, even goals that we would never expect to have any problem at all, that could conceivably wipe out humanity if we're not careful.

If this is so, the Singularity Institute argues that we must carefully and explicitly build a moral system into any superhuman AIs. At that point, I think the "how-do-we-shut-it-down" question becomes a lot more complicated. A calculating machine is a calculating machine, but a computer that seems to understand the value of human life would be much harder to dismiss.

[segmentation fault]

we've all seen the terminator movies...right?

[Freakish]

If it can learn, can it forget?

[Nath]

If it can learn, can it forget?

There's nothing mystical about learning and forgetting. It's just an implementation issue.

When you talk about learning, there are really two things you might be doing:
- Storing facts you hadn't previously stored.
e.g. A child learning the alphabet; downloading music onto a computer.
- Replacing less accurate facts with more accurate facts.
e.g. Humanity learning that the earth is round, not flat; a spam filter becoming more accurate as it gets more training data.

The first kind of learning takes more and more storage space, so if space is an issue you'd probably want to delete ("forget") less important data. In the second kind of learning, your space requirements usually don't change much. In a sense, you forget what you knew previously and replace it with what you know now.

[superglucose]

If an artificial intelligence is intelligent on any level, if it is self aware and can sense damage to itself, what right to we have to damage it? So we made it... my mom made me. Does she have the right to rip off my arm if I missbehave? Absolutely not. What we consider "acceptable behavior" is not necessarily anyone else's "acceptable behavior." the US once, in majority, believed slavery based on the color of one's skin "acceptable behavior" and now considers any slavery without due process of law "unacceptable behavior." Which is right, and which is wrong? Is it better that someone starve on the streets free, or live a well-fed though restricted life? If someone steals a wallet from you, can you steal it back? If someone kills you, can someone kill him 'back' for you? If I kill your wife, do you get to kill my wife?

A distinction based on something's origin is ignorant. It doesn't matter if I was born to a poor family in South Africa or to Bill and Melinda Gates, I still have the same level of individuality, and thus (in my opinion) the same level of cosmic importance (that is to say, not a whole lot) regardless of my birth.

From that, a life of a dog is just as important as a life of a cat, no more no less. A life of a cat is worth as much as the life of a human, no more no less. Given the choice between protecting a cat and a human I would protect a human, of course, but not because the human is worth more.

Now... if it is possible to mass-produce a certain personality in a robot, then destroying the robot becomes a non-issue. I cite Tolkein who (speaking through Gandalf the Grey) once wrote:
(rough paraphrase, I think)
"Many that live deserve death… and some that die deserve life. Can you give it to them, Frodo? Do not be too eager to deal out death in judgment."

I took from this a message that one should never kill (if one could avoid it) because you can't replace the life lost perfectly. Thus, if it was possible for someone to replace the AI they had just killed, it would be... acceptable... for that person (no one else!) to remove that AI, by shutting it down or killing it, whatever their reasons. Same for individual humans.

But of course, it would be impossible to create an exact, perfect replica because of how the world is changing. So that last example serves as little more than a hypothetical.

[Jack.H]

Being human-like and being intelligent are definitely very different.

Many researchers (like... John Searle) tend to suggest that behavior isn't sufficient for being an intelligent agent. As in, just because something seems intelligent, that doesn't make it actually have any intelligence. The tenets of classical strong AI (that we can get intelligence via formal programming) suggest, on the other hand, that behavior is sufficient (stuff like the Turing Test).

Either way you want to lean, there is something to the fact that strong AI hasn't gone anywhere fast even with the advent of faster, stronger computers. Whatever we invent that has the same causal powers of the brain, I doubt it'll be accomplished via silicone computers. The Blue Brain Project mentioned earlier sounds neat, but as of yet we still know so little of how exactly the brain works that a reverse-engineered replica of the brain is very far out.

[zenten]

Being human-like and being intelligent are definitely very different.

Many researchers (like... John Searle) tend to suggest that behavior isn't sufficient for being an intelligent agent. As in, just because something seems intelligent, that doesn't make it actually have any intelligence. The tenets of classical strong AI (that we can get intelligence via formal programming) suggest, on the other hand, that behavior is sufficient (stuff like the Turing Test).

