PSI Test - Mark2

[MortFurd]

In this thread, (http://forums.randi.org/showthread.php?postid=5737710#post5737710) I posted a program to test psychic ability to influence a random number generator. There was some interest shown by other sceptics in the thread. As there was also a notable shortage of psychics participating in that thread, I am starting this thread to try and attract the attention of more psychics and also to get the attention of more sceptics.

I am looking for suggestions for improvement from both sides, as well as volunteers to actually put it to use.

The program as originally posted has some short comings.
Problems:
1. It hasn't been proven to be a fair random number generator.
2. It requires the user to manually start each test run.
3. It requires the user to manually tabulate the results.
4. Getting the result requires the user to read a value from the screen at the correct moment.

Problems 2 through 4 are separate from problem 1, but they all have the same solution. I will implement an "Autotest" function tonight that evaluates each test run and starts the next one and tabulates the results into an onscreen display (numeric and graphic).

This will solve the user interaction problems, but it will also allow the fairness question to be solved. Start the program, hit autotest, ignore the computer for a day or two, read the results. If the results are the expected 50/50 break down of up/down, then all is good.

Does anyone have any suggestions to make, critique of the whole idea, questions to ask?

[dafydd]

In this thread, (http://forums.randi.org/showthread.php?postid=5737710#post5737710) I posted a program to test psychic ability to influence a random number generator. There was some interest shown by other sceptics in the thread. As there was also a notable shortage of psychics participating in that thread, I am starting this thread to try and attract the attention of more psychics and also to get the attention of more sceptics.

I am looking for suggestions for improvement from both sides, as well as volunteers to actually put it to use.

The program as originally posted has some short comings.
Problems:
1. It hasn't been proven to be a fair random number generator.
2. It requires the user to manually start each test run.
3. It requires the user to manually tabulate the results.
4. Getting the result requires the user to read a value from the screen at the correct moment.

Problems 2 through 4 are separate from problem 1, but they all have the same solution. I will implement an "Autotest" function tonight that evaluates each test run and starts the next one and tabulates the results into an onscreen display (numeric and graphic).

This will solve the user interaction problems, but it will also allow the fairness question to be solved. Start the program, hit autotest, ignore the computer for a day or two, read the results. If the results are the expected 50/50 break down of up/down, then all is good.

Does anyone have any suggestions to make, critique of the whole idea, questions to ask?

The reason for the lack of psychics is easy to understand,there are no psychics.Problem solved!

[MortFurd]

As promised, the much improved PSI-Test Mark2.

17242

For anyone interested, you'll need to download the attached file and the appropriate PureData-extended package for your computer. PureData available here. (http://puredata.info/downloads)

Windows (http://downloads.sourceforge.net/pure-data/Pd-0.41.4-extended.exe)
Apple with Intel processor (http://downloads.sourceforge.net/pure-data/Pd-0.41.4-extended-macosx104-i386.dmg)

Mark2 has the planned autotest feature and some indicators. Nothing fancy, but it should do the job.

I am currently letting it run free to see if it is a fair RNG. I'll post a screen shot tomorrow of the results.

Here's a picture of it running:
http://forums.randi.org/imagehosting/thum_109764ba81b2a6516b.png (http://forums.randi.org/vbimghost.php?do=displayimg&imgid=19499)


On the readout:
Positive and negative should be clear enough. "Miracles" need some explanation. If you flip a coin, you get heads or tails - and once in a great while one lands on edge and stays there. In the PSI Tester, the equivalent is when the sum is exactly zero. This is very unlikely for a good many reasons, but it can happen. Since it is so unlikely, I've dubbed the occurance a "Miracle."

As always, feedback appreciated.

[Hokulele]

Cool! I will download this and give it a try from my home computer.

[MortFurd]

After about 5 hours of free running.
Total number of test runs: 313
Positives: 163
Negatives: 150
http://forums.randi.org/imagehosting/thum_109764ba862e868036.png (http://forums.randi.org/vbimghost.php?do=displayimg&imgid=19505)
While I was typing, it has gone to 318, 165, 153.

