Staredit Network > Forums > Lite Discussion > Topic: Humans Need Not Apply
Humans Need Not Apply
Oct 6 2014, 6:11 am
By: Apos  

Oct 6 2014, 6:11 am Apos Post #1

I order you to forgive yourself!

Kinda late, but here is an interesting video from C.G.P Grey. It talks about how most of our jobs can be automated to be more cost efficient, eventually replacing us by robots, computer programs, etc.

He claims this is not science-fiction, and is currently happening right now.



Probably relevant: Snip




Oct 12 2014, 6:21 am MasterJohnny Post #2



If I get my master's degree in math, I don't feel like I could be replaced by a robot anytime soon...



Philosophy deals with unanswered questions. Religion deals with unquestioned answers. I am a Mathematician

Oct 12 2014, 2:03 pm Roy Post #3

An artist's depiction of an Extended Unit Death

Quote from MasterJohnny
If I get my master's degree in math, I don't feel like I could be replaced by a robot anytime soon...
That's some optimism you have there. What makes you think you'll be irreplaceable? I feel like mathematics and computer science should be among the easiest careers to replace, as it can be a pure software solution.

Once we can achieve a human-like artificial intelligence, we're done. There will be literally nothing we can do that a robot could not. We just need to crack the code and turn this organic computer into a mechanical one. Then the AI can write smarter and more efficient AIs and design new hardware to run them on ad infinitum, and they'll be solving problems in minutes that would have taken us decades. A self-driving car, for example, would be a completely solved problem once we have human AI (though we'll probably have self-driving cars before then).

The question is: how long will it take for us to design an artificial AI that can function the same as a human brain? The video suggests for many jobs, we don't even have to wait for this.




Oct 12 2014, 5:28 pm Apos Post #4

I order you to forgive yourself!

Quote from Roy
Quote from MasterJohnny
If I get my master's degree in math, I don't feel like I could be replaced by a robot anytime soon...
That's some optimism you have there. What makes you think you'll be irreplaceable? I feel like mathematics and computer science should be among the easiest careers to replace, as it can be a pure software solution.

Once we can achieve a human-like artificial intelligence, we're done. There will be literally nothing we can do that a robot could not. We just need to crack the code and turn this organic computer into a mechanical one. Then the AI can write smarter and more efficient AIs and design new hardware to run them on ad infinitum, and they'll be solving problems in minutes that would have taken us decades. A self-driving car, for example, would be a completely solved problem once we have human AI (though we'll probably have self-driving cars before then).

The question is: how long will it take for us to design an artificial AI that can function the same as a human brain? The video suggests for many jobs, we don't even have to wait for this.

Kind of a scary though. From the video, about 45% of the working force can easily be replaced by robot labor for cheaper. I'd assume that many people being unemployable would start a war?




Oct 12 2014, 5:30 pm NudeRaider Post #5

The entire universe has been neatly divided into things to (a) mate with, (b) eat, (c) run away from, and (d) rocks -Terry Pratchett

Quote from Roy
A self-driving car, for example, would be a completely solved problem once we have human AI (though we'll probably have self-driving cars before then).
We already have that: http://en.wikipedia.org/wiki/Google_driverless_car




Oct 12 2014, 5:35 pm jjf28 Post #6

Oh bother...

Quote from MasterJohnny
If I get my master's degree in math, I don't feel like I could be replaced by a robot anytime soon...

What do you know that wolfram alpha doesn't? :P



Rs_yes-im4real - Clan Aura - jjf28.net84.net

Reached the top of StarCraft theory crafting 2:12 AM CST, August 2nd, 2014.

Oct 12 2014, 5:38 pm Roy Post #7

An artist's depiction of an Extended Unit Death

Quote from Apos
I'd assume that many people being unemployable would start a war?
Ideally, it would start economic reform. (I assume that's why you linked to UBI at the end of your post.)

Google's self-driving car is not consumer-approved, so "we" is subjective here. They still have some technical hurdles to overcome, which is why it's not a completely solved problem. But yes, the timeline of having them by 2024 is most likely going to be before human AI (and other companies are aiming for even earlier, such as 2020).




Oct 12 2014, 5:56 pm Zoan Post #8

Math + Physics + StarCraft = Zoan

I think math - especially higher level math - requires the most creativity out of any subject, including art related things. That would be extremely difficult to engineer into an A.I.

Also, as a side note, I think that human decisions are only made through random low-level radiation of atoms within the human brain. So, if people were to somehow measure the probability of each individual atom within people's brains' of radiating, they could predict those people's actions. I also think that external stimuli probably affect how the atoms in our heads radiate, and thus influence our actions.