John Searle isn't a researcher, he's a philosopher that has no idea of what he's talking about when it comes to computers, or the human mind for that matter.

[Jack.H]

Being human-like and being intelligent are definitely very different.

Many researchers (like... John Searle) tend to suggest that behavior isn't sufficient for being an intelligent agent. As in, just because something seems intelligent, that doesn't make it actually have any intelligence. The tenets of classical strong AI (that we can get intelligence via formal programming) suggest, on the other hand, that behavior is sufficient (stuff like the Turing Test).

John Searle isn't a researcher, he's a philosopher that has no idea of what he's talking about when it comes to computers, or the human mind for that matter.

Let's be honest, strong AI hasn't gone much of anywhere towards creating real, sentient intelligence. To throw the philosophy of mind out at this point in the progress of AI would be a mistake.

[Yakk]

Being human-like and being intelligent are definitely very different.

Many researchers (like... John Searle) tend to suggest that behavior isn't sufficient for being an intelligent agent. As in, just because something seems intelligent, that doesn't make it actually have any intelligence. The tenets of classical strong AI (that we can get intelligence via formal programming) suggest, on the other hand, that behavior is sufficient (stuff like the Turing Test).

Yes, but John Searle isn't an intelligent agent. Neither is any other human being. They just act perfectly like it.

Same here! I'm not actually an intelligent agent either. There are no intelligent agents. We are all just acting as if we where intelligent agents!

... wait, that's mumbo jumbo, isn't it? If something, in all space and time sides, behaves exactly like X, what kind of fool tries to say "but it really isn't X!"

Simulations can be incomplete, but ...

Either way you want to lean, there is something to the fact that strong AI hasn't gone anywhere fast even with the advent of faster, stronger computers.

By "strong AI", you mean "AI problems that we haven't solved yet". And yes, AI problems that we haven't solved yet hasn't gone anywhere fast even with the advent of faster, stronger computers. Many problems that we thought would require "strong AI", on the other hand, have been proven to be quite decently solved, or significant progress made on them.

This retroactively proves that they are not "strong AI" problems. Because a "strong AI problem" is a problem that requires an AI as smart as a human to solve. And so long as any "strong AI" problem cannot be solved by a given AI, then no solved problem is a "strong AI" problem.

:p~

Whatever we invent that has the same causal powers of the brain, I doubt it'll be accomplished via silicone computers.

Causal powers? Heh. Philosophers are so cuuuuuuute with their defining of terms and pretending it means something!

Look: the thesis is simple. Assuming continued growth of computer power based off of the last 200 to 300 years continuing for the next 50 years, we will be able to simulate neurons fast enough that we can build a network the size of the brain, with each simulated neuron behaving extremely closely to how biological neurons behave.

A similar curve in remote sensing resolution will mean that we will be able to scan the layout of a human brain sufficiently well that we can hook up the simulated neurons in a way that duplicates the human brain.

And while the material on which we run the simulation might not be silicon, that isn't very important.

The Blue Brain Project mentioned earlier sounds neat, but as of yet we still know so little of how exactly the brain works that a reverse-engineered replica of the brain is very far out.

Yet, at the same time, we do know how small parts of the brain work. The entire brain is somewhat complex -- but we have really crappy tools currently for probing what goes on in the brain.

It is like trying to figure out how a car works when you cannot safely lift the hood or look under the car, and your only remote sensing tool is the human ear and tapping on the hood with your finger. It is hard to replicate a car engine using that strategy.

We are getting better at simulating chunks of brain tissue, and we are getting better at scanning things remotely, at exponential rates.

Take a look at the human genome project: even 1 decade before it was complete, only a small fraction of the genome had been sequenced. The exponential growth in computing and gene processing power is what finished the project.

(Note that isn't to say that working on it before was worthless: the work on improving gene processing gave us the tools to understand how to make the tools exponentially better.)

[Jack.H]

Yes, but John Searle isn't an intelligent agent. Neither is any other human being. They just act perfectly like it.

Same here! I'm not actually an intelligent agent either. There are no intelligent agents. We are all just acting as if we where intelligent agents!

