Allow me, if you will, to take us years into the future (assuming humanity doesn't kill itself first). Humanity has been able to build androids.
Some are used for physical or dangerous work, others for service, other's still for 'personal use' (ew). They must be programmed to do their tasks.
What about 'care giving'.
To do so, humanity decides we have to program 'morality' into them, as morality is programmed into humans all the time - seems like a doable task.
For discussion:
Will humanity ever be able to program morality into a machine?
Why or why not?
Programming morality
Moderator: Moderators
-
- Prodigy
- Posts: 3187
- Joined: Wed Nov 11, 2020 11:32 am
- Has thanked: 1510 times
- Been thanked: 824 times
- Difflugia
- Prodigy
- Posts: 3044
- Joined: Wed Jun 12, 2019 10:25 am
- Location: Michigan
- Has thanked: 3276 times
- Been thanked: 2022 times
Re: Programming morality
Post #21Part of the question is whether or not human brains are actually solving the problems that they think they are. One computer science approach to "solving" unsolvable (or even just expensive) problems is to solve a different problem that is cheaper, but still has enough overlap that the solution is correct for the most common inputs. Since it's known that human brains do this, it's possible that humans can't solve uncomputable problems, either. An example of this is risk/reward analysis. Human brains are slow at doing math, so they evaluate some problems by performing computationally cheaper analysis that is "close enough." Casinos are profitable because they craft games that are on the wrong side of a boundary condition and human brains get the wrong answer ("keep gambling" instead of "cut your losses and go home").nobspeople wrote: ↑Mon Mar 28, 2022 1:37 pmSeems to be experts that are trying to solve this problem. Will they? Only time will tell. I wouldn't bet against them given enough time.Sherlock Holmes wrote: ↑Mon Mar 28, 2022 1:07 pmWell just to be clear the limitations on computability are inescapable, it is not a question of time or resources, it is not a current limitation but an absolute one, all any computer can do is compute computable things.
There's a significant position that human brains can't actually compute things that computers can't. The question, then, is to identify the techniques that humans use to "solve" uncomputable problems and adapt those methods to computers. Machine learning using neural networks is one of those methods and it's impressive both in speed and emergent accuracy, but it's suboptimal. Because it's not computationally optimal, the dataset used for training the network can bias the computed solutions. This has resulted in things like machine AI being racist because of racist training data even though researchers were unaware that the data were racist in the first place.
I have little doubt that computers can be programmed for morality, but it will likely mirror the same failings that human morality shows. In fact, it's likely that if computed morality does begin to surpass human morality, human beings will see the differences as incorrect output. It will be the same situation as sitting next to someone at a blackjack table that's convinced that I can "take the dealer's bust card" if I strictly play the odds.
My pronouns are he, him, and his.
-
- Prodigy
- Posts: 3187
- Joined: Wed Nov 11, 2020 11:32 am
- Has thanked: 1510 times
- Been thanked: 824 times
Re: Programming morality
Post #22I updated the quoted link to show how it's being worked on currently.Sherlock Holmes wrote: ↑Mon Mar 28, 2022 2:23 pmHave what? it is impossible, logically impossible. Did I misunderstand you perhaps?nobspeople wrote: ↑Mon Mar 28, 2022 2:18 pmNot saying it's an issue now. The link I provide earlier says this as well. But it also says computer scientist are working on the problem and in a couple of decades, the expect to have it.Sherlock Holmes wrote: ↑Mon Mar 28, 2022 2:12 pmNo, this is a question of rigorous logic, mathematics. These are provable claims, here's the proof that the halting problem is non-computable:nobspeople wrote: ↑Mon Mar 28, 2022 1:37 pmSeems to be experts that are trying to solve this problem. Will they? Only time will tell. I wouldn't bet against them given enough time.Sherlock Holmes wrote: ↑Mon Mar 28, 2022 1:07 pmWell just to be clear the limitations on computability are inescapable, it is not a question of time or resources, it is not a current limitation but an absolute one, all any computer can do is compute computable things.nobspeople wrote: ↑Mon Mar 28, 2022 12:56 pmThat there may not be computers that can do what humans do, I would add 'not yet'. I remember being told 'computers can't do this or that' when I was young, now many computers can. With advancements, who knows what computers will bring to the tableSherlock Holmes wrote: ↑Mon Mar 28, 2022 12:48 pmMy post was in error, this is what I intended to write:Bust Nak wrote: ↑Mon Mar 28, 2022 12:25 pmThat's the point. A computer doesn't have to be an purely algorithmic machines, the fact that such machines cannot solve non-computable things is a non-issue, because no one should be expecting a non-algorithmic computers to be solve such non-computable problems. If you objection is with the term "computers" itself then replace it with a generic "machines."Sherlock Holmes wrote: ↑Mon Mar 28, 2022 11:52 am But a "computer" solves computable problems, it cannot solve non-computable, that's why they're called "computers", there are non-algorithmic computers, an analog computer (though sometimes unpredictable) does not compute non-computable things.