Basically, I think if we made a robot wired in such a way that it had atoms that radiated at random intervals, and somehow were to account for each atom's probability and analyse them together as a whole, we could get a robot that behaved 'randomly but predictably,' just like a person.

Though I also might just be an idiot/crazy person.

I should also add that when I say radiate, I don't necessarily mean actual radiation - I mean, the way I use it accounts for other quantum related phenomena, like tunneling and stuff. Then again, I've had little quantum and don't reallly know what I'm talking about. So don't take what I say extremely literally, but allow yourself some interpretation - just try to get the jist of my idea, if you care lol.

Post has been edited 5 time(s), last time on Oct 12 2014, 6:08 pm by Zoan.



\:rip\:ooooo\:wob\:ooooo \:angel\: ooooo\:wob\:ooooo\:rip\:

Oct 12 2014, 7:48 pm MasterJohnny Post #9



Quote from Roy
Quote from MasterJohnny
If I get my master's degree in math, I don't feel like I could be replaced by a robot anytime soon...
That's some optimism you have there. What makes you think you'll be irreplaceable? I feel like mathematics and computer science should be among the easiest careers to replace, as it can be a pure software solution.

I don't think we can ever really create ai that can do proofs. I also don't think computer scientists would be easy to replace. Who would fix these robots? Another robot?


Quote from jjf28
Quote from MasterJohnny
If I get my master's degree in math, I don't feel like I could be replaced by a robot anytime soon...

What do you know that wolfram alpha doesn't? :P

Wolfram alpha is more for computation. It can't do proofs. Like you can ask if there exists a bijection between the integers and the natural numbers and wolfram cannot tell you if such a bijection exists



Philosophy deals with unanswered questions. Religion deals with unquestioned answers. I am a Mathematician

Oct 12 2014, 8:42 pm Roy Post #10

An artist's depiction of an Extended Unit Death

Quote from MasterJohnny
I don't think we can ever really create ai that can do proofs. I also don't think computer scientists would be easy to replace. Who would fix these robots? Another robot?
What makes a human brain unique to being able to do mathematical proofs? Or to write software? Is there something about the way we think that cannot ever possibly be replicated?

Chess was once assumed to be a logical game in which no computer could possibly beat a skilled player in. Of course, this was ages ago, technologically speaking, as the first computer to beat skilled players was in the 1980s, and ever since 2005-2006 there has been no human that could beat the most skilled chess computer. Now we feel the same way about Go, and I'm looking forward to when an AI can dominate that game as well (we're getting reasonably close already, with only a 3-4 stone handicap for the AI on a full-sized board). That's not to say that these specialty AIs are going to eliminate jobs: it's more to illustrate that what we assume is not possible to replace with a specialty AI can, in fact, be replaced by it. But a true revolution won't occur until we have a human AI, rather than one specifically designed for a given task.

And yes, robots would make robots, and robots would fix robots. That would be a trivial task (one that I believe we already do today, granted with human oversight).

Quote from MasterJohnny
Wolfram alpha is more for computation. It can't do proofs. Like you can ask if there exists a bijection between the integers and the natural numbers and wolfram cannot tell you if such a bijection exists
We already have AI that can come up with the more straightforward mathematical proofs, and while they have not yet developed complex mathematical proofs, they can verify them. In fact, computer-assisted proofs exist through these means, such as the four color theorem.

Is this replacing mathematicians today? No, not even close. Will it ever in the future? Without a doubt in my mind.




Oct 12 2014, 9:54 pm Sand Wraith Post #11

she*

Quote from Roy
Quote from MasterJohnny
If I get my master's degree in math, I don't feel like I could be replaced by a robot anytime soon...
That's some optimism you have there. What makes you think you'll be irreplaceable? I feel like mathematics and computer science should be among the easiest careers to replace, as it can be a pure software solution.

Once we can achieve a human-like artificial intelligence, we're done. There will be literally nothing we can do that a robot could not. We just need to crack the code and turn this organic computer into a mechanical one. Then the AI can write smarter and more efficient AIs and design new hardware to run them on ad infinitum, and they'll be solving problems in minutes that would have taken us decades. A self-driving car, for example, would be a completely solved problem once we have human AI (though we'll probably have self-driving cars before then).

The question is: how long will it take for us to design an artificial AI that can function the same as a human brain? The video suggests for many jobs, we don't even have to wait for this.

Wait, what? Are you sure about that considering the Halting problem and NP-hard problems with respect to NP Completeness? (I haven't studied yet NP-hard and NP completeness.) Both problems are, to my knowledge, probably the two main problems for which I expect computers themselves never to be able to find solutions, and for which, potentially, is not achievable in the time frame presented by logistics with respect to environmental sustainability.