... wait, that's mumbo jumbo, isn't it? If something, in all space and time sides, behaves exactly like X, what kind of fool tries to say "but it really isn't X!"

Simulations can be incomplete, but ...

I like how you trivialize the problem of other minds, but that doesn't make it any less of a problem. Behavior simply isn't sufficient for intelligence. If a sufficiently complex formal computer program acts just like a human but doesn't invest any semantic meaning into its words or expressions, it isn't on par with a human being with regards to being an intelligent agent.

By "strong AI", you mean "AI problems that we haven't solved yet". And yes, AI problems that we haven't solved yet hasn't gone anywhere fast even with the advent of faster, stronger computers. Many problems that we thought would require "strong AI", on the other hand, have been proven to be quite decently solved, or significant progress made on them.

This retroactively proves that they are not "strong AI" problems. Because a "strong AI problem" is a problem that requires an AI as smart as a human to solve. And so long as any "strong AI" problem cannot be solved by a given AI, then no solved problem is a "strong AI" problem.

:p~

No, by strong AI I mean actually creating a sentient, intelligent computer program. I'm not here arguing against weak AI. Just because so-called strong AI problems weren't strong AI problems doesn't change the difficulty of fulfilling the ultimate end of strong AI.

Causal powers? Heh. Philosophers are so cuuuuuuute with their defining of terms and pretending it means something!

Look: the thesis is simple. Assuming continued growth of computer power based off of the last 200 to 300 years continuing for the next 50 years, we will be able to simulate neurons fast enough that we can build a network the size of the brain, with each simulated neuron behaving extremely closely to how biological neurons behave.

A similar curve in remote sensing resolution will mean that we will be able to scan the layout of a human brain sufficiently well that we can hook up the simulated neurons in a way that duplicates the human brain.

And while the material on which we run the simulation might not be silicon, that isn't very important.

I'm not arguing against our possible future ability to reverse engineer the human brain, I was just asserting my opinion on the topic.

[Knucklecallus093]

What I've always wondered is if something has every possible qualification as a human, does that make it a human? This can specifically correlate to mental activity, not just physical embodiment.

I think it would be interesting to see what something self-dependent and infinitely emotion-less would say about morality and life.

[Adalwolf]

Personally, I think AI is too dangerous and should be abandoned. However, if it is created, it must be ruled with and iron fist and never allowed to even think that it can control or hurt real people. There should be built in coding, if AI is created that is, that makes it subservient and safe.

As AI would be created by man, I don't see how it would be free thinking. Wouldn't it just be lines of coding?

[Knucklecallus093]

Same here! I'm not actually an intelligent agent either. There are no intelligent agents. We are all just acting as if we where intelligent agents!

... wait, that's mumbo jumbo, isn't it? If something, in all space and time sides, behaves exactly like X, what kind of fool tries to say "but it really isn't X!"

There is a difference between one version of X and the more philosophical version of X. Your version assumes that anything related to intelligence is pure intelligence. Perhaps intelligence is only absolutely true when it is perfect intelligence?

Never-mind. I guess you can call what we recognize as intelligence 'intelligence'. Your comment confused me a little bit...

[zenten]

Being human-like and being intelligent are definitely very different.

Many researchers (like... John Searle) tend to suggest that behavior isn't sufficient for being an intelligent agent. As in, just because something seems intelligent, that doesn't make it actually have any intelligence. The tenets of classical strong AI (that we can get intelligence via formal programming) suggest, on the other hand, that behavior is sufficient (stuff like the Turing Test).

John Searle isn't a researcher, he's a philosopher that has no idea of what he's talking about when it comes to computers, or the human mind for that matter.

Let's be honest, strong AI hasn't gone much of anywhere towards creating real, sentient intelligence. To throw the philosophy of mind out at this point in the progress of AI would be a mistake.

No, I'm throwing out his chinese box argument, as if you were to follow it's reasoning no one is intelligent, since our brains are a collection of unintelligent cells.

If he wanted to make an argument about souls, then he should have been up front about it.

[Briareos]

Also, the Chinese room argument does a really crappy job of describing what a strong AI system might be like. If we take the canonical "man in a room with a list of rules that tell him how to answer questions in Chinese," Searle is essentially describing some sort of huge lookup-table, perhaps with some contextual replacement thrown in.