"But a "computer" solves computable problems, it cannot solve non-computable, that's why they're called "computers", there are no non-algorithmic computers, an analog computer (though sometimes unpredictable) does not compute non-computable things.
OK lets look at that example?Bust Nak wrote: ↑Mon Mar 28, 2022 12:25 pmA non-computable function, as in one particular non-computable function? Yeah I can . But I cannot give any example of a machine that can solve general non-computable functions.Give me an example of a machine - any machine - that can compute a non-computable function, can you?
I'm not defending anything he claims, I merely pointing out that there are such views and that they seem to imply that the human brain can solve non-computable problems, if that's true then no computer can do what we can.Bust Nak wrote: ↑Mon Mar 28, 2022 12:25 pmIs there? We face the very same limitations as an machines. We can analyse non-algorithmic problems, but we can't solve them. Sometimes we make mistakes, sometimes we can't halt.As I mentioned there are strong indications that the human brain can solve non-algorithmic problems.
Well, he is not here to defend his claims. Care to have a go?the theoretical physicist Roger Penrose for example wrote about this in detail. If this is true then clearly no algorithmic (digital or analog) machine can do what the brain can do, it is fundamentally beyond the reach of machines.
https://futurism.com/the-evolution-of-a ... programmed
"It may be quite a bit further out, but to computer scientists, that means maybe just on the order of decades."
That said, I don't think humanity will ever be able to agree on if a computer will have morality as people have different views on morality that's ever changing.
Have a great, potentially godless, day!
-
- Prodigy
- Posts: 3187
- Joined: Wed Nov 11, 2020 11:32 am
- Has thanked: 1510 times
- Been thanked: 824 times
Re: Programming morality
Post #23In one of the articles, the author suggest the same thing as the bolded statements you provided. The article isn't saying (at least that I read) AI will have better morals than humanity, just that it may be possible, once they understand the thinking and brain activity on such moral issues, to program an AI to work 'as a human mind' would work with such queries.Difflugia wrote: ↑Mon Mar 28, 2022 2:27 pmPart of the question is whether or not human brains are actually solving the problems that they think they are. One computer science approach to "solving" unsolvable (or even just expensive) problems is to solve a different problem that is cheaper, but still has enough overlap that the solution is correct for the most common inputs. Since it's known that human brains do this, it's possible that humans can't solve uncomputable problems, either. An example of this is risk/reward analysis. Human brains are slow at doing math, so they evaluate some problems by performing computationally cheaper analysis that is "close enough." Casinos are profitable because they craft games that are on the wrong side of a boundary condition and human brains get the wrong answer ("keep gambling" instead of "cut your losses and go home").nobspeople wrote: ↑Mon Mar 28, 2022 1:37 pmSeems to be experts that are trying to solve this problem. Will they? Only time will tell. I wouldn't bet against them given enough time.Sherlock Holmes wrote: ↑Mon Mar 28, 2022 1:07 pmWell just to be clear the limitations on computability are inescapable, it is not a question of time or resources, it is not a current limitation but an absolute one, all any computer can do is compute computable things.
