Unlike statements of fact, which require no further work on our part, lies must be continually protected from collisions with reality. When you tell the truth, you have nothing to keep track of. The world itself becomes your memory, and if questions arise, you can always point others back to it. You can even reconsider certain facts and honestly change your views. And you can openly discuss your confusion, conflicts, and doubts with all comers. In this way, a commitment to the truth is naturally purifying of error.
I think this is actually a myth. It’s appealing, to us who love truth so much, to think that deviating from the path of the truth is deadly and dangerous and leads inevitably to dark side epistemology. But there is a trick to telling lies, such that they only differ from the truth in minor, difficult to verify ways. If you tell elegant lies, they will cling to the surface of the truth like a parasite, and you will be able to do almost anything with them that you could do with the truth. You just have to remember a few extra bits that you changed, and otherwise behave as a normal honest person would, given those few extra bits.
Worse, you can simply let people catch you, then get angry with them and bully them into accepting your claims not to have lied out of a mix of imperfect certainty and conflict avoidance. By doing this you condition them to accept the radical form of dominance where they have the authority to tell you what you are morally entitled to believe.
By doing this you condition them to accept the radical form of dominance where they have the authority to tell you what you are morally entitled to believe.
You’re not actually disagreeing with Harris. Crafting efficient lies that behave as you describe is hard, particularly on the spot during conversation. Practice helps, and having your interlocutor’s trust can compensate for a lot of imperfections, but it’s still a lot of work compared to just sharing everything you know
Hm, that gives me an idea: study lying as a computational complexity problem. Just as we can study how much computing power it takes to distinguish random data from encrypted data, we can study how much computing power it takes to formulate (self-serving) hypotheses that take too much effort to distinguish from the truth.
Just a thought...
(Scott Aaronson’s paper opened my eyes on the subject.)
I don’t know much about the problem in question, but there’s a related open problem in number theory.
Suppose I am thinking of a positive integer from 1 to n. You know this and know n. You want to figure out my number but are only allowed to ask if my number is in some range you name. In this game it is easy to see that you can always find out my number in less than 1+log2 n questions.
But what if I’m allowed to lie k times for some fixed k (that you know). Then the problem becomes much more difficult. A general bound in terms of k and n is open.
This suggests to me that working out problems involving lying, even in toy models, can quickly become complicated and difficult to examine.
Are you familiar with the seemingly similar question about the prisoners, king, and coin? I don’t know the name, but it goes like this:
There are n prisoners in separate rooms, each with a doorway to a central chamber (CC) that has a coin. One by one, the king takes a random prisoner into the CC (no one else can see what is going on), and asks the prisoner if the king has brought all prisoners into the CC by now. The prisoner can either answer “yes” or “I don’t know”. If he says the former and is wrong, all prisoners are executed. If he’s right, they’re released.
If If he says “I don’t know”, he can set the coin to heads or tails. The king may turn over the coin after a prisoner leaves (and before he brings the next in), but he may only do so a finite k number of times in total. (This is a key similarity to the number of lies in the problem you describe).
The prisoners may discuss a strategy before starting, but the king gets to listen in and learn their strategy. So long as the game continues, every prisoner will be picked inifinte times (i.e. every prisoner can always expect to get picked again).
Is it possible for the prisoners to guarantee their eventual release?
The answer is yes, and there’s a known bound on how long it takes. (Got this from slashdot a long time ago.)
Edit: Found it. Here’s the discussion that spawned it, and here’s the thread that introduces this problem, and here’s a comment with a solution. Apparently, the problem has a name it goes by.
Edit2: This also serves as a case study in how to present a problem as succinctly as possible. The only thing I got wrong about its statement was that the king chooses the order of the prisoners going into the CC (rather than it being random), although given the constraint that each prisoner is eventually brought in infinite times, and the strategy must work all the time, I don’t think it changes the problem.
Maybe I wasn’t clear. The blockquoted part is (my phrasing of) the problem statement. In the slashdot thread (and this is all from memory), several correct, bounded solutions were posted. I’ll try to find the thread. (IIRC the original phrasing had a cup instead of a coin.)
The intuition behind the existence of a solution is that the prisoners can effectively send infinite one-bit messages between each other, while the king can only block a finite number of them, so they just need to choose a leader and run some “message accumulator” protocol that will reach a certain state when all prisoners are certain to have been in the CC.
Edit: Wow, that was actually easy to find. Here’s the discussion that spawned it, and here’s the thread that introduces this problem, and here’s a comment with a solution. Apparently, the problem has a name it goes by.
Being embarrassed about your knowledge is anathema to rational conversation. You can see it in drug policy debates, where nobody talks about how relatively harmless marijuana is, for fear that people might know that they smoke it. You can see it in censorship debates, where no community member is going to stand up and say “hey, this porno doesn’t violate my standards, in fact it’s pretty hot”. We can stand around pretending to be good people, or we can get at the truth.
I’m more willing to admit to lying here, because I trust you guys more than most people to take that admission only for what it is, and no more.
Being embarrassed about your knowledge is anathema to rational conversation. You can see it in drug policy debates, where nobody talks about how relatively harmless marijuana is, for fear that people might know that they smoke it. You can see it in censorship debates, where no community member is going to stand up and say “hey, this porno doesn’t violate my standards, in fact it’s pretty hot”. We can stand around pretending to be good people, or we can get at the truth.
You sound like you’re advocating radical honesty. It seems like there should be a middle ground of making sure relevant information is introduced, but doing it in a way that minimizes derailing self-disclosure (or self-disclosure that could cost you in status).
Also, arguing from personal experience can be form of defection, shifting the conversation to an arena where one’s convincingness is proportional to one’s willingness to lie. (I think I have some comments saved that say that better than I can.)