Oct 12 2014, 10:56 pm Roy Post #12

An artist's depiction of an Extended Unit Death

Quote from Sand Wraith
Wait, what? Are you sure about that considering the Halting problem and NP-hard problems with respect to NP Completeness? (I haven't studied yet NP-hard and NP completeness.) Both problems are, to my knowledge, probably the two main problems for which I expect computers themselves never to be able to find solutions, and for which, potentially, is not achievable in the time frame presented by logistics with respect to environmental sustainability.
It needn't solve unsolvable problems to be a human AI.

In regards to the Halting problem, an extension to my IDE can tell me when my code has infinite recursion or will never exit a loop. Programs can recognize if code will terminate to the same ability as we can, by understanding the code itself. The Halting problem simply states that there is no algorithm that exists that can determine if a program will terminate. Computers cannot solve this problem in the same way that you couldn't solve the problem if given a program that either runs for an arbitrarily long time or never terminates.

What you're suggesting is that computers will have to solve problems that humans cannot solve in order to qualify as human AI, but really they just have to be able to solve the same problems we're capable of solving.




Oct 12 2014, 11:11 pm Sand Wraith Post #13

she*

Quote from Roy
Quote from Sand Wraith
Wait, what? Are you sure about that considering the Halting problem and NP-hard problems with respect to NP Completeness? (I haven't studied yet NP-hard and NP completeness.) Both problems are, to my knowledge, probably the two main problems for which I expect computers themselves never to be able to find solutions, and for which, potentially, is not achievable in the time frame presented by logistics with respect to environmental sustainability.
It needn't solve unsolvable problems to be a human AI.

In regards to the Halting problem, an extension to my IDE can tell me when my code has infinite recursion or will never exit a loop. Programs can recognize if code will terminate to the same ability as we can, by understanding the code itself. The Halting problem simply states that there is no algorithm that exists that can determine if a program will terminate. Computers cannot solve this problem in the same way that you couldn't solve the problem if given a program that either runs for an arbitrarily long time or never terminates.

What you're suggesting is that computers will have to solve problems that humans cannot solve in order to qualify as human AI, but really they just have to be able to solve the same problems we're capable of solving.

good point




Oct 13 2014, 12:12 am MasterJohnny Post #14



Quote from Roy
We already have AI that can come up with the more straightforward mathematical proofs, and while they have not yet developed complex mathematical proofs, they can verify them. In fact, computer-assisted proofs exist through these means, such as the four color theorem.

Is this replacing mathematicians today? No, not even close. Will it ever in the future? Without a doubt in my mind.

What AI do we have that can do straightforward proofs? Computer-assisted proofs currently are just proofs that check a large but finite amount of cases. They don't really "think".



Philosophy deals with unanswered questions. Religion deals with unquestioned answers. I am a Mathematician

Oct 13 2014, 12:59 am jjf28 Post #15

Oh bother...

Any activity that can be broken up into sequential steps can be programmed with enough time, smarts, and computing power behind it.

How humans create proofs (from my experience):

1.) Propose a rule (random with a bias towards principles already shown inductively)
2.) Lay out definitions (use natural language)
3.) Do a quick check for counter-examples
4.) Select an approach (ex: pure algebra, using series', using iffs, bias towards what works with similar problems)
5.) Determine what the approach would have to result in
6.) Use all known theorems, identities, and so fourth to manipulate the 'equations', most related manipulators first (by recalling similarities to other proofs) with a hint of randomness, tend towards more simplified states unless it returns to a previously explored, simplified state
7.) If the problem isn't solved return to 6 (many times)
8.) If the problem still isn't solved return to 4 (till no more relevant approaches exist)
9.) Retrace steps, see if the steps within the proof can be simplified
10.) Present steps

All seems quite doable, heuristics for similarities, giant databases of theorems and identities, randomness here and there... of course new approaches to proofs would remain in the domain of humans unless someone fleshed out steps for this, which i'm sure is possible :)

edit: O, and humans of course 'break the rules' sometimes, where there's room for that (where something is prudent but not necessary) the computers could have a weighted random chance of breaking a rule.

Post has been edited 2 time(s), last time on Oct 13 2014, 1:07 am by jjf28.



Rs_yes-im4real - Clan Aura - jjf28.net84.net

Reached the top of StarCraft theory crafting 2:12 AM CST, August 2nd, 2014.

Oct 13 2014, 1:27 am Roy Post #16

An artist's depiction of an Extended Unit Death

Quote from MasterJohnny
What AI do we have that can do straightforward proofs?
Well, in the 1950s a program proved that the sum of two even integers is even. (If you're expecting any particularly exciting mathematical proofs, I'm sorry to disappoint.) If you're interested, this particular study is called Automated Theorem Proving.