There's a significant position that human brains can't actually compute things that computers can't. The question, then, is to identify the techniques that humans use to "solve" uncomputable problems and adapt those methods to computers. Machine learning using neural networks is one of those methods and it's impressive both in speed and emergent accuracy, but it's suboptimal. Because it's not computationally optimal, the dataset used for training the network can bias the computed solutions. This has resulted in things like machine AI being racist because of racist training data even though researchers were unaware that the data were racist in the first place.
I have little doubt that computers can be programmed for morality, but it will likely mirror the same failings that human morality shows. In fact, it's likely that if computed morality does begin to surpass human morality, human beings will see the differences as incorrect output. It will be the same situation as sitting next to someone at a blackjack table that's convinced that I can "take the dealer's bust card" if I strictly play the odds.
Have a great, potentially godless, day!
Re: Programming morality
Post #24Yes I misunderstood, you were referring to morality and I thought you were talking about the halting problem.nobspeople wrote: ↑Mon Mar 28, 2022 2:28 pmI updated the quoted link to show how it's being worked on currently.Sherlock Holmes wrote: ↑Mon Mar 28, 2022 2:23 pmHave what? it is impossible, logically impossible. Did I misunderstand you perhaps?nobspeople wrote: ↑Mon Mar 28, 2022 2:18 pmNot saying it's an issue now. The link I provide earlier says this as well. But it also says computer scientist are working on the problem and in a couple of decades, the expect to have it.Sherlock Holmes wrote: ↑Mon Mar 28, 2022 2:12 pmNo, this is a question of rigorous logic, mathematics. These are provable claims, here's the proof that the halting problem is non-computable:nobspeople wrote: ↑Mon Mar 28, 2022 1:37 pmSeems to be experts that are trying to solve this problem. Will they? Only time will tell. I wouldn't bet against them given enough time.Sherlock Holmes wrote: ↑Mon Mar 28, 2022 1:07 pmWell just to be clear the limitations on computability are inescapable, it is not a question of time or resources, it is not a current limitation but an absolute one, all any computer can do is compute computable things.nobspeople wrote: ↑Mon Mar 28, 2022 12:56 pmThat there may not be computers that can do what humans do, I would add 'not yet'. I remember being told 'computers can't do this or that' when I was young, now many computers can. With advancements, who knows what computers will bring to the tableSherlock Holmes wrote: ↑Mon Mar 28, 2022 12:48 pmMy post was in error, this is what I intended to write:Bust Nak wrote: ↑Mon Mar 28, 2022 12:25 pmThat's the point. A computer doesn't have to be an purely algorithmic machines, the fact that such machines cannot solve non-computable things is a non-issue, because no one should be expecting a non-algorithmic computers to be solve such non-computable problems. If you objection is with the term "computers" itself then replace it with a generic "machines."Sherlock Holmes wrote: ↑Mon Mar 28, 2022 11:52 am But a "computer" solves computable problems, it cannot solve non-computable, that's why they're called "computers", there are non-algorithmic computers, an analog computer (though sometimes unpredictable) does not compute non-computable things.
"But a "computer" solves computable problems, it cannot solve non-computable, that's why they're called "computers", there are no non-algorithmic computers, an analog computer (though sometimes unpredictable) does not compute non-computable things.
OK lets look at that example?Bust Nak wrote: ↑Mon Mar 28, 2022 12:25 pmA non-computable function, as in one particular non-computable function? Yeah I can . But I cannot give any example of a machine that can solve general non-computable functions.Give me an example of a machine - any machine - that can compute a non-computable function, can you?
I'm not defending anything he claims, I merely pointing out that there are such views and that they seem to imply that the human brain can solve non-computable problems, if that's true then no computer can do what we can.Bust Nak wrote: ↑Mon Mar 28, 2022 12:25 pmIs there? We face the very same limitations as an machines. We can analyse non-algorithmic problems, but we can't solve them. Sometimes we make mistakes, sometimes we can't halt.As I mentioned there are strong indications that the human brain can solve non-algorithmic problems.