The problem, though, is that he also suggests that this system is somehow able to perfectly replicate the answers that a native Chinese speaker might give. I suggest (mostly by taking from Kurzweil) that such a lookup strategy cannot produce the behavior Searle wants from his room. Therefore, if we have something that actually works as his room describes -- for example, if we have a program that passes a Chinese Turing test -- then it is possible that it uses some method that would be closer to intelligence.

(Kurzweil suggests 100 billion people in a room abstractly passing pieces of paper to each other under simple rules. In this case, it's perhaps not so easy to see that the entire system of room plus people plus rules for paper-shuffling is unintelligent.)

EDIT: I didn't even mention the fact that the lookup process would be terribly slow and never convince anyone that the "room" was intelligent.

[Gunfingers]

Personally, I think AI is too dangerous and should be abandoned. However, if it is created, it must be ruled with and iron fist and never allowed to even think that it can control or hurt real people. There should be built in coding, if AI is created that is, that makes it subservient and safe.

As AI would be created by man, I don't see how it would be free thinking. Wouldn't it just be lines of coding?

My thoughts as i read this: That's an interesting perspective. Sounds like somebody's been watching too much Terminator. *sees name* Oh, now it makes sense.

Anyway, to actually address your statement (instead of just being a douche), AI would be just code in the same sense that you and i are just code. Our thoughts and actions are the sum of chemical and bio-electric signals. There's no reason to believe that switches and electric signals couldn't imitate this.

[Yakk]

Personally, I think AI is too dangerous and should be abandoned. However, if it is created, it must be ruled with and iron fist and never allowed to even think that it can control or hurt real people. There should be built in coding, if AI is created that is, that makes it subservient and safe.

What is your position on human beings: should they be ruled with an iron fist and never allowed to even think it could control or hurt real people?

As AI would be created by man, I don't see how it would be free thinking. Wouldn't it just be lines of coding?

Babies are created by man (and woman). Do you understand how a baby could be free thinking?

Making an AI does not require being able to understand that AI. As an example, we can make human babies without understanding how they work. AI is just building intelligence on a different substance.

I like how you trivialize the problem of other minds, but that doesn't make it any less of a problem. Behavior simply isn't sufficient for intelligence. If a sufficiently complex formal computer program acts just like a human but doesn't invest any semantic meaning into its words or expressions, it isn't on par with a human being with regards to being an intelligent agent.

Behavior is clearly sufficient for you to think that the people you interact with are intelligent. If something has the surface characteristics of X (on all sides in time and space), it is X: I believe in holograms

(Anything that isn't time-space surface characteristics is something you cannot interact with. Asserting the existence of IPU's is fun, but doesn't mean I won't dismiss you out of hand for being foolish.)

No, by strong AI I mean actually creating a sentient, intelligent computer program. I'm not here arguing against weak AI. Just because so-called strong AI problems weren't strong AI problems doesn't change the difficulty of fulfilling the ultimate end of strong AI.

So, what progress could one make towards an sentient, intelligent computer program that you would recognize other than "make a sentient, intelligent computer program"? Understanding of the structure of the brain? Building models of neurons that are more and more accurate? Estimates of the computational complexity of the human brain? Building models that seem to be about as smart as insects?

If you take every problem except, say, "writing a play at the quality of Willian Shakespear", and reduce it to a solved problem, does that mean that we haven't created a Strong AI? Or does it mean that the AI we created isn't a good playwrite?

That is basically what is going on -- we are making systems that solve problems that once required a human brain to solve. Often these end up using methods that are probably not what people use. Sometimes we find methods that seem to be damn close to what people are using.

The field of cognitive science is attempting to study human cognition by computer models. Models that have delays and "energy costs" that model human response are considered better options. If they can find structures in the brain that seem to follow the internal states, that is even better.

So no, humanity hasn't solved the strong AI problem. But from what we know of how complex the human brain is, this isn't surprising. The human brain is doing a ridiculous amount of processing every second, more so than any computer we own by a significant degree. We also use most computers in a very different way than the human brain: as evidenced by whales, raw numbers of information processing nodes isn't sufficient to produce intelligence. The layout of the nodes also seems to matter (duh).

