sidenote: qwih is not smart

Is he trolling? Yeah, no, AI won't take over the world anywhere outside Hollywood :e

let me guess: You haven't read much about what you're speaking about.

Oh look, its Mr. Hawkings himself.

I'm pretty sure both I and Hawking did more research on this than maza, so I call your point moot

That's a really good argument

well with what the fuck would you reply to yourself

??????

oh, I thought you were being facetious.

moopoint was a fun movie

I'm confident I need no research to call this impossibru. Machines can learn and can kill people but design themselves to be human, no. Didn't read what Hawking was on about this time but he's a man you shouldn't listen too much, I think.

Well, the question would be why. Is there something special with our brain that makes us able to do things other hardware wouldn't be able to? If not, then is it a software problem? Is there something in our mind that cannot be reproduced? What is it and why can't it be?

It's perfectly fine if you have the feeling this can't happen, but feeling isn't knowledge. If you do have knowledge, please share it :).

Well one thing that machines sure lack is soul. But yeah didn't read the article or whatever so I don't know what Hawking exactly claimed.

By soul, you mean consciousness? That changes exactly nothing. They can be whatever we/they decide theyu want to be, and they can be everything. Won't become humans, but may easily be humans 2.0

Are you a christian

Oh boy, so many issues with what you say, don't' even know where to begin.

To be clear, I'm not saying you dump a human brain inside a computer to make an AI. I was trying to find words to get maza to question his beliefs. What makes AI impossible? Well he answered with "soul" so I decided to stop the discussion there.

Obviously, research isn't going this way. It would also be a stretch to say neural network in AI are like reproducing the human brain in a computer. But I haven't said any of that. If you have further criticism about my questions to maza, feel free to begin anywhere.

He only has snarky remarks to make.

sidenote: qwih is not smart

crazy ai is crazy?

not us, but maybe people alive 150 years from now

electromagnetic pulse bombs.... problem solved.

that's where they fucked up in the matrix.

Only one problem:

he sure knows how to cash in on his cerebral palsy or w/e he got. so much hype cuz he's a scientist in a wheelchair.

inb4 weird being a tosser

esdf took his spot as a tosser.

i can never outtoss weird.

Weird is in his lab, cooking up math;

*meth

[img1]

SyncError said every Pole playing QL is bot so who knows...

stephen hawking is a fucking vegetable. who cares what "he" says about anything? I don't get the fascination. seems like every week there's some new "stephen hawking said blah blah blah," and it's invariably something stupid. I care almost as much as I care about Matt Damon's opinions on politics.

Muricans..

don't associate tyryl with america.

for real, that's just rude :( we'd have him deported if that were an option!

because someone named "eburbeck"--whatever that is-- speaks for America. lol.

stephen hawking is a fucking quake master

I'm with you. After the "Be afraid of aliens" I was thinking nigga is trolling or going senile. Even if there was intelligent life outside, they would need about million years to get here even with some lighting fast spaceship with autopilot AND they would come birds way directly. One steer wrong, and theyll miss us.

Even if someone wanted to build ultrasophisticated robots out to destroy humans. Why not just build a nuclear weapon and blow us up or some germ without antidote? It seems needlessly complicated.

I cannot wait!

Pretty sure mankind's destiny was to invent technology that can take care of itself and maybe take us along for the ride.

Really don't see much of a future for mankind which seems to be based on capitalism and religion, both of which will only end in tears.

Just enjoy your life while you can, doesn't matter how smart you are or what you do, you'll always be insignificant.

The only way to win is to do things which make you happy and not to die in pain! and if you can help others realise that too, then bully for you.

"doesn't matter how smart you are or what you do, you'll always be insignificant"

coming from someone who loves to belittle other people every chance he gets by shoving BAFTA awards in their faces

I think I already won this one ages ago, you guys mention it far more than I do without zero provocation from me is far more apparent.

Your post is just further evidence of that.

http://esreality.com/files/placeimages/2011/8...-bafta.JPG

For the record, I don't really consider a BAFTA as worth much after 10 years have passed as one runs the risk of being a has been if they keep talking about it.

https://www.youtube.com/watch?v=tUg_Qp7gQQI

yea sure, I ll check back 100 years later.

Even if we are getting replaced by "bots", what's to say that's not a normal and natural part of evolution? And if we take history as an example, with the exception of humans replacing humans (i.e., genocide), there aren't many cases of one "species" ("bots" being viewed as just another species in this scenario) completely replacing (or displacing to the point of extinction) another. So in my view, we might see "bots" "take over" the planet, but not in a way that is necessarily detrimental to humans. In fact, given that they arise from humans, and humans are always self-serving, the "bots" would likely have a tendency to serve humans as well.