[MortFurd]

What I need now is more volunteers to run the program and see if it is a good, clean RNG. I also need assistance from someone who knows statistics who can tell me if the results are normal or skewed in some way.

To my eyes, the imbalance looks bad. If so, how bad is it? Are there methods in statistics to handle biases in random numbers? Say you know that the RNG produces X percent more positives than negatives over time. Can that be compensated when trying to determine if there was an external influence?

[MortFurd]

I've reworked the GUI and simplified it. I think I also have a handle on the uneven distribution. I seem to have introduced a bias by trying to eliminate any bias in the noise generation. Irony in action. :)

I'm going to let it run over night. If it looks good, then I'll post it tomorrow.

[Darat]

As promised, the much improved PSI-Test Mark2.

17242

For anyone interested, you'll need to download the attached file and the appropriate PureData-extended package for your computer. PureData available here. (http://puredata.info/downloads)

Windows (http://downloads.sourceforge.net/pure-data/Pd-0.41.4-extended.exe)
Apple with Intel processor (http://downloads.sourceforge.net/pure-data/Pd-0.41.4-extended-macosx104-i386.dmg)

Mark2 has the planned autotest feature and some indicators. Nothing fancy, but it should do the job.

I am currently letting it run free to see if it is a fair RNG. I'll post a screen shot tomorrow of the results.

Here's a picture of it running:
http://forums.randi.org/imagehosting/thum_109764ba81b2a6516b.png (http://forums.randi.org/vbimghost.php?do=displayimg&imgid=19499)


On the readout:
Positive and negative should be clear enough. "Miracles" need some explanation. If you flip a coin, you get heads or tails - and once in a great while one lands on edge and stays there. In the PSI Tester, the equivalent is when the sum is exactly zero. This is very unlikely for a good many reasons, but it can happen. Since it is so unlikely, I've dubbed the occurance a "Miracle."

As always, feedback appreciated.

Great - now I can be even lazier! Thanks.

[Hokulele]

I downloaded Mark 2, but I will wait for Mark 3 before playing with it with any seriousness.

[GenghisKhan]

I would love to test it out! (I just posted on the old thread so I will re-post here).

Question: will it work on a Mac?

ETA: Read your post more carefully, answered my own question. >.<

[Mirrorglass]

Let me compliment your work as well. This is a really cool program, and I'd love to see this test go better than the last one.

That being said, I haven't actually tried it yet. :p I'll try to get to it later this week.

[sadhatter]

I again suggest a twenty sided dice rolled on camera. It was the original random number generator, and if they can effect something like a computer program, there is no reason they cannot effect a dice. ( both would require known science to be suspended).

Heck, i will offer my time as the dice roller if need be.

[Mirrorglass]

Well, to be honest, if there's a test where the anti-psi of the skeptic is a reasonable excuse, I'd say a test where the skeptic is throwing the dice and the psychics are scattered around the world is one. Maybe that's why no believer has even acknowledged this idea, much less responded to it?

[sadhatter]

Well, to be honest, if there's a test where the anti-psi of the skeptic is a reasonable excuse, I'd say a test where the skeptic is throwing the dice and the psychics are scattered around the world is one. Maybe that's why no believer has even acknowledged this idea, much less responded to it?

Well then i am sure a believer could get some twenty sided dice ( they are less than a quarter a piece) and provide the rolling. Seeing as we have enough rpg people around here, cheating would be fairly easy to spot.

And on the subject of anti psi. You realize that if they fail ( which they will) it is going to come down to " well a skeptic programed it then!".

With the dice, and a believer rolling said dice in a manner that cheating could not occur ( it is damned near impossible to throw six sided dice in a way that will influence them, let alone a 20 sided dice..) , this would eliminate at least one excuse when the inevitable occurs.

But really, there is never, ever going to be a situation , online , testing in this manner, that is not going to give the psychics an excuse when they fail. I mean lets say we let a believer write the program, and it still fails. Well that person must have been a skeptic in actuality.