When you tell the truth, you have nothing to keep track of. The world itself becomes your memory, and if questions arise, you can always point others back to it.
As any decent defense attorney will tell you: if you’re accused of something you didn’t do, this is still an extremely bad approach.
For a defendant, lying is the only thing worse than telling the truth. Telling the truth is still often a terrible idea, particularly for a person accused in the formal American legal system.
(Edited to change meaning to what I originally intended but typed incorrectly. Original words were “For a defendant, the only thing worse than lying is telling the truth,” but the above is what I had intended.)
Sam Harris, “Lying”
I think this is actually a myth. It’s appealing, to us who love truth so much, to think that deviating from the path of the truth is deadly and dangerous and leads inevitably to dark side epistemology. But there is a trick to telling lies, such that they only differ from the truth in minor, difficult to verify ways. If you tell elegant lies, they will cling to the surface of the truth like a parasite, and you will be able to do almost anything with them that you could do with the truth. You just have to remember a few extra bits that you changed, and otherwise behave as a normal honest person would, given those few extra bits.
Worse, you can simply let people catch you, then get angry with them and bully them into accepting your claims not to have lied out of a mix of imperfect certainty and conflict avoidance. By doing this you condition them to accept the radical form of dominance where they have the authority to tell you what you are morally entitled to believe.
*where you have the authority to tell them (?)
Yep. Sorry.
You’re not actually disagreeing with Harris. Crafting efficient lies that behave as you describe is hard, particularly on the spot during conversation. Practice helps, and having your interlocutor’s trust can compensate for a lot of imperfections, but it’s still a lot of work compared to just sharing everything you know
Hm, that gives me an idea: study lying as a computational complexity problem. Just as we can study how much computing power it takes to distinguish random data from encrypted data, we can study how much computing power it takes to formulate (self-serving) hypotheses that take too much effort to distinguish from the truth.
Just a thought...
(Scott Aaronson’s paper opened my eyes on the subject.)
I don’t know much about the problem in question, but there’s a related open problem in number theory.
Suppose I am thinking of a positive integer from 1 to n. You know this and know n. You want to figure out my number but are only allowed to ask if my number is in some range you name. In this game it is easy to see that you can always find out my number in less than 1+log2 n questions.
But what if I’m allowed to lie k times for some fixed k (that you know). Then the problem becomes much more difficult. A general bound in terms of k and n is open.
This suggests to me that working out problems involving lying, even in toy models, can quickly become complicated and difficult to examine.
Are you familiar with the seemingly similar question about the prisoners, king, and coin? I don’t know the name, but it goes like this:
The answer is yes, and there’s a known bound on how long it takes. (Got this from slashdot a long time ago.)
Edit: Found it. Here’s the discussion that spawned it, and here’s the thread that introduces this problem, and here’s a comment with a solution. Apparently, the problem has a name it goes by.
Edit2: This also serves as a case study in how to present a problem as succinctly as possible. The only thing I got wrong about its statement was that the king chooses the order of the prisoners going into the CC (rather than it being random), although given the constraint that each prisoner is eventually brought in infinite times, and the strategy must work all the time, I don’t think it changes the problem.
Doesn’t your comment on Slashdot indicate that there is no solution?
Maybe I wasn’t clear. The blockquoted part is (my phrasing of) the problem statement. In the slashdot thread (and this is all from memory), several correct, bounded solutions were posted. I’ll try to find the thread. (IIRC the original phrasing had a cup instead of a coin.)
The intuition behind the existence of a solution is that the prisoners can effectively send infinite one-bit messages between each other, while the king can only block a finite number of them, so they just need to choose a leader and run some “message accumulator” protocol that will reach a certain state when all prisoners are certain to have been in the CC.
Edit: Wow, that was actually easy to find. Here’s the discussion that spawned it, and here’s the thread that introduces this problem, and here’s a comment with a solution. Apparently, the problem has a name it goes by.
This is the comment that provoked mine. Your link and this do seem to be solutions, though.
There are some comments I wish I could delete from slashdot … and this site, for that matter … such as the parent.
Not that I am implying that it is normal to be honest, haha.
It is customary to add at the end of such confessions, “or so I’m told”, which is technically not a lie but merely an implicature.
Being embarrassed about your knowledge is anathema to rational conversation. You can see it in drug policy debates, where nobody talks about how relatively harmless marijuana is, for fear that people might know that they smoke it. You can see it in censorship debates, where no community member is going to stand up and say “hey, this porno doesn’t violate my standards, in fact it’s pretty hot”. We can stand around pretending to be good people, or we can get at the truth.
I’m more willing to admit to lying here, because I trust you guys more than most people to take that admission only for what it is, and no more.
You sound like you’re advocating radical honesty. It seems like there should be a middle ground of making sure relevant information is introduced, but doing it in a way that minimizes derailing self-disclosure (or self-disclosure that could cost you in status).
Also, arguing from personal experience can be form of defection, shifting the conversation to an arena where one’s convincingness is proportional to one’s willingness to lie. (I think I have some comments saved that say that better than I can.)
As any decent defense attorney will tell you: if you’re accused of something you didn’t do, this is still an extremely bad approach.
Definitely. If questions arise you should always point others back to your attorney! ;)
For a defendant, lying is the only thing worse than telling the truth. Telling the truth is still often a terrible idea, particularly for a person accused in the formal American legal system.
(Edited to change meaning to what I originally intended but typed incorrectly. Original words were “For a defendant, the only thing worse than lying is telling the truth,” but the above is what I had intended.)
Don’t defence attorneys (at least in the USA) heartily recommend shutting up as opposed to lying?
Yes.
.
Greater than signs are only necessary at the beginning of the paragraph, by the way.
Thanks, fixed.
Should “Lying” be italicized and not in quotes, since it’s a book?