Well, he is not here to defend his claims. Care to have a go?the theoretical physicist Roger Penrose for example wrote about this in detail. If this is true then clearly no algorithmic (digital or analog) machine can do what the brain can do, it is fundamentally beyond the reach of machines.
https://futurism.com/the-evolution-of-a ... programmed
"It may be quite a bit further out, but to computer scientists, that means maybe just on the order of decades."
That said, I don't think humanity will ever be able to agree on if a computer will have morality as people have different views on morality that's ever changing.
-
- Savant
- Posts: 9856
- Joined: Mon Feb 27, 2012 6:03 am
- Location: Planet Earth
- Has thanked: 189 times
- Been thanked: 266 times
Re: Programming morality
Post #25My last response still apply, so use a different term. Call my proposal a thinking machine, if "computer" is to be narrowly defined only as something that solves computable problems.Sherlock Holmes wrote: ↑Mon Mar 28, 2022 12:48 pm My post was in error, this is what I intended to write:
"But a "computer" solves computable problems, it cannot solve non-computable, that's why they're called "computers", there are no non-algorithmic computers, an analog computer (though sometimes unpredictable) does not compute non-computable things.
It was a joke. I was referring to a simple file comparison app being able to tell you if an input file marches byte by byte, to a pre-selected non-halting algorithm, if there is a match, then it can 100% accurately tell you that the algorithm will not halt.OK lets look at that example?
If that's true then no Turing machines can do what we can. It's quite a jump to say no machines can, even if Penrose is correct.I'm not defending anything he claims, I merely pointing out that there are such views and that they seem to imply that the human brain can solve non-computable problems, if that's true then no computer can do what we can.
-
- Prodigy
- Posts: 3187
- Joined: Wed Nov 11, 2020 11:32 am
- Has thanked: 1510 times
- Been thanked: 824 times
Re: Programming morality
Post #26OK that makes sense. Thanks for the clarificationSherlock Holmes wrote: ↑Mon Mar 28, 2022 2:50 pmYes I misunderstood, you were referring to morality and I thought you were talking about the halting problem.nobspeople wrote: ↑Mon Mar 28, 2022 2:28 pmI updated the quoted link to show how it's being worked on currently.Sherlock Holmes wrote: ↑Mon Mar 28, 2022 2:23 pmHave what? it is impossible, logically impossible. Did I misunderstand you perhaps?nobspeople wrote: ↑Mon Mar 28, 2022 2:18 pmNot saying it's an issue now. The link I provide earlier says this as well. But it also says computer scientist are working on the problem and in a couple of decades, the expect to have it.Sherlock Holmes wrote: ↑Mon Mar 28, 2022 2:12 pmNo, this is a question of rigorous logic, mathematics. These are provable claims, here's the proof that the halting problem is non-computable:nobspeople wrote: ↑Mon Mar 28, 2022 1:37 pmSeems to be experts that are trying to solve this problem. Will they? Only time will tell. I wouldn't bet against them given enough time.Sherlock Holmes wrote: ↑Mon Mar 28, 2022 1:07 pmWell just to be clear the limitations on computability are inescapable, it is not a question of time or resources, it is not a current limitation but an absolute one, all any computer can do is compute computable things.nobspeople wrote: ↑Mon Mar 28, 2022 12:56 pmThat there may not be computers that can do what humans do, I would add 'not yet'. I remember being told 'computers can't do this or that' when I was young, now many computers can. With advancements, who knows what computers will bring to the tableSherlock Holmes wrote: ↑Mon Mar 28, 2022 12:48 pmMy post was in error, this is what I intended to write:Bust Nak wrote: ↑Mon Mar 28, 2022 12:25 pm
That's the point. A computer doesn't have to be an purely algorithmic machines, the fact that such machines cannot solve non-computable things is a non-issue, because no one should be expecting a non-algorithmic computers to be solve such non-computable problems. If you objection is with the term "computers" itself then replace it with a generic "machines."
"But a "computer" solves computable problems, it cannot solve non-computable, that's why they're called "computers", there are no non-algorithmic computers, an analog computer (though sometimes unpredictable) does not compute non-computable things.
OK lets look at that example?
