|
On August 29 2014 05:54 LSB wrote:Show nested quote +On August 29 2014 05:42 Acrofales wrote:On August 29 2014 05:35 LSB wrote:On August 29 2014 05:31 Acrofales wrote:On August 29 2014 05:13 The_Templar wrote:On August 29 2014 04:38 ComaDose wrote: how much someone knows about statistics and random number generation would also affect how well they made a random string of numbers so it would vary greatly change from person to person.
can you tell us what your point was and what the answer is if there is one? my answer is that it could be either we don't know. The point I made is that, in isolation, both are far more likely to be human generated, and there was therefore no way to actually tell. Nobody agreed with me, and everyone found it obvious that the second one was computer generated and not the first. Of course this was correct. I don't think you phrased that properly, because I don't really see why either of the strings is "far more likely" to be generated by a human than by a computer. I do agree that the underlying assumptions for stating the second one is computer-generated are tenuous... and a better argument is that in isolation it is not easy to state which is which. As LSB's math above shows, a computer will only generate a similarly lopsided string in 3% of the cases, so it's not exactly a "typical" outcome for a random string generator either. @LSB: you have to make some assumptions. Otherwise all you're saying is that a string similar to the bottom one is less likely to be generated by a computer than the top one, in which you are throwing away the information that you know the other one is generated by a human... and it's not so that we know absolutely nothing about humans and therefore should simply assign to them the one that is less likely to be generated by a computer. Just because you have data doesn't mean you have or should incorporate in it a model. In fact, in this case incorporating the data would induce a huge amount of error, rather than simplify it. I disagree. As long as you do it in a principled manner. I think I could make a fairly simple Bayesian classifier that does better than random at predicting human strings looking at "longest string of subsequent digits" as one of the features. Perhaps "deviation from the expected number of 1s" is another one, although I have no evidence to back the second one up. This is the fatal trap I which I am pointing out that you are falling into. You have three assumptions 1) Computer behaves a certain way 2) A typical human behaves a certain way 3) The specific human who picked the number sequence behaves like a typical human I make one. See the difference?
No, you make 2. The first and the last. You just say that your specific human picking the sequence, instead of behaving like a typical human, behaves like ANYTHING that isn't a computer, and therefore the string least likely to be generated by a computer is the most likely to be generated by a human. The likelihood of that assumption being true is rather low: it is far more likely that a specific human behaves like a typical human. We can then devise experiments to figure out how a typical human behaves, and presto, we have a scientific approach!
|
On August 29 2014 03:52 LSB wrote:Show nested quote +On August 29 2014 00:18 The_Templar wrote: There was an argument in my information and coding class today about two binomial strings, where I was the only person who thought my point was valid at all. 1010101001 0001110101 0110100110 1001010100 1001001101 1000111010 0111101101 1110111111 1011001111 1100010110
Which of these is randomly generated, and which of these was created by a human?
For a serious answer. Assumption #1: One of the strings is Human Generated, One of the Strings is Computer Generated Assumption #2: The computer picks 0 and 1 at true random. String 1 Has 24 Ones, this seems to be the one most likely to be generated by a random number generator String 2 Has 33 Ones The chance of observing 33 or more successes in 50 trials is 1.64%, double this if you want to include the chance of 17 or less heads for 3.28% which is less than the 5% value typically used for "statistical significance" Thus it is far more likely the first is randomly generated. My statistics is rusty so correct me if I'm wrong plox.
I don't think this is right. Both strings are single draws from the complete set of all possible strings. Both are equally likely to come out of a random number generator. The number of ones is only relevant the other way around: "if I draw a string with a random number generator how much ones is it likely to have?"
What you do here is similar to having 3 stones in a bag, labeled 1,2,3 and after being presented with 2 stones labeled 2 and 3 you are asked: "which of these comes from the bag and which is made by us?" You say: "well in the bag are 2 stones with odd numbers so the one from the bag is more likely to be odd, so the 3 probably comes from the bag and the 2 is handmade."
edit: for clarity, you can for example also say in this situation "well in the bag are 2 stones smaller than 3, so the one from the bag is mor likely to be smaller than 3, so the 2 probably comes form the bag" and then you get a contradiction
From this also follows that you can't use the fact that one of them is given by RNG so you have to use assumptions about humans, which I'm staying out of.
|
On August 29 2014 02:39 The_Templar wrote: There are only two strings, I just happened to divide them into groups of ten In that case, there are two conflicting analyses that point to both strings being unlikely outcomes of a process that churns out 50 1s or 0s with equal probability, which then agrees with your earlier point that both strings are poor examples.