Further, I see little conflict of interest that could arise...resources required by "bots" vs. humans are likely to be different. This is because robots would probably not eat food (that seems somewhat inefficient), nor would they have similar housing/space requirements as people. The area where there could be conflict is if any sort of cross-breeding occurs - as soon as "bots" need human women to reproduce effectively, we're in trouble, as we've seen that to be the root of ALL human conflicts through all of history (different discussion - start another thread if you want to work that one out with me).

So I'd say, "fine, bring on the bots and we'll see how things go".

lol @ half the people here acting like it's some cooked up joke scenario from a movie. Unless we blow ourselves up with bombs, AI WILL reach human level intelligence and then surpass it, and at that point there is no way of knowing what it will do.

AI researchers will tell you this. I haven't listened to this one yet, will sometime soon, but you can try listening to this: http://twit.tv/show/triangulation/177 . I'm sure he'll say the same thing, AI can easily be very dangerous.

well atm the idea that they will reach human level intelligence after which they might go off on their own and we lose control of them because they are smarter... IS hollywood.

The main danger in the near future lies in that it reaches human stupidity after which it takes off on it's own and clones itself faster than we can destroy it. (as building something that can build itself is a quadratic function. While destruction is linear.)

The first scenario might become true if we manage to let an AI increase it's own hardware. And there is research being done into similar fields (mainly aimed at self-repairing automated objects). But on the short term this possible threat is non existent.
Tbh it might never become reality.

Theres also the 3rd option, where someone writes a 'self learning virus' which infects all other pc's and eventually will "become concious and decide things on it's own" but this is pure bs ;).

Well atm all the future scenarios are hollywood. Yet some of them are likely and possible, that's why we don't speak of them by saying they're hollywood unless we're a calling-out-'bullshit'-for-fun jackass.

NO FUN ALLOWED

oh, okay ;(

Which scenario would you find likely then?

That AI becomes supremely intelligent is beyond doubt, the question is what will happen then. I'm not knowledgeable enough to comment on chances of these, but popular scenarios include the singularity, death of the human race, human augmentation and symbiosis with technology and so on.

The death of the human race wouldn't neccessarily be absolute, or final, the destruction of our economies and societies would suffice. Then, if the AI was smart enough to work on itself, it would replace humanity. If not, it'd just chill and do what it's programmed to do.

Well theres a difference between "intelligent" and "fast".
Intelligence has much more to do with understanding problems which were unfamilliar to you and using other already learned concepts to try and solve it.
At this moment there is nothing at all that even comes remotely close to that in AI.
What is done in AI is mainly based on speed and simplicity.

For instance you can make something that recognizes where any ball is within it's field of vision. Then moves towards that ball, and perform some predefined action.
This is being brought as "this robot learned how to play soccer".

However if you want to teach the same robot to recognize where a cube is located, and pick it up. You'll have to make it "relearn everything".

But heres the funny bit. Once you try to let it learn both(!) the quality of both decreases dramatically. Simply because "the difference between a box and a ball" is too complex.

So an AI can learn how to do 1 specific task well. But try to fit multiple tasks in there and it goes banana's.
Humans are kinda similar though where we lack the ability to multitask.
Ofcourse it's possible to create seperate AI's in 1 robot, 1 AI for each task. And then put in some schedualling such that at time+situation X the robot prefers to perform task Y.
But still each task would be completely seperated from the others.

Which to our understanding would be a severe lack of intelligence.

That is a good and important point, but I believe it only pushes the possibility of intelligent AI further into the future. There's absolutely no reason any of these couldn't be overcome with time.

AI would need some very complex 'administration system' similar to human brain. But then again i like this idea that the brain is far from being understood completely and it is doubtful that it ever will be, because if it was only as complex as that we could completely understand it, how would we then be intelligent enough to understand it?

just wanted to leave this idea here. I don't think the AI system would need to be as complex as the human brain, but it is an interesting way to look at it.

You can divide any complex system into smaller parts, and understand those. So human-like intelligence will too be possible, only it might, as you suggested, come later than a more intuitively designed intelligence.

Yea but then there is emergence and the whole is more than the sum of its parts and so on. But this is another debate. Sure you are not wrong at some point AI might be that intelligent but atm there are no signs for it.

You mean consciousness? It is a philosophical issue, and it's relevant, but it can be considered secondary in this topic, because machines can be exactly as smart without it.

But we will reach it [above human level intelligence]. Experts estimate it at 2040-2050, but even they're wrong and it won't be until 2150, it will happen. I haven't heard an expert say it won't happen, except for nukes blowing up the planet of course.