I applaud your efforts, but it is going to turn into a case of fighting the hydra. The people like rory, who just want to yank our chains, will talk in riddles and keep doing so for pages. The people like limbo, who may be legit , will keep complicating the testing protocol , not be available when the test needs to be done, and have at least one member in reserve on the excuse o matic for when failure occurs.

Honestly , until we get at least one honest, clear concise psychic believer, i don't think a test like this is going to go anywhere. It gives the trolls a chance to troll and waste people's time. And there are far too many excuses for failure to do any good to the legitimate believers when failure occurs.

[Mirrorglass]

Sadly this thread seems to be suffering from a severe lack of woos. At this rate, our experiment is going to end up proving a random number generator generates random numbers :p.

I like the dice idea, and would like to do it. But like you said, the psychics won't do it. Or at least, Limbo won't do it, or any other examination not exactly like what he wants. There don't seem to be any other believers available either. Still, I think there's some merit in trying; at least we're setting a precedent (if not the first one) for times when people claim they can do something paranormal. And it'll be interesting to point new arrivals to this test when they make their claims. Who knows, it might even convert someone.

As for the trolls, well, if I'm spending my time in threads about the paranormal, I don't think I can really blame the trolls for wasting it. ;)

[MortFurd]

Sadly this thread seems to be suffering from a severe lack of woos. At this rate, our experiment is going to end up proving a random number generator generates random numbers :p.

Snip..

Right now I'd be happy to prove the !"§§§$%§%/ thing is a fair RNG. I'm up to Mark5, and it still seems to be slightly lop sided.

[Hokulele]

Right now I'd be happy to prove the !"§§§$%§%/ thing is a fair RNG. I'm up to Mark5, and it still seems to be slightly lop sided.


Whoops, sorry, that must be my psi. I'll try to tone it down a bit. ;)

[Mirrorglass]

Maybe you have hidden psychic abilities?

What kind of tests are you using to decide whether it's lopsided? I can't claim to be an expert on this stuff, but a random number generator provides random numbers, so it's to be expected that they don't behave exactly as expected. The numbers should average out at the middle on a long enough time scale, but there's no guarantee they will during one hour, or even a few days.

[MortFurd]

Maybe you have hidden psychic abilities?

What kind of tests are you using to decide whether it's lopsided? I can't claim to be an expert on this stuff, but a random number generator provides random numbers, so it's to be expected that they don't behave exactly as expected. The numbers should average out at the middle on a long enough time scale, but there's no guarantee they will during one hour, or even a few days.
I'm using the program's own "autotest" feature. It picks 600 random ups and downs, one every 100 milliseconds and generates a running average. When it hits 600, the average is taken as the result. If the average was above 0, one point for Up, if it is below zero, one point for Down. It then resets the counter and clears the average and starts again. It tabulates the total number of test runs and the number of ups and downs.

The longest period I've let it run was over 10 hours. That was 682 test runs, 322 Ups and 360 downs. That looks far too one sided to me. I keep hoping a statistician will wander in and eyeball the numbers, or maybe tell me that it can be used any way as long as we do a long run and determine which way it is biased and by how much. No statisticians as of yet.

Maybe it is normal for the numbers to be almost 3 percent above/below average. Idunno.

[sadhatter]

Right now I'd be happy to prove the !"§§§$%§%/ thing is a fair RNG. I'm up to Mark5, and it still seems to be slightly lop sided.

K.I.S.S While i respect the hell out of you being able to program ( something i gave up at roughly the time duke nukem was going 3d) a 2 sided dice would work just as well ( if we are still going the ones and zeros route) ( look in your pocket or couch cushions if you have never seen a two sided dice.). And you wouldn't be wasting your time for some wooster to say " well that is cool but your positronic vibrations have made it an inherently skeptical program. "

[sadhatter]

I'm using the program's own "autotest" feature. It picks 600 random ups and downs, one every 100 milliseconds and generates a running average. When it hits 600, the average is taken as the result. If the average was above 0, one point for Up, if it is below zero, one point for Down. It then resets the counter and clears the average and starts again. It tabulates the total number of test runs and the number of ups and downs.