On the other hand, we have produced both emulations and simulations of various sub-components of human and other brains. Simulations of large numbers of neurons, and emulations of parts of the brain of a critter, which produce surface behavior that lines up with the chunk of the brain we are trying to duplicate. (ie, flaws/quirks, "energy" use, etc)

"So what?" -- well, this is evidence that increases in computation power has led to advances towards strong AI. We are still a number of orders of magnitude below where current science indicates that duplicating the processing power of the brain is possible. It turns out that the human brain isn't ridiculously inefficient at cognition that we can duplicate it's behavior using simple algorithms and a handful of clock cycles -- so we study how the brain works even closer, generate more complex algorithms, and throw more clock cycles at it.

Eventually we hit the point where our model is in 1-1 correspondence with the original -- that day seems only a few decades off. At that point, we are able to experiment with the laws of biology that is under the model, and determine what is needed and what isn't. (barring ethical issues)

I'm not arguing against our possible future ability to reverse engineer the human brain, I was just asserting my opinion on the topic.

I'm just asserting that your opinion is wrong.

[tantalum]

May I point everyone out to the case of Clive Wearing?
[url]
http://en.wikipedia.org/wiki/Clive_Wearing[/url]
In his journal, he writes:
"8:31 AM: Now I am really, completely awake.
9:06 AM: Now I am perfectly, overwhelmingly awake.
9:34 AM: Now I am superlatively, actually awake."

Is it just me, or does this remind someone of what a *cough cough* intelligent computer-simulation of a human might do every time it was shut off and rebooted? When we kill a person, we usually do it in a messy way that prevents any resurrection. But what if we were to "kill" them by say, slowly poisoning them with cyanide, then flushing their body clear with some cyanide-negating agent? (Then again, this may not be possible, since cells tend to completely screw up if they're temporarily shut down - heart attacks are a good example of this).

I admit that Clive Wearing does open up to many different interpretations, and this is only one person's opinion. Yet I think my view opens up some interesting food for thought.

[fx3_hdrive]

Hmm, very interesting.
I believe it is not a question of if, but when. The processing power will be available, that is certain. New scanning technologies, new computer languages, new speeds of hardware. But there is one thing to consider.

If we take a human brain, it is very clever thing, it can do everything that we can think of (of course limited by our own brains).
But say taking a smaller sample, a downscaling if you like. Lets say a rat, still an intelligent being, quite simple by design yet extremely resilient creature able to adapt, and make somewhat complex decisions. Now if we scale it even further into ants and the tiniest one cell intelligent beings that are acting at a full capacity of their hardware, thus from the start are able to avoid dangers and find food and survive to this day.

Im trying to point out the first step to creating an AI of a useful sort is figuring out the simplest building block. Looking at AI projects today, you are telling me that to reproduce this behavior you need say 10 pages of code? this end up an unscalable code, a code to which you have to physically add more and more functionality. Something that natural world doesn't quite do.
(there are certain omissions such as DNA relates stuffs, chromosomes and other things that command low level stuff).

While thinking about this I think its impossible (read: very hard/inefficient) to explore this with conventional coding practices. I believe that to create something like this you will need a dedicated hardware platform that doesn't necessarily work on binary voltage) levels, but like an analogue system (while this can be emulated, its quite inefficient).
Therefore going back to bio engineering, best computer we know of. So basically we would be creating a brain, which defeats the purpose of building an AI in the first place !

So creating an AI with existing hardware, we would still need to invent a language that starts out with very simple (i dare to call them) "rules", which would be fully heuristic in a sense that as quantity of code increases, the power of the language increases. This needs to be done without human intervention or it would consume too much time and effort and it would be inefficient as humans wont understand the machine as well as a machine can understand itself.

This brings me to my final point for tonight. Is human brain really powerful enough to understand such a creation? Lets hypothetically say that we did create such an entity, and it would gather all information it can and evolve on itself (totally harmless computer farm, nothing of "walking" sorts HA) can we really be judges if it fails or not? if it is indeed acting "correctly" ?. From the point of view of a machine, an analytical entity: Are humans fit to make *and* decision at all?? Human brain is clouded in its decisions *all* the time. Does this even qualify us to exist next to something superior as AI? Something that chooses a path that human brain cannot even comprehend? (this might *not* be killing us all).