Was this in a class about human bias in what a random process should look like (underestimate frequency and length of runs)?
Edit:
Actually, I disagree with LSB's analysis because by his assumption #1, we have to include conditional probabilities and Bayesian analysis i.e. whether a high number of 1s with a high number of runs is less likely to be generated by a computer or by a human which is a silly path to take.
The question is flawed and whoever proposed it originally was not careful enough to make up a contrived example.
|
i stand by my assumption that if you consider both options bit by bit they are equally as likely to be generated by a computer which makes the question unanswerable. to assume the one with more consecutive bits is false makes an assumption about the human that is not disclosed.
|
your Country52796 Posts
On August 29 2014 09:56 xes wrote:
Was this in a class about human bias in what a random process should look like (underestimate frequency and length of runs)? It was about human bias in picking random numbers. I argued that a human that might just alternate 1s and 0s could just as easily put more ones than zeros.
|
Why aren't phone cameras built into phones sideways so that idiots who record with their phone vertically have normal video?
|
i agree that default position should be widescreen
|
On August 29 2014 10:19 ComaDose wrote: i stand by my assumption that if you consider both options bit by bit they are equally as likely to be generated by a computer which makes the question unanswerable. to assume the one with more consecutive bits is false makes an assumption about the human that is not disclosed. I think given the context of the class you can make the assumption / empirical evidence on human bias in picking "random" numbers.
Both string 1 and string 2 are rare samples from the pool of 50-length encodes of 50 consecutive Bernoulli trials. String 2 is more rare by around an order of magnitude, comparing the unlikelihood of small runs with the unlikelihood of high 1s.
So P(String 1 generated by computer) > P(String 2 generated by computer)
But you know that whatever string wasn't computer generated is human generated, so we are actually comparing
P(String 1 generated by computer|String 2 generated by human) to P(String 2 generated by computer|String 1 generated by human)
To figure out this, you have to make an assumption about human bias because you need to know P(String 1 generated by human) and P(String 2 generated by human).
In addition, because there is empirical evidence that human generated "binomial" data tends to minimize runs and keep total number of successes and failures equal, this effectively explains away (by virtue of the conditional) the fact that String 2 has a statistically improbably number of 1s in the context of the question.
|
On August 29 2014 11:35 xes wrote:Show nested quote +On August 29 2014 10:19 ComaDose wrote: i stand by my assumption that if you consider both options bit by bit they are equally as likely to be generated by a computer which makes the question unanswerable. to assume the one with more consecutive bits is false makes an assumption about the human that is not disclosed. I think given the context of the class you can make the assumption / empirical evidence on human bias in picking "random" numbers. Both string 1 and string 2 are rare samples from the pool of 50-length encodes of 50 consecutive Bernoulli trials. String 2 is more rare by around an order of magnitude, comparing the unlikelihood of small runs with the unlikelihood of high 1s. So P(String 1 generated by computer) > P(String 2 generated by computer) But you know that whatever string wasn't computer generated is human generated, so we are actually comparing P(String 1 generated by computer|String 2 generated by human) to P(String 2 generated by computer|String 1 generated by human) To figure out this, you have to make an assumption about human bias because you need to know P(String 1 generated by human) and P(String 2 generated by human). In addition, because there is empirical evidence that human generated "binomial" data tends to minimize runs and keep total number of successes and failures equal, this effectively explains away (by virtue of the conditional) the fact that String 2 has a statistically improbably number of 1s in the context of the question.
But is that really a fair assumption to make?