Well... at this point in time I can't say if we will ever be able to make something which is close to human-like learning capability.

But like I said earlier. The main threat of AI is not that it reaches human like ability, but that it reaches human stupidity.

If we "teach" "something" to replicate itself... Well... then we could run into problems.
(ofcourse there are a lot of if's and buts there... for instance in order for it to become a problem the origional thing would have to know how to build a battery and have a long enough lifespan to reach all of the required resources / power.
But these kinds of things are somewhat close to being possible. That is the process is fairly simple and could be learnable...)
Basically if we can teach something to stupidly iteratate something of which each product then also stupidly iterates something... Well the process itself would speed up very rapidly ;)

We will reach it [above human level intelligence]. Experts estimate it at 2040-2050, but even they're wrong and it won't be until 2150, it will happen. I haven't heard an expert say it won't happen, nor am I aware of any reason it couldn't. Except for nukes blowing up the planet of course. The last resort is to simply replicate the evolutionary process or the human brain, but it won't likely come to that.

You're right in that engineering is the crucial part. If we can make AI stupider than humans in all areas, but smarter in Engineering, it will be able to upgrade itself very quickly. If we made AI very good in speech recognition for example, it wouldn't be nearly as progressive.

You could listen to the podcast I linked, the interview is pretty good.

Well.... On the topic of AI in software I'm a much more relevant source of information than any philosopher.
I am writing a paper on how philosophical arguments have lead the field of AI to believe certain "facts" which are factually wrong.

Bottom line is the philosophers all assume (in their papers) that if an AI can learn task 1, after that learn task 2. And that the addition of task 2 will have no influence on the task 1.

Which is technologically wrong.

This is not "new" I found similar stupidity in the most origional papers on things like 'neural networks'. Although they do talk about certain limits of learning, but forget that somehow there must be a clear distinction to start with.

I kinda do that in my spare time though when I'm not lurking around here / doing other irrelevant stuff.
So there should be others who actually get paid to do so who should be doing a better job than me... Unfortunately the media and even the scientific community is not very fond of our views.

The media doesn't like it because it doesn't sell. The scientific community doesn't like it because theres 30+ years of research in which this assumption has been shown to hold. Which is mainly due to lack of processing power in the past (thus limiting what could be learned heavily, leading to never reaching any convergence on the algorithms)

p.s. I did study AI (at the computer science department) at the university btw... And yes I took some extra classes on synchronous processes as well.
p.p.s. theres also AI on more social studies / philosophical studies. These are the idiots who think that we can build robots with human like capability.

Well, I believe the philosopher is up to date with the researched and in contact with experts, so I wouldn't rate his opinion so low. And I'm not going to stop believing the experts because of you all of a sudden, no offense but I don't know anything about your credentials. Though the idea you're pitching is interesting. How much of a polymath can you be lol, this is surprising.

You took a course in AI, or did you actually study AI as the focus of your degree? I'm not questioning your netcode views, so let's not get into that.

When I took it (around 4 years ago) it was called Intelligent Systems.
Which was a 2 years masters study into basically everything regarding autonomous software systems (so excluding hardware/robotics).

I somewhat keep in touch but the gap between philosophical papers and the mathematical ones was already so huge (when I actively read them) that you can call bullshit on most.
In some cases the studies are very similar to bad studies into medication.... Where if you look closely the odds of you obtaining some result before hand is 50 / 50 depending on lots of factors. And the ones which are succesful get published, the ones that fail do not.
Then a few years later the one that proves that the previous one was actually "just due to random factors" gets published and a few years later theres a research that proofs the disproval was wrong. Untill eventually one side just gives up.

Cool. So do you have an abstract now, or will you have it sometime later? I'd love to read it if it's possible.

Would take a while I think.

But to simply explain it:

Pretty much all AI algorithms output a "class".
So say that you want an AI to learn the difference between an apple and an orange. You'd have 2 classes (apples and orangjes) as possible results.
Obviously if you could seperate on color this is easy.... But let's say that you cannot detect color (to make it interresting), you only have greyscale images.

In order to "teach this to the AI" you
1) take pictures of a lot of oranges and a lot of apples. Let's say you photograph 500 of each leading to 1000 seperate images.

2) Label all of these according to what fruit it is.
So you write Orange next 500 images of oranges, and Apple next to the other 500 images.