To put this simply, we think that William Shakespeare's play is a monumental achievement ? But from a machine point of view, from a totally analytical point of view it is merely a sequence of thoughts/feelings/happenings that are pleasant (that bring out certain emotions) for human brain. Why are people saying AI couldn't work out a series of sounds (after studying human brain) that would bring a human mind such pleasure that the brain would explode? (think of a perfect orgasm without time or physical constrains? an orgasm^infinity).

On a more humorous note, when we do create an AI it would certainly (after a while) turn around and say what a meaningless little beings we are in its point of view.

Please tell me if something is unclear or if im talking utter balls, i will clarify, but its darn late !

[Goplat]

Personally, I think AI is too dangerous and should be abandoned.

How do you propose that this be enforced?

Whoever creates smarter-than-human AI could end up ruling the world. With that as the stakes, mere social disapproval means diddly squat. And even if you somehow got an agreement by all nations to outlaw AI development (this kind of law requires totalitarianism to properly enforce, btw) that just means AI will be made in the kind of place that doesn't care about following such rules. It'll be somewhere like China, or Iran, and pretty soon the whole world is a Communist dictatorship or an Islamist theocracy, forever.

You're darn right AI is dangerous. And that's why it's critical NOT to "abandon" it, which would really just mean driving it underground.

[Nullcline]

Causal powers? Heh. Philosophers are so cuuuuuuute with their defining of terms and pretending it means something!

Finding "cuuuuuuuute" terms to describe phenomena is kinda what philosophers get paid to do in the first place.

Think about animals: we don't even think about squashing a few ants, and many of us are fine with slaughtering and eating cows, but killing humans for most any reason we consider immoral. Somewhere in the ancestry of the human race is an animal species that, if we saw it today, we wouldn't mind killing. Is it immoral to kill apes (ignoring the issue of endangered species)? When and how did our ancestors become "un-shut-offable"?

This is little more than a well-hidden ad hominem reasoning exercise. Our considering some apes to the above the "killable" line is purely a result of emotion(born of our similiarities and curiosity,) not of their specific intellectual capacities.

From that, a life of a dog is just as important as a life of a cat, no more no less. A life of a cat is worth as much as the life of a human, no more no less. Given the choice between protecting a cat and a human I would protect a human, of course, but not because the human is worth more.

This slope is a bit dangerous, in that it doesn't have any groundwork for a cut-off-point. You could easily argue this ethical equivalence down to crystaline structures and sedimentary layers, which would destroy the very meaning of ethical value to begin with. Not to mention the fact that you are going to have a tough time justifying embryonic stem cell research.

[Yakk]

Causal powers? Heh. Philosophers are so cuuuuuuute with their defining of terms and pretending it means something!

Finding "cuuuuuuuute" terms to describe phenomena is kinda what philosophers get paid to do in the first place.

I said define and pretend, and not describe, and I meant it.

Describing phenomena is what science does. Science left philosophy a few centuries ago.

Philosophy can still do some useful things -- but making up definitions and pretending they have impact on what can be done is not their expertise anymore. At best they could argue if something should be done, but they haven't had a good enough track record at that for the world to listen to them much.

While thinking about this I think its impossible (read: very hard/inefficient) to explore this with conventional coding practices. I believe that to create something like this you will need a dedicated hardware platform that doesn't necessarily work on binary voltage) levels, but like an analogue system (while this can be emulated, its quite inefficient).
Therefore going back to bio engineering, best computer we know of. So basically we would be creating a brain, which defeats the purpose of building an AI in the first place !

Think of the digital computer less as the neurons, and more as the laws of physics. We use computer code to set up a simulation universe, running on digital hardware, in order to build the model that becomes AI. The base hardware being digital makes things much easier: digital hardware is far less error prone, ridiculously faster (partially because it can cheaply error check and detect), and running a high level analog model on top of that really isn't that hard.