The longest period I've let it run was over 10 hours. That was 682 test runs, 322 Ups and 360 downs. That looks far too one sided to me. I keep hoping a statistician will wander in and eyeball the numbers, or maybe tell me that it can be used any way as long as we do a long run and determine which way it is biased and by how much. No statisticians as of yet.

Maybe it is normal for the numbers to be almost 3 percent above/below average. Idunno.

Wait a tick.

Okay, so the woosters say they can effect a random number generator. Well wouldn't it be easier to set up a biased generator ( lets say 99.99999999999 4' s or all 23's with the ability to make a random other number.) and then see if they can effect it? The gist of thier ability seems to be effecting a random number generator, so wouldn't this be the more definitive test? There would be little room for argument if in 20 or 30 runs with this type of RNG you saw the number they were going for 30 numbers in a row.

And it would save you the programing trouble of making a really random RNG.

[MortFurd]

K.I.S.S While i respect the hell out of you being able to program ( something i gave up at roughly the time duke nukem was going 3d) a 2 sided dice would work just as well ( if we are still going the ones and zeros route) ( look in your pocket or couch cushions if you have never seen a two sided dice.). And you wouldn't be wasting your time for some wooster to say " well that is cool but your positronic vibrations have made it an inherently skeptical program. "
My bold. While I actually can program and do so for a living, I'm not sure this falls under the general description. In PureData, you draw a diagram of what you want done. There's a bunch of different function blocks, and you place them and wire them like you would some electronic parts. Far different from what I would do in a normal language.

PD is really a trip to work in. It is normally running and you have to make a real effort to make it NOT run, so most of the time when I work with it it is like working on a TV with the power on - except for the not getting zapped or fried part if you screw up.

Anyway, the claim was that the psychics could mangle an electronic RNG over the internet. Preparatory to getting organized with a couple of competing teams and setting up a web site, I thought it might be good to see if the folks can actually influence an RNG. I know I wouldn't want to something like that publically if I hadn't tried it privately first.

[MortFurd]

Hoping that randomly selected sites on the web are accurate:
If N is the number of trials and we have two equally probable states for a result, then one standard deviation is:
sqrt(N)/2
So, for the above mentioned numbers:
sqrt(682)/2=13.06

Would therefore expect to see 341+-13 as results. That would be between 354 and 328. Actual numbers were 360 and 322. That is more than one standard deviation, but less than the two standard deviations. Since you would only expect to get within one standard deviation about 67 percent of the time, the values aren't that far out of whack.

Anybody know more about this stuff than my random googling?

[Christian Klippel]

Right now I'd be happy to prove the !"§§§$%§%/ thing is a fair RNG. I'm up to Mark5, and it still seems to be slightly lop sided.

Since you are already using PureData (hey, you might want to join #dataflow on freenode some day...), you can use the audio input. Take a microphone input (or line input + mic preamp), connect a diode to the mic input, crank up the gain. The thermal noise of the diode will make a nice RNG.

Greetings,

Chris

[MortFurd]

Since you are already using PureData (hey, you might want to join #dataflow on freenode some day...), you can use the audio input. Take a microphone input (or line input + mic preamp), connect a diode to the mic input, crank up the gain. The thermal noise of the diode will make a nice RNG.

Greetings,

Chris
Very good idea, and if I were trying to build one for my own use I might have actually gone that way.

This thing needs to be downloadable and usable by any random psychic. If they gotta bust out the soldering iron, then I don't think it'll fly.

[MortFurd]

In the end, KISS wins. I've got two separate testers running on separate machines. Both have outputs within one standard deviation, one more ups and one more downs.

I was trying WAY to hard to ensure a fair output. Turns out, the noise is fair all by itself and all my efforts did was to add a bias.

As a bonus, the Mark 5 uses 0.1 percent CPU time as opposed to ~25 percent for everything up to Mark4.