The primary problem with this approach (I think) is that a computer cannot discern an interesting proof from an uninteresting one, and synthetic reasoning expands exponentially, meaning the computer will create a very large number of uninteresting proofs. In this aspect, the program has to be "guided" by giving it a goal, I believe (but I could be mistaken: it certainly could be much more automated today than it was several decades ago, or maybe they're solving by means other than synthetic reasoning). I need to do more research instead of guessing at potential limitations.

Here's an interesting blurb from the article:
More ambitious was the Logic Theory Machine, a deduction system for the propositional logic of the Principia Mathematica, developed by Allen Newell, Herbert A. Simon and J. C. Shaw. Also running on a JOHNNIAC, the Logic Theory Machine constructed proofs from a small set of propositional axioms and three deduction rules: modus ponens, (propositional) variable substitution, and the replacement of formulas by their definition. The system used heuristic guidance, and managed to prove 38 of the first 52 theorems of the Principia.

And:
Interactive provers are used for a variety of tasks, but even fully automatic systems have proved a number of interesting and hard theorems, including at least one that has eluded human mathematicians for a long time, namely the Robbins conjecture. However, these successes are sporadic, and work on hard problems usually requires a proficient user.

Quote from MasterJohnny
Computer-assisted proofs currently are just proofs that check a large but finite amount of cases. They don't really "think".
Proof by exhaustion is a mathematical proof. Though you are correct that it's not a "thinking" process as it is more a verification of a given conjecture.

Do you think it impossible for a computer to "think"? A learning algorithm is trivial to write (I've written one myself, and I'm not a super-genius or anything), and it's certainly replicating a method of thinking. Of course, these trivial algorithms are for specialized tasks, as I mentioned earlier. However, with IBM's Watson as an example, we're paving the way for Cognitive Computing (or Neuromorphic Engineering), which is getting us even closer to human-like intelligence.

Post has been edited 6 time(s), last time on Oct 13 2014, 1:54 am by Roy.




Oct 15 2014, 3:16 pm Sacrieur Post #17

Still Napping

Top level executives and business owners aren't immune from automation either. I believe the video overlooks the fact that companies can be run better and more effectively with a robot brain at the helm. It's logistics, computers are good at that.

In the end I'm not sure what will happen? It's likely some sort of neo-communism will take hold. Or something like Star Trek.



None.

Oct 15 2014, 5:46 pm Apos Post #18

I order you to forgive yourself!

Guess our future lies in us playing video games. Oh wait, that can probably be automated too....




Oct 16 2014, 2:07 am Oh_Man Post #19

Now on ICCUP, channel donuts

I don't know about AI replacing managerial roles, Sac. Those roles require a lot of interaction with humans. It's basically a human resources role.




Oct 16 2014, 2:20 am jjf28 Post #20

Oh bother...

But if we're supposing the roles below them have been replaced by AIs... then the human resource component of management would no longer exist.



Rs_yes-im4real - Clan Aura - jjf28.net84.net

Reached the top of StarCraft theory crafting 2:12 AM CST, August 2nd, 2014.

Options
  Back to forum
Please log in to reply to this topic or to report it.
Members in this topic: None.
[06:14 pm]
jjf28 -- there's also an interesting caveat in the Snowden case: he fled, he didn't stay or return to face justice (after he could be reasonably certain that there would be a court case and not a secret assasination), so he kinda abandoned the moral high-ground
[06:05 pm]
jjf28 -- NudeRaider
NudeRaider shouted: I like the logic http://i.imgur.com/uBxxgKY.png but state secrets are a valid thing, right? I mean secrets that (better) benefit the public if they are kept away from the public. Or is that just another brainwashed train-of-thought?
if we're significantly less safe because of the information he leaked (and I don't know enough to make that judgement) then it's the classic balancing act of safety/authoritarianism vs freedom/privacy
[05:10 pm]
IskatuMesk -- you can take my secrets from my cold dead buttcheeks
[04:47 pm]
NudeRaider -- I like the logic http://i.imgur.com/uBxxgKY.png but state secrets are a valid thing, right? I mean secrets that (better) benefit the public if they are kept away from the public. Or is that just another brainwashed train-of-thought?
[03:48 pm]
NudeRaider -- maybe you should start numbering them instead of giving them names eliminates that problem and makes things easier all around
[03:47 pm]
NudeRaider -- Mini Moose 2707
Mini Moose 2707 shouted: idk like 15 probs
after 10 or so you stop counting...
[02:46 pm]
Dem0n -- I thought it was gonna be a dick
[02:35 pm]
TF- -- all you need is a pipe and several cans of spinach
Please log in to shout.


Members Online: Roy, Azrael