We teach humans that repeating patterns are not random and then say that humans are unable to do random numbers because they favor non-repeating patterns. Its a tainted assumption only true because we teach the subjects to do so. Isn't that more a cultural bias than a "human" bias?
|
On August 29 2014 11:59 Thieving Magpie wrote:Show nested quote +On August 29 2014 11:35 xes wrote:On August 29 2014 10:19 ComaDose wrote: i stand by my assumption that if you consider both options bit by bit they are equally as likely to be generated by a computer which makes the question unanswerable. to assume the one with more consecutive bits is false makes an assumption about the human that is not disclosed. I think given the context of the class you can make the assumption / empirical evidence on human bias in picking "random" numbers. Both string 1 and string 2 are rare samples from the pool of 50-length encodes of 50 consecutive Bernoulli trials. String 2 is more rare by around an order of magnitude, comparing the unlikelihood of small runs with the unlikelihood of high 1s. So P(String 1 generated by computer) > P(String 2 generated by computer) But you know that whatever string wasn't computer generated is human generated, so we are actually comparing P(String 1 generated by computer|String 2 generated by human) to P(String 2 generated by computer|String 1 generated by human) To figure out this, you have to make an assumption about human bias because you need to know P(String 1 generated by human) and P(String 2 generated by human). In addition, because there is empirical evidence that human generated "binomial" data tends to minimize runs and keep total number of successes and failures equal, this effectively explains away (by virtue of the conditional) the fact that String 2 has a statistically improbably number of 1s in the context of the question. But is that really a fair assumption to make? We teach humans that repeating patterns are not random and then say that humans are unable to do random numbers because they favor non-repeating patterns. Its a tainted assumption only true because we teach the subjects to do so. Isn't that more a cultural bias than a "human" bias?
I actually doubt it's taught. Pattern recognition is a very very basic skill for us. But if you feel like doing a PhD, you can try doing anthropological fieldwork to test the hypothesis that the human bias in random generation is actually a cultural thing.
|
On August 29 2014 11:59 Thieving Magpie wrote: But is that really a fair assumption to make?
We teach humans that repeating patterns are not random and then say that humans are unable to do random numbers because they favor non-repeating patterns. Its a tainted assumption only true because we teach the subjects to do so. Isn't that more a cultural bias than a "human" bias? Sure, one argument is that for all behavior studies "human" really refers to "western (particularly American) undergraduate/university students."
But I think pattern recognition is a distinct part of human cognition, and indeed animal cognition as well (most studies on birds and rats).
Yet as another discussion point though, humans (i.e. American university students, presumably also Caucasian male) seem to do worse at understanding probability than animals, presumably because our believes interfere with our ability to impartially react to empirical data. A pretty hilarious (although small sample size) study is where pidgeons perform better than humans at optimal strategies in the Monty Hall Problem http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3086893/
Finally, I posit that it is impossible to extract this randomness bias from what is learned vs innate. If you went to some random tribal island and asked "give me a sequence of 50 1s and 0s and try to make it random" the very question already assumes the cultural context of what we (in the Eurocentric definition) consider random.
|
Canada11355 Posts
How come liquibets are so slow to update?
|
i just saw 10.10.10.10 in the beginning of the first string and instantly thought that was in no way made randomly by a human. one just doesn't, unless on purpose.
i'd even guess that the next 2 numbers in the first string would be 1 1
|
What the hell was the characters name that saved Nikolai Rostov's life in his first battle against the French in War and Peace? Nikolai brought him home with him during his leave from the army.
Seriously this is driving me mad.
|
Why do we have this thread when Google already exists?
|
On September 03 2014 03:41 Jett.Jack.Alvir wrote: Why do we have this thread when Google already exists?
#powertotheplebs #liquidgoogle
|
I’m looking to join a christian religion, but there are so many different denominations to choose from. Which one’s the correct one to believe in so that God will let me into Heaven?
|
To answer that question, you will need a competent spirit medium or classical necromancer. Then, just contact enough spirits of each denomination so you can make a statistically significant call as to which gives you the highest chance to get into heaven. To try to reduce selection bias etc, you will need to be very careful when selecting which spirits to talk to.
|
On September 04 2014 00:27 Epishade wrote: I’m looking to join a christian religion, but there are so many different denominations to choose from. Which one’s the correct one to believe in so that God will let me into Heaven? how did you pick that god out of the millions you could have chosen from too?
|
On September 04 2014 01:03 ComaDose wrote:Show nested quote +On September 04 2014 00:27 Epishade wrote: I’m looking to join a christian religion, but there are so many different denominations to choose from. Which one’s the correct one to believe in so that God will let me into Heaven? how did you pick that god out of the millions you could have chosen from too? Random hat drawing!
|
|
|
|