There's still some details on the GUI to clean up, then I'll post the finished product this evening. Time to go to work now. :)

Screenshots:
http://forums.randi.org/imagehosting/thum_109764bab098506d4c.png (http://forums.randi.org/vbimghost.php?do=displayimg&imgid=19521)

http://forums.randi.org/imagehosting/thum_109764bab099d8ea07.png (http://forums.randi.org/vbimghost.php?do=displayimg&imgid=19522)

[MortFurd]

This is it. Mark5. 17275
I trust it as fair to the best of my ability to tell, and it is fairly easy to operate.

Operating instructions:

Set the microphone input level for your operating system all the way down.
Unplug microphone.
Start PureData.
Turn "Calculate audio" off
Load the PSI-Tester
Set "Calculate audio" on
Click "Auto-Test" on the PSI-Tester
If you get a constant stream of zero, you may need to connect a microphone and try again. If you then get non-zero results, consider getting a plug that fits the microphone jack but which doesn't have a microphone attached. Some sound cards may switch to digital mode when there is no microphone plugged in.


Baseline

Start the tester as described above.
Let run for 2 two hours (120 Total_Tests)
At the end of the 120 runs, "Positives" and "Negatives" should both be between 50 and 70.
If they are significantly different, try again. (Two hour run)
If they are still outside the above boundaries, your sound card may be picking up periodic interference from somewhere in your computer.
In this case, I wouldn't trust the PSI-Tester. Try a different computer.
If the results are within the bounds, make a screen shot and keep a copy with the test results you are going to generate.


PSI-Test

Decide BEFORE you start how many test runs (Total Tests) you will make.
Decide BEFORE you start which direction you will try to force the results.
Decide BEFORE you start how many times you will repeat the entire sequence.
Make your test and try ONLY to force in the chosen direction and make all repeats as planned.
Note the results of each run along with the direction you were trying to force.
I suggest making as screen shot, printing it out and taking notes in the margin.
Repeat the PSI-Test process, but force in the OPPOSITE direction.


By the time someone has completed all of the above, maybe one of the resident statistics masters will have taken mercy on me and will explain how to calculate what is a significant variation and what is not.

[not daSkeptic]

Unplug microphone.

On some systems this may result in no noise. The input on my Mac Mini, for instance, is a dual-purpose optical/electrical jack. It has been my experience that the lack of any electrical connection causes it to run in optical (digital) mode, and the lack of a signal in that mode is interpreted as a solid stream of zeroes. However, I don't know the details of your particular software -- it may have some means of forcing the jack into electrical (analog) mode.

[MortFurd]

The operating method in the background has been greatly simplified comapred to the first versions.

Operating principle:

Every 100mSeconds (in the default 60 second test) one sample is taken from the sound card.
The sign of the sample is taken and added to the average for the current run.
Repeat sample and average 600 times, updating the scrolling display each time with the new average.
At number 600, see if the average is above, below, or zero.
Update the totals
Reset the average.
Repeat until stopped.


I've found that the running average is within 1 standard deviation of 300, and that about 1 time 100 runs the average will be zero.

At any rate, is there a statistician in the house?

[MortFurd]

On some systems this may result in no noise. The input on my Mac Mini, for instance, is a dual-purpose optical/electrical jack. It has been my experience that the lack of any electrical connection causes it to run in optical (digital) mode, and the lack of a signal in that mode is interpreted as a solid stream of zeroes. However, I don't know the details of your particular software -- it may have some means of forcing the jack into electrical (analog) mode.
Urk. In that case you'll have to plug in the mic and hope that turning the gain all the way down is protection enough against sounds from the room. Alternatively, you could clip the plug off of a broken mic or earphone and plug it in.

I didn't consider such dual use jacks, and PureData hasn't got a way to control it anyway.

Thanks for the tip.

[MortFurd]

I found this site. (http://graphpad.com/quickcalcs/chisquared1.cfm) It tells me that the test runs of 600 points vary significantly from being perfectly random.

