Slashdot Mirror
Random Number Generator That Sees Into the Future
hackajar writes "Red Nova news has an interesting article about a random number generating black box that may be able to see into the future. From the article: "according to a growing band of top scientists, this box has quite extraordinary powers. It is, they claim, the 'eye' of a machine that appears capable of peering into the future and predicting major world events"."
If it can sense future events, that would make it less random, right? To me, that almost sounds like pre-determined events (how far into the future this pre-determination is good for, you decide), so it really isn't "random".
"If you were plowing a field, which would you rather use? Two strong oxen or 1024 chickens?" --Seymour Cray
/. needs a "trivially debunked hogwash" category. This belongs with the "battery stickers" story from a few weeks ago.
-jcr
The only title of honor that a tyrant can grant is "Enemy of the State."
Believing in superstitious quackery like this black box has serious ramifications.
Perhaps you've heard of the scientific method?
It sounds like quakery, but so did flight and travel to the moon 150 years ago.
The appropriate stance is "I'll believe it when they prove it", not "that can't be true." Rabid atheism is no more scientific than wicca.
Red Nova usually has good articles, but every once in awhile, one shows up that belies evidence of lack of scientific rigour. This is the case here.
An example (from the article:)
It was a preposterous idea at the time. The results, however, were stunning and have never been satisfactorily explained.
This sentence is prejudicial because it biases the results as being "stunning", without describing who finds the results stunning.
"Never satisfactorily explained" also presumes that someone finds it worthy of needing explanation.
Again and again, entirely ordinary people proved that their minds could influence the machine and produce significant fluctuations on the graph, 'forcing it' to produce unequal numbers of 'heads' or 'tails'.
"Proved"? Pretty strong words with no supporting detail. Once I read sentences like this, I discount an article as being scientifically unfounded.
In response to the parent post:
No, the laws of chance do not say any such thing. In fact, the laws of chance say exactly the opposite.
I believe you're misinterpreting the laws of chance.
If you have two choices chosen at random over a series (a 1 and a 0; or heads and tails on a coin), there is a high probability that one of the choices will be chosen a significantly higher number of times than the other.
Significant as a percentage? Unlikely.
Over time, the percentage disparity will decrease to near zero, but the total numerical disparity is likely to increase.
This is a trivial statement. If n flips has m total disparities, n+x flips will have between m and m+x disparities. It is therefore impossible for the total number of disparities to decrease, and almost guaranteed that it will increase.
The only significant measure of disparity is that of percentile disparity. And if you measure percentile disparity on a scale equivalent to the number of events being measured, it will in fact appear to be a nearly flat line on the graph.
The thing that bothers me about this "experiment" is that it presumes to assert that people can control a machine that generates random events, without describing the algorithm by which those random events are produced. Trying to simulate true randomness (indeed, what is random?) is a huge topic within math, statistics, and computer science; yet, it's not mentioned once within the article.
Classic case of weak correlation.
Worse, the correlation suggests the causation post-facto. Nobody even guesses there will be a correlation until there's an effect. And if there's no effect, nobody discounts the box's output.
Sad. Innumerate. Stupid.
The only question is if they actually have the data to back it up (some graphs would be nice).
I would like detailed instructions on how to construct a stream of random numbers with behaviors that correlate to outside events as they describe, so that I can repeat their experiments myself and see if I can reproduce the same effect. Tabletop reproduction isn't always possible in science (e.g. historical sciences like archaeology, paleontology, cosmology- remember that, you creationists) but in this case reproduction of results should be easy. (If this were real.)
At the very least I want to know how to generate a stream of random numbers that reproduces this effect, how to recognize a prediction when it arrives in the stream, and how to assign a P-value for associations between random stream events and real world events. Unless we move past the sort of ex post facto "predictions" of past events, there is nothing new here. It looks like a repetition of work already done by Nostradamus.
Try reading the papers of the people running the damn research program. They do have several websites on it.