I'm not defending anything he claims, I merely pointing out that there are such views and that they seem to imply that the human brain can solve non-computable problems, if that's true then no computer can do what we can.
https://futurism.com/the-evolution-of-a ... programmed
"It may be quite a bit further out, but to computer scientists, that means maybe just on the order of decades."
That said, I don't think humanity will ever be able to agree on if a computer will have morality as people have different views on morality that's ever changing.
Have a great, potentially godless, day!
Re: Programming morality
Post #27It's not a jump if we are speaking of symbol manipulation (which is all that a digital computer actually is) we can seemingly do symbol manipulation but a TM can only ever do symbol manipulation.Bust Nak wrote: ↑Tue Mar 29, 2022 4:48 amMy last response still apply, so use a different term. Call my proposal a thinking machine, if "computer" is to be narrowly defined only as something that solves computable problems.Sherlock Holmes wrote: ↑Mon Mar 28, 2022 12:48 pm My post was in error, this is what I intended to write:
"But a "computer" solves computable problems, it cannot solve non-computable, that's why they're called "computers", there are no non-algorithmic computers, an analog computer (though sometimes unpredictable) does not compute non-computable things.
It was a joke. I was referring to a simple file comparison app being able to tell you if an input file marches byte by byte, to a pre-selected non-halting algorithm, if there is a match, then it can 100% accurately tell you that the algorithm will not halt.OK lets look at that example?
If that's true then no Turing machines can do what we can. It's quite a jump to say no machines can, even if Penrose is correct.I'm not defending anything he claims, I merely pointing out that there are such views and that they seem to imply that the human brain can solve non-computable problems, if that's true then no computer can do what we can.
There's a great book I encourage people to checkout too, it's What Computers Still Can't Do: A Critique of Artificial Reason, by Hubert Dreyfus.
In there the author argues that there's basically no evidence whatsoever that the brain is a symbol manipulator or uses symbol manipulation and that therefore equating computer with brains is without serious foundation.
-
- Savant
- Posts: 9856
- Joined: Mon Feb 27, 2012 6:03 am
- Location: Planet Earth
- Has thanked: 189 times
- Been thanked: 266 times
Re: Programming morality
Post #28[Replying to Sherlock Holmes in post #27]
Why are you still talking about what Turing machines can and cannot do, when my point was what Turing machines cannot do is irrelevance since Turing machines aren't the only kind of machines around? I just don't see how your post has address anything I said in mine.
Why are you still talking about what Turing machines can and cannot do, when my point was what Turing machines cannot do is irrelevance since Turing machines aren't the only kind of machines around? I just don't see how your post has address anything I said in mine.
Re: Programming morality
Post #29We must face the fact that all man made machines are limited to computable problems, they cannot address non-computable problems, that's what I am saying here. Couple that with the claim (by Penrose and others) that the human brain seems able to solve non computable problems then we have the impasse.Bust Nak wrote: ↑Tue Mar 29, 2022 11:45 am [Replying to Sherlock Holmes in post #27]
Why are you still talking about what Turing machines can and cannot do, when my point was what Turing machines cannot do is irrelevance since Turing machines aren't the only kind of machines around? I just don't see how your post has address anything I said in mine.
Now if you are saying we have or could have a man-made mechanism that is able to solve non-computable problems then yes, that might be capable of "morality" but I don't see how we can get such a machine since all machines obey the laws of physics, computable functions.
-
- Savant
- Posts: 9856
- Joined: Mon Feb 27, 2012 6:03 am
- Location: Planet Earth
- Has thanked: 189 times
- Been thanked: 266 times
Re: Programming morality
Post #30[Replying to Sherlock Holmes in post #29]
Okay, since we are talking about impasse, here is my bottom line: Replicating a human brain neuron by neuron is merely a technical challenge, it is achievable in principle. Such a replica would then be a man made machine capable of addressing any problem a human can. Whether such problems include non-computable problems or not is a different matter all together.
Okay, since we are talking about impasse, here is my bottom line: Replicating a human brain neuron by neuron is merely a technical challenge, it is achievable in principle. Such a replica would then be a man made machine capable of addressing any problem a human can. Whether such problems include non-computable problems or not is a different matter all together.