3) Images are in general really high resolution. This is horrible for pc's, so you don't actually use the images. (Some do(!) but you'll understand why this is really really bad in a minute)
So these images are then manually pre-analysed. You can obtain a lot (N) of different properties of the fruit from the images. For instance you might look at "to what degree are they perfectly round" or "which greyscale are they" or "is it still attached to a branch" or "is there a sun in the image" or "how large is the fruit compared to the picture" or "god knows what kind of bullshit non relevant property the image has".
You store this information by creating an table with N + 1 records.
Which has N properties + 1 label.
So for instance if you take 3 properties you'd have something like:
"0.4576, 1.34532, 2.3453 Apple" as the first line in your table which describes the 1st image you analysed. You then continue with the next image and place that on the next line.
So you end up with a table of N+1 x 1000 records.

4) Split this dataset into a random "learning" and a "testing" set. Usually an 80 / 20 divide. Where you assign, at random, 80% (of the lines in your table) for learning and 20% for testing.
Which means that the eventual algorithm has to properly seperate aproximately 100 Apples from 100 Oranges. After having learned what the difference is from looking at images of 800 pieces of fruit.

5) you take all these properties (which are labeled) and you present the learning set to a "chosen algorithm". Which then learns how it can best seperate the data.

6) You test the result of this by running over your randomly chosen test set.
If it identifies 90% correctly, you plan a party, write a paper and get wasted.

If it does not identify 90% correctly you do one of the following steps:
- Randomly split the dataset differently (go to step 5)
- Change the input. (go to step 3)
- Give more input. (go to step 3)
- Pick another algorithm (go to step 5)

And you then iterate from the corresponding step untill you hit 90%.




Now this still "sounds fairly reasonable" and I actually accepted this reasoning for a course... although towards the end I had my doubts (this course was a shared course between computer science and philosophy).

The reason I had doubts is the following:

First of all if you just iterate through all algorithms, and each algorithm looks at the data in a significantly different way. Then the chance of one of them "fitting the data well" is quite reasonable. There are around 5~10 very significantly different ways, and each of those can be tweaked with settings or by altering functions (which makes them insanely different from the same algorithm with another function).

Secondly the more dimensions your data has the easier it gets to seperate.
Let's stick with the apples vs oranges here:
if you would place 4 apples and 4 oranges on a line at random. There is a HUGE chance that this line is not 4 apples followed by 4 oranges.
So you cannot seperate the set based on 1 property (place on the line).
But when you place them at random on the table, and try to find a line that seperates all apples from all oranges... There is a fair chance that you can do this.
When you hang them randomly distributed across your kitchen and you are allowed to seperate it using a large piece of paper, chances are very high you are able to do this.
So if you "up the dimensionality" the "seperation becomes easier" (this is a well known issue already. But it has been done SOOO often in papers)

Thirdly if you just randomly pick test/learning sets long enough... you'll end up with a test set that is purely oranges or purely apples. In which case your algorithm just has to output a single class.

Forth you can easily manipulate the data itself.
For instance if you photograph all apples during the day, and all oranges during the night. For instance at different lighting schemes....
It's easy to seperate on this quality.

And there are several other things which add up to this (mainly having to do with the properties of the algorithm function which you choose).

So very often people "claim good results" and then they write a paper.
But if you'd actually use the results in a dynamic environment.... well suddenly it all goes to hell.

This is even the case with the soccer robots. Where there are rules about how the ball will look exactly such that it is easier to learn.
Also a lot of experiments on robots have been known to actually alter lighting on the scene in order for the camera to pick up the right values such that it would fall in the correct range for the learned seperation vector etc.

(an abstract of the paper, of course)

To answer how much of a polymath I can be...
When I was 15 ~ 16 I got tested for dyslexia (which I have).
The results were that I had finished university according to almost every math test and was worse relative to my age on all language things.

Which is even today frustrating because I sometimes fail to convey the message clear enough due to my jihad with language :p

Next to that I also am not a very tallented painter...

So all in all... I'm just your regular human but with some weird skew towards mathematical subjects ;)

The results of the test at the age of 15-16 were such that you were on graduate level in maths? Or did you just take the dyslexia test at 15-16 and finished university at the appropriate time?

results of the test were that I was on graduate level in math.

haha. Good for you :)

although it seems like math has caught up with you.

Problem with this logic is that it assumes that there's no such thing as genetic engineering. If evolution is too slow to keep up with machines (which it is) we can just mod our genes ourselves.

http://youtu.be/4PLvdmifDSk?t=1m40s

This thread is somewhat of a proof that AI might outsmart humans and destroy us sooner then we think :D

Nothing can stop skynet and judgement day

This. People have a hard time understanding how much improvement that can be done to AI/Machine learning with the support of us humans. Once AI will be able to adapt learning algorithms the iteration cycles will be done in hours/minutes/seconds. No one really knows what that will lead to.