Typical data:
325 ups, 275 downs

P value and statistical significance:
Chi squared equals 4.167 with 1 degrees of freedom.
The two-tailed P value equals 0.0412
By conventional criteria, this difference is considered to be statistically significant.


However, it also tells me that the long term results ARE fair.
Current results of the long term test:
2559 test runs, 1301 ups, 1258 downs.

P value and statistical significance:
Chi squared equals 0.723 with 1 degrees of freedom.
The two-tailed P value equals 0.3953
By conventional criteria, this difference is considered to be not statistically significant.

So, the noise from the sound card is of pretty crappy quality. It isn't, however, biased in any particular direction.

Basically, the noise produces "streaks" where either ups or downs dominate but in the long run the streaks are evenly distributed.

[not daSkeptic]

How long do these streaks tend to last?

[MortFurd]

I've watched the displays for a while, and started and stopped test runs.

I've seen it do three of each direction (three up, three down) in a row, then go to flip flopping back and forth. I've also seen one side get 10 ahead of the other, and then slowly level out with not quite even flip flops.

Any tests done would naturally have to be repeated, and done with the user trying to make the average go in the opposite direction.

I can fumble around and get stuff off of websites, but I'm not a statistician and never will be. Some one who's good at that stuff would have to determine how many test runs and test sets would be needed - and how many times it would have to be repeated to be significant.

I'm just a programmer who took on the intellectual challenge of putting together a tool to test a claim of pschic powers.

[Mirrorglass]

I think this is an excellent tool for testing the kind of claim Limbo made in the previous thread. Obviously, a single run on this proves nothing, but if one can do ten runs that each deviate upwards, that would seem to indicate psychic powers.

Of course, since the tool works the way it should, it's unlikely we'll ever get a psychic to try it, much less post their results. Still, the next time someone makes a claim of MPK, it'll be fun to have a place to point them to.

Great work, MortFurd.

[BobR]

The reason for the lack of psychics is easy to understand,there are no psychics.Problem solved!dafydd, this appears to be a thread for real skeptics, not trolls. I realize that sometimes it's easy to confuse the two, but trust me, this forum has plenty of outlets for your sentiments on other threads. Knock yourself out.

[BobR]

I've just run into this thread, and I think the open-mindedness and spirit of investigation is commendable. My idea about psi ability is that either everyone has it or nobody does. Might want to check out this site: http://www.random.org/, which exposes a TRNG (based on atmospheric noise) as a web service.

[BobR]

Wait a tick.

Okay, so the woosters say they can effect a random number generator. Well wouldn't it be easier to set up a biased generator ( lets say 99.99999999999 4' s or all 23's with the ability to make a random other number.) and then see if they can effect it? The gist of thier ability seems to be effecting a random number generator, so wouldn't this be the more definitive test? There would be little room for argument if in 20 or 30 runs with this type of RNG you saw the number they were going for 30 numbers in a row.

And it would save you the programing trouble of making a really random RNG.
I like what you are saying, sadhatter, except with respect to this analogy. Suppose we are trying to determine whether it is possible to lift a one ounce object off of a table. If we provide a one ton object and it can be lifted, we have definitively shown that a one ounce object can be. However, if we are unable to lift the one ton object, we haven't determined anything about the one ounce object one way or the other. Thus we are not really providing an experiment that focuses on the hypothesis. Maybe psi ability is something that can be developed from something weak to something strong, as the "woosters" also claim.

[Dr H]

Wait a tick.

Okay, so the woosters say they can effect a random number generator. Well wouldn't it be easier to set up a biased generator ( lets say 99.99999999999 4' s or all 23's with the ability to make a random other number.) and then see if they can effect it? The gist of thier ability seems to be effecting a random number generator, so wouldn't this be the more definitive test? There would be little room for argument if in 20 or 30 runs with this type of RNG you saw the number they were going for 30 numbers in a row.

And it would save you the programing trouble of making a really random RNG.

I'm intrigued by this idea and haven't really seen it considered elsewhere.