You are passing judgement on their work only on the basis of a PopSCI level article written for a 9th grade audiance.
Most, if not all, of your issues are addresed on the project sites.
Except they shouldn't be straight lines at all. They should take random directions all the time. Sometimes even very big ones. A flat line is one of the least random things produced in the world. If GC existed according to how REGs worked, atomic clocks would randomly lose percision around major events.
I mean a scientist is quoted as saying "Our data shows clearly that the chances of getting these results by fluke are one million to one against." I would actually place the chance much much lower, I mean a million to one is nothing really. The odds of 30 coin flips in any order is a million to one. The real problem is prediction. The question is whether the model can predict into the future what events will cause blips and the magnitude of the event.
I'd do something interesting, but my server can't handle a slashdotting.
I'd like to see how many "special" sequences they had which were NOT followed by an event they deemed special.
"Long run is a misleading guide to current affairs. In the long run we are all dead." (John Maynard Keynes)
No, the laws of chance do not say any such thing. In fact, the laws of chance say exactly the opposite. If you have two choices chosen at random over a series (a 1 and a 0; or heads and tails on a coin), there is a high probability that one of the choices will be chosen a significantly higher number of times than the other. Over time, the percentage disparity will decrease to near zero, but the total numerical disparity is likely to increase.
The article may very well be about pseudo-science. However trying to counter it with pseudo-reasoning and confusing distinct, well-defined statistical properties doesn't advance the cause of science. In fact it looks not only bad but desperate.
My professor in statisitics would probably have pitched an eraser at you for suggesting what amounts to an oxymoronic "high probability of the improbable." If the probability is 1:1,000,000, then in one million experiments there is a finite probability (1:1,000,000) that you may see the event once, and a lesser finite probability you would see it more than once. If something improbable turns up "significantly" as you phrase it then you check to see if the dice are honest.
In fact, the mean value of a normally distributed series of random numbers should trend toward a constant value. In the case of runs of 0s and 1s, it should trend toward 0.5 and approximate it more closely as the experiment runs.
The variance should tend to increase as less probable values fill the wings of the bell curve. The longer the series of random values the more nearly normal that trend should be and the greater the potential variance may be, since with a longer experiment you can actually acquire less probable runs that simply could not occur earlier. For instance you need to toss a coin a minimum of 20 times to have even the possibility of achieving 1:1,000,000 odds (1:1048576, actually 2^20). You would need to toss a good many more times than that before you could legitmately begin to worry of a 1:1,000,000 occurence did not show up.
That's how Los Vegas makes a living. The rubes always hope that the improbable will kiss them on the neck. In fact nothing you say actually contradicts the quote you are trying to criticize. They are discussing the mean results while you are talking about the variance.
The simplest explanation for their "correlation" is simply coincidence of highly improbable runs temporarily skewed the data. Remember that the experiment has been running for years so some really improbable runs are possible. They need a lot more disasters before they can actually test an argument based on a statistical improbability.
------ The only greater hazard to your liberty than n politicians is n+1 politicians.
5 parts in 10000 is nothing. The probability theory guarantees that there are many experiements where such results are randomly produced. It's the same as with stock market. Many people use various insane trading schemes. Some of these randomly get rich. Those that consistently get rich claim that their schemes work, which is, of course, bogus.
Technical analysis is the same. 250 people go to a seminar, half of them decide to get an account, buy books and software and start trading on FOREX or commodity markets. After a year 64 of them are still in the black. After two years 30 of them have profits for two years. After six years there probably will still be 2 guys, with BMWs, Rolexes and stuff. Wait one more year - one of them will lose anything, but the lucky one will decide to give seminars on technical analysis or write 100$ books on trading.
Those guys in Princeton are idiots. They are wasting their time and university's money. Their claims are ridiculous and they deserve to be fired and sent to work in the trash sorting plant. That way we can put their skills in finding valuable stuff in random shit to good use.
Future Wiki -- If you don't think about the future, you cannot have one.