I did 5 trials each, of random mashing (54-61%) and going to the rhythm of the music in my head (69-78%). Seems to imply that physical/digital output is reflective of some internal rhythm/pattern.
1) impulse - predictable by algo (60-75%)
2) self regulate - to stop yourself from doing what is intuitive (50ish%)
3) game the system - understand algo to some degree, predict it's guess, and do opposite (0-30%)
Edit:
Actually... I'm wondering if this could be used as a more easily accessible test of perseveration and set-shifting? The winsconsin card sort is time consuming, clunky, has inherent variance and strategic reasoning issues, and requires an interpreter etc. This simple program gives you a score immediately and the instructions are clearer -> less extraneous variables, greater focus on variable of interest, less resources required. This could be a real game changer!
I kept flipping a coin until I was bored of it. hovered around 50-55%. For a while it stayed at 100% and then suddenly plummeted, which was a bit weird, I guess
The categories make sense. Self supervision in this context isn't as good as taking advantage of the imperfect system. 50% for an ideal predictor would mean fully random, if a player manages to get to 0% they aren't really random, they are being completely unpredictable in contrast to random which is predictable half of the time.
I relied on my programmer's sense of randomness and got 35%, the approach that inputs can repeat, patterns cannot.
Given that random is 50% predictable a long sequence of unpredictable results would look the most out of place. Thinking about patterns that are less than 50% predictable, given ideal predictor, blows my mind and I'm not sure if I'm missing something obvious or I can't find information about that stuff.
It would have to have a minimal noise content and maximal structure or information.
If this was applied to a signal from space (or any message hidden in noise) and the observer was judging whether it was artificial I think I'd group it as follows:
Predictability of 0-49 % (especially 0-1%) now the initial assumption would be that the predictor is broken or the signal specifically avoids that type of detection, what if the predictor is perfect though? If the sequence is sufficiently short and never repeats it would be a message or at least structured information of some kind.
If the sequence would continue into infinity it would likely mean artificial and low noise, cannot be described by a function, model or probability distribution, has little to no redundant information as there is no error correcting pattern, if it was a message it would sort of be written in an infinite alphabet. The implications seem monumental, a structure so complex and noiseless that can't be modeled, has no pattern? All the methods of prediction and comparable sources of random signal find something more random than possible? Infinite meaningless data structure?
50 %, is random noise, natural and no message.
51-90ish %, mixture of artificial and natural sources, has repeating pattern, likely carries redundant information, the usual alien tv broadcast or something we emit into the void everyday, natural processes.
90-99 %, back to a mix of natural structures, stellar/other regular emissions, distorted communications
100%, a constant directed artificial energy, artificial dampening
I'm surprised by how engaging it is, kind of the simplest game. Strongly reminds me of the Divinity's Original Sin conflict resolution, where in co-op both players differing on their overall plot decision would have to play rock, paper, scissors to see which choice affects the world.
Interesting how rps was used to have players settle in-game moral/ethical disputes, sometimes over who gets to live.
I came up with a system that got me 24 places in without error... but then it falls apart and I don't know why (not an area I'm familiar with).
DDDDDD
F
DDDDD
FF
DDDD
FFF
DDD
EXPLODE
It's weird. It projects the representation of all previous strings of five, but then the very first prediction it makes goes against this pattern? And it continues to predict variation from previous strings up until a certain point... then doesn't?
It "remembers" how you gamed the system previously.
Hado posted while I was making this so I am going to share my findings...
There is an algorithm for repeats.
There is an algorithm for alternates.
There is an algorithm for which letter you "prefer", meaning which one you pick more often.
It has various "resets" that are layered for larger "chunks". I wouldn't be surprised if all it was was different "time" layers.
It basically goes through "checks" for itself in what it is recognizing as the "correct" guess or not on the length of the pattern.
Typically it has a "memory" of about 5-7 places, but as I said first, it can remember how you gamed the system before. I experimented with repeating the same letter for a fair bit and then switched, and then switched again, which gave me 2 reds for quite a while and I was getting like 6 in a row on a regular basis. It then it depends which letter you switch after a streak of the same letter, which is where what letter you prefer comes into play.
I kept it at about 52-53% for a while, but I think the odds are against you the longer you play. Basically, The longer you play, the better 50% looks, just like how the longer the fibb sequence goes on the more and more it looks like Phi (actually it might be cool if someone could try phi as a way to game the system but IDK if this can be done. I tried Fib in terms of repeating the same letter for the number represented as the next number in sequence to Fib and this is how I found out it has an algorithm for repeats built in.)
The difference between the human and the AI here is that the AI isn't trying to get "perfect" just better than you on average.
When I tried to be random, I scored in the 60-70% range. When I left it to intuition, I scored in the 50-58% range initially, then it suddenly dropped to 40-47%.
Kept at this for a while, and stayed pretty consistently at 40-45%.
I'm a bit weirded out now, need to think about this. There's got to be a psychological explanation for this somewhere.
A local radio station was giving the viewers a chance to win cash if they called in and were the first to guess their randomly selected number from 1-100. Using this on-off switch phenomena as an example, I told someone who was playing that their odds of guessing might increase slightly if they also used a random generator to pick the number, because people pick certain numbers more than others. Maybe the chances would be slightly even better if someone were to know which numbers are least picked.
Anyways, I played with this a little. I didn't see much sense trying to trick the game, because there was no easy way I could know what it's predicting.
I tried punching in all 32 combinations of the 5-grams, but got 57%. I was thinking it might have given it a harder time predicting, but I guess not.
Then I worked up an order of d's and f's that would produce a perfect 0 score:
This site uses cookies to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies. We have no personalisation nor analytics --- especially no Google.
.