With the RNG, presumably anything which is going to affect the randomness of the output is going to have to influence some part of the generation or the outputting of the RNG. Problem is, if someone claims an effect that's not very far above expected noise effects, there is going to be endless statistical quibbling.

But suppose you set up an algorithm -- as elaborate as you like -- that takes a seed number for input and always generates the same output number from that seed input (seed could be a single number from an RNG). Then you set it up to loop some arbitrary but large number of times. You would not see the output, or know what the output number is, until after the run is complete, at which time you print out the results.

What you would expect to see is a single number, repeated N times. If the alleged PSI "force" were able to act on any part of the process, you would instead see (at least) two different numbers, N, and M, output. With a reasonably well-shielded computer the amount of noise you'd expect to see from a process like this would be extremely low, and a sudden string of (say) "4's" in a list of five million "23's" would stand out like a sore thumb. The process would be essentially infinitely replicable, and statistical analysis of the results would be fairly simple and straight-forward.

As I say, I've never seen this particular method described before, but it's simple enough that it seems unlikely that someone hasn't given it a try. Anyone know if someone has?

[MortFurd]

snip...

But suppose you set up an algorithm -- as elaborate as you like -- that takes a seed number for input and always generates the same output number from that seed input (seed could be a single number from an RNG). Then you set it up to loop some arbitrary but large number of times. You would not see the output, or know what the output number is, until after the run is complete, at which time you print out the results.

What you would expect to see is a single number, repeated N times. If the alleged PSI "force" were able to act on any part of the process, you would instead see (at least) two different numbers, N, and M, output. With a reasonably well-shielded computer the amount of noise you'd expect to see from a process like this would be extremely low, and a sudden string of (say) "4's" in a list of five million "23's" would stand out like a sore thumb. The process would be essentially infinitely replicable, and statistical analysis of the results would be fairly simple and straight-forward.

snip


That type of thing exists in just about every programming language I know (above assembler, anyway.) The thing is that the algorithms are deterministic - the program merely rearranges the bits according to a pattern. Putting in the same seed will always yield the same output. If it doesn't, your computer is so broken that the program wouldn't run at all.

What you are proposing is no less than that the pschic should make the computer calculate wrong or store the data wrong. Either of those two implies that the CPU or the RAM can be influenced by external events. It also assumes you can influnce the RAM or the CPU to the necessary extent so precisely that everything else functions normally.

To do that you'd have to be able to either make the CPU calculate wrong only at the exact instant that it is processing your numbers (at over 1Ghz - **Snork**) or that you can make the exact bits in your RAM work wrong - good luck, there's over 16 Billion bits in your Ram (for 2GB of RAM.) I would have a devil of a time determining which bits belong to a particular variable, if I could do it at all. And you'd have to change the values at the right time - if you change it just before the CPU stores the latest calculated value, the your psychically changed value will get overwritten.

All of that is the reason for trying to make a true RNG from existing hardware. Trying to change the output of a PRNG (as you suggest) is an excercise in futility. A hardware RNG at least gives you the theoretical possibility of doing something. All you have to do is induce noise in a single place, and that spot will not affect the normal function of the system but will affect the output. That's what this PSI test program does. All you have to do is to induce a little bit of biased noise on the line in or mic line of your sound card. Surely that is easier than the complicated machinations needed to manipulate a properly functioning CPU and RAM.

[Robo Sapien]

I'm not 100% sure on this, but I don't think computers can actually produce true random numbers. My understanding is that most programming languages include methods to create seemingly random numbers by using the current system time (in milliseconds) as a variable in a really complex formula. So it is highly improbable, but not impossible, to replicate a "randomized" result.

[MortFurd]

I'm not 100% sure on this, but I don't think computers can actually produce true random numbers. My understanding is that most programming languages include methods to create seemingly random numbers by using the current system time (in milliseconds) as a variable in a really complex formula. So it is highly improbable, but not impossible, to replicate a "randomized" result.
None that I know of do it that way.

Some seed the random number generator for the time when you start your program, some the first time you use the PRNG. On all of them that I know of, you can seed the generator yourself and thus force a particular sequence to be produced.