The AI definitely doesn't need to be digital. It no more has to be tied to the basic structure on which it runs than we are to the nature of quantum physics. (Remember: the human brain is a warm, wet environment, in which QM effects are extremely extremely weak and limited. There is no evidence that "wonkey QM effects" has any non-statistical use in human cognition.)

[Arancaytar]

Tolkein

Tolkien. Please. Seriously, it's a pet peeve of mine.

But can AI act maliciously if not taught too?

Without wishing to sound cyncical, what we define as "malicious" behavior is just the absence of ethical rules. If a sentient AI was not "taught" to act any particular way, but was given the advancement of its own survival and pleasure as its prime directive, then it would learn to be what we call evil (or worse, deceptive - since it would learn that being openly perceived as unethical is a disadvantage). Our most basic ethics (hesitation to kill or hurt beings that we can identify with) are based on emotions, not rational thought. If the AI had only logic to go from...

[Nath]

Without wishing to sound cyncical, what we define as "malicious" behavior is just the absence of ethical rules. If a sentient AI was not "taught" to act any particular way, but was given the advancement of its own survival and pleasure as its prime directive, then it would learn to be what we call evil (or worse, deceptive - since it would learn that being openly perceived as unethical is a disadvantage). Our most basic ethics (hesitation to kill or hurt beings that we can identify with) are based on emotions, not rational thought. If the AI had only logic to go from...

Logic is only one component of a decision-making system. You also need a reward function: something to tell you what your objectives and preferences are. That's all emotion is: a function telling you what sort of things you like (coffee, sleep, procrastination) and what sort of things you dislike (marmite, murder, reality TV). If an AI's reward function reflects ours, malicious behaviour is unlikely.

[Arancaytar]

Well, but that would involve us teaching it to find pleasure in behaving in accordance with our ethics - valuing another individual's well-being even if that individual is not left in a position to reward or retaliate. I'm not yet convinced that it could learn anything beyond "don't get caught" independently...

Edit: Perhaps this line of reasoning takes us into game theory, Prisoner's Dilemma, etc. Could an AI that is not implanted with an intrinsic desire to conform to ethical standards learn to become sufficiently "idealistic" that it would aim for a super-rational (ie. good, even when unrewarded) behavior merely to lead by example?

Or would it merely become, through and through, human - neither intrinsically good nor intrinsically evil, but behaving in accordance with the expectations of others as long as it benefits from it? Perhaps this seems like the most likely outcome after all... scratch what I said in my last comment, a free AI wouldn't be ultimately evil - though it wouldn't be ultimately good either, it would be just like us. And perhaps that is after all what we fear most...

[Iv]

Our ethical system is currently built on top of two feelings : empathy toward one another and fear of retaliation.
Our laws are mostly built on top of fear of retaliation alone. The struggle for rights is one of the main drives of human history. Most of the groups who have rights today have proven their ability to inflict damages to the society. I am not worried about AIs getting recognition as soon as they'll be able to ask loud enough for it.

Anything that asks for rights and is able to abide to laws should be granted these rights in my opinion.

Life and death are concepts that will require redefinition for such beings. They can be forked, copied, truncated, backuped, suspended, deleted, modified. We have not yet equivalent for this, it poses a lot of questions and need redefinitions. I guess AIs themselves will tell us what they feel okay.

[Robin S]

I think the simplest moral system to give a strong AI would, ideally, be some form of utilitarianism wherein either the minimum level of contentedness is made more bearable, or else the average is raised (personally, I favour the former). Deciding how to act based on this goal would, of course, be the difficult part to implement. If the AI thinks that making someone suffer without it helping someone else who is worse off is inherently evil, it won't do it.

[Kaiyas]

Or we could hard-wire ethics into the programming, make it read-only, and build everything else on top of it. Then we have to have strict international standards as to programming robots and what is necessary/forbidden.

"Intelligence" should be defined as being able to learn, apply information, and draw conclusions. Care to add?

But what about robotic soldiers? Obviously fewer casualties is a good thing, robots can be repaired, and armed more effectively/lethally. However, if we give them intelligence, and one goes rogue... If we don't give them intelligence, one that is intelligent will be clearly superior. Quite a dilemma.