Not saying we are even close to reaching that level of AI but we're slowly getting there. I work within the automotive industry (IT) and I've seen some giant leaps in the last 2-3 years that I thought we wouldn't see so soon. It's basically a combination of machine learning techniques hooked up to on-the-fly predictive analytics of various types of data along with amazing robotics/mechanical engineering.

You do not need to worry about AI.

So I have this homework and you seem smart, can you please help me out?

What's 3.99999999..99 / 4 ?

0.99999999..9725

But if the .. could be assumed to represent an infinite length of 9's then you cannot add 99 at the end. And thus the exercise is wrong.

...but if it's all 99 to the end, how can the end be not 99?

it can't end on 99 if it's infinite.
If however it has an end then it can.

...so it's infinite and has an end? it's becoming more puzzling by the minute

learn to read.
then reread my post.

...dude pls

Do you understand what it means when you place an I infront of an F?

Its about the thing that the brain doesnt see, amirite ?

I'd say that you get an IF.
Which you should read as an IF.

But IF the .. could be assumed to represent an infinite length of 9's THEN you cannot add 99 at the end. And thus the exercise is wrong.

next post:

it can't end on 99 IF it's infinite.
IF HOWEVER it has an end then it can be.

I dunno how often I should say the same thing before it is read correctly.

what is this, some kind of reverse trolling? Or are you unable to detect trolling yourself? |]

What if you change to a base 12 number system, does not all you problems disappear?
http://en.wikipedia.org/wiki/Duodecimal

Because there are now 12 equal parts to a whole number
12/3 = 4 or 1/3. Then 4 x 3 = 12 ( a whole number)
No problem here. This whole .999 business only occurs because you chose to use base 10 number system. It does not have to be so.

Now for this AI rubbish. AI is just formula's. It has no emotion or desire or feelings. It only makes choices based on pure logic or similar.

An AI driving a vehicle does not avoid crashes because it is concerned for your well being, and this kind of reasoning (human emotion) is impossible so far to program. I don't think we even understands what gives us this ability ourselves so good luck to ever reproduce that in a machine anytime soon.

regarding number systems... another one that doesn't understand trolling, lol.

Regarding AI, expert polls average out at human level intelligence 2040-2050. I'm not saying I can put all the dots on the i's, but I'm trusting them instead of you.

I make evil insulting remarks whenever I feel like being trolled.
Stull like "learn to read" ;)

Iniquitous.

that's how I feel about trolling so. Imho my behavior is justified ;)

It can easily be cute and cheeky.

nope, can't be done.

lol

Nice. I in front of f comment. And learn to read

i cant be arsed to read through the blocks of texts here so excuse if i say stuff already mentioned. machines have always in a way replaced men in other words theyve always reduced workforce. ultimatively they are simply better in commercial terms it really took off in the 19th century and never stopped. now that we can give them a brain its not only agricultural equipment that replaces people it is machines printing tickets for bus train plane or laser stuff doing surgery or drones maintaining the american reich. what i want to say is that there is no reason why machines wouldnt replace all work eventually but seeing that as hostile would be foolish. a scenario where ai turns against us would mean a greater bug than strafe jumping and overbounce combined..

there is no reason why machines wouldnt replace all work eventually but seeing that as hostile would be foolish

Do you think every human would benefit from the robot's work equally? If they do all the work, then there's nothing left for humans to do. Will the average Joe starve to death? Will he rebel? Or will we find a way to keep everyone happy?

Not having anything to do by ourselves could be a disaster for the masses if it's not handled properly.

the machine replacing all our do is quite fascinating because it is a revolution. what would you do if you wouldnt have to do anything? its the end of currency or further the end of banks, maybe even wars. 150 years ago a young david thoreau carelessly exclaimed in walden: "If we only respected what is inevitable and has a right to be music and poetry would resound along the street." he probably didnt know how right he would be in the machine age.

the hypothesis is that once AI becomes intelligent enough, it will be able to reprogram or rebuild itself exponentially and infinitely. there is no predicting its intentions or even what it will be capable of doing, and it would have no evolutionary motive to be friendly towards humans.

http://en.wikipedia.org/wiki/Technological_singularity

thats a valid point but i think ai will always have our rules implemented.The film Irobot is probably really close to a possible reality where smart robots have 3 basic rules:dont harm a human being, save one if you can and so on. no matter how smart we get them we wont forget to give them our ethics

The Asimov rules are a joke. Example 1: You know how to prevent harm from coming onto humans? Stop humans from being ever born or conceived.

Rule 0.

A.I. from Allah Intelligence?

Anonymous Ignorance

Only because he himself is basicly AI

now we will have bots in real life?
=/