Experimental data to support simulation hypothesis

There exists (probably) a correlation between random evens and there is data directly showing that. An exploration and data analysis of 78840000000 coin flips.

Published in

Becoming Human: Artificial Intelligence Magazine

8 min readSep 12, 2021

Preface for the sane reader

I am aware that the data anomalies that I will present here have a range of more pragmatic explanations in contrast to being proof to the simulation hypothesis.

There could a bug in the code — that’s why I am attaching all the code, intermediate data, and links to raw data and kindly asking to review it.
There could be a mathematical/statistical error in data analysis. Statistics is a tricky science and this code uses it a lot. So, if you are able to point out the errors — please do so.
There could be an undiscovered issue with data generation/data collection processes, faulty devices, gamma rays hitting the RAM, or sophons messing up with the data. We are looking at extremely small statistical deviations here.
At the end of the day — even if the data and analysis are valid, there are hundreds of other fringe theories that could be used to explain observed phenomena.

But, I choose the simulation hypothesis to explain everything, because of YOLO, that’s why. Not like a have an academic title or respect to lose. So, grab a coffee, I will show you some magic.

The Princeton Noosphere experiment

a.k.a. Global Consciousness Project.

This is a hard territory to navigate as it is borderline with all possible fringe science and outright quackery. And maybe it still holds many secrets because a proper scientific community is afraid to touch it.

This research originated in the (in)famous US Government “Stargate Project” experiment. That’s right, it is an official fact — there was a USA-funded project dedicated to the research of paranormal and extrasensory abilities in sleep-deprived, LSD-high, ex-hippies US Army soldiers. The telekinesis was researched in one of the branches, and I can imagine that after the constant failures to bend a spoon by tripping soldiers, scientists have started to look for less sturdy objects to bend.

Random Number Generators

The less sturdy object they found is dead simple: take a truly random number generator (based on a physical process of quantum nature — tunnel effect in diode or any thermal noise) that generates 0s and 1s with a 50/50 probability and ask an “operator” to temporarily change the distribution of those probabilities by the power of magic/thought/concentration/midichlorians by a statistically significant number.

This idea alone was very fruitful. Hundreds, if not thousands of papers were published in academic journals, passing the review barrier by obscuring telekinesis with “micro-PK phenomena”. Some claimed total success, some — failure to reproduce, and the official scientific consensus reached at the end of the 90s.

No micro-PK exists, all successful experiments are bad design, selective reporting, wrong statistical analysis, or fraud. You can all go home.

The Global Consciousness Project experiment

Global Consciousness Project (GCP) is the last large-scale experiment with random number generators and micro-PK. It is actually still ongoing — the website has all the details and raw data and results, and I urge you to take a look. But TLDR is following:

They placed 10 to 50 (active in different years) quantum random number generators across the globe. Each generator produces 100 random 0s or 1s per second and streams the data to the central database, where it is stored. The devices are built in a way that mathematically guarantees exactly 0.5 probability (with XOR filtering) of 0s and 1s.

The GCP hypothesis is following:

Great event synchronizes the feelings of millions of people, our network of RNGs becomes subtly structured. We calculate one in a trillion odds that the effect is due to chance. The evidence suggests an emerging noosphere or the unifying field of consciousness described by sages in all cultures.

GCP results and its criticism

The main result and claim of the GCP project fit into one picture.

The read of this chart is the following: GCP authors “cut out” 500+ fragments of data corresponding in time to “worldwide global events”. For each event, they calculate the cumulative Z-score metric. Mathematically it is a number of standard deviations from the mean and intuitively you can read it like “how probable was the deviation from 50% expected proportion of 0s and 1s”. It turns out, if you aggregate the data from all events — highly, highly, highly improbable (1*1⁰¹³). According to GCP authors, this proves their hypothesis of global consciousness. And also it’s pretty easy for independent researchers to replicate — the raw data is all there available.

Issues with the experimental data

A non-trivial number of reproduction attempts were made by independent institutions. The TLDR:

The data generation, collection, and processing don’t seem to have issues. Some RNGs tend to break and go offline, some could produce spikes of highly improbable sequences in a short time due to power spikes, some have a bias in standard deviations (coming from XOR), but all of that is bad data (rotten eggs on slang) is cleared in preprocessing stage leaving us this data that has reasonable behavior and total Z score, well in the margin of “acceptable”.
The weakest point is of this project is the arbitrary selection of “important” events and their timespan. For example, the very strong correlation of RNGs observed during the 9/11 attack can be fully diminished by slightly changing the analysis timeframe (link). This is a very strong signal, that observed results can be explained by the author’s unconscious bias and unaccounted multiple hypothesis testing corrections.

My idea and data analysis

The GCP hypothesis has inverted the chicken and egg. It’s not the events that cause RNGs correlations, but the subtle changes in RNGs behavior could cause some events to happen, but not necessarily “global and important”. Anyways, I will leave my ungrounded hypothesis for later, let me show you the data.

Hypothesis: there exists a correlation between the Z-score of independent RNGs.

Imagine an alien civilization, deciding to run a simulation to study the properties of life. Even having extreme compute powers, it would make sense to limit the computation to a certain “space” region, cut it off from the rest of the observable universe with a fundamental limit (speed of light), and ALSO to kindly push the evolution into more “interesting” and less probable state, where life actually arises from proto-mess (quite low probability event). Taking into account that our macroscopic world operates on a basis of well-explored laws and we see no obvious violations of them, the only place where an “external” force can influence our world would be quantum level happenings. And this influence could be as gentle as changing the probability of a single quantum event, that will propagate through the universe as a butterfly effect. But here comes the physical reality — single wave collapse has a pretty high chance to go unnoticed and be consumed by surrounding thermal noise. So, the “magic” touch should be slightly more strong.

What GCP did — they just cherry-picked such moments.

My hypothesis: there exists a correlation between independent RNGs.

This is the main claim of this analysis. I have just repeated the GCP analysis but removed the key weak point — the arbitrary selection of time slices for analysis. This analysis just runs over all of 2014 (randomly chosen year) with a varied sliding window.

Here I show you the only important chart for this claim.

What does this mean?

First, for every round of analysis, I randomly split all RNGs into non-intersecting subsets, having data from different RNGs. Let’s name them “test” and “control” for clarity.

Then I scan the all-year data with a sliding window of a 2 seconds (which was shown to have the strongest effect in preliminary exploratory data analysis) over the “test” subset and calculate the probability of each observation. Then, I select time episodes where probability was below a certain threshold (title of the chart) and calculate the probability of state that we observe in the “control” stream. And then the same is repeated, but switching the “test” and “control”.

For sanity check every time I also calculate the probability of the state of the second stream not in the moments when the first stream is breaking the probability threshold, but in the randomly chosen moments.

The whole process is repeated 30 times and, by applying the central limit theorem, the standard deviation of each observation and p95 confidence interval is calculated.

So, the legend for each subplot bar is following:

1: Probability of independent “control” stream in moments when “test” stream in breaking a threshold. (Direct Test)

2: Probability of independent “control” stream in random moments (Direct Test Random)

3: Probability of independent “test” stream in moments when “control” stream in breaking a threshold. (Inverted mean)

4: Probability of independent “control” stream in random moments. (Inverted Random)

Conclusion.

Under all sanity checks, all bars show exhibit random and fully independent behavior, while leading to a mean probability of 0.5 with respect to confidence intervals. 0.25 and 0.1 thresholds are clearly showing breaking this rule with p<0.05 and others are showing directional results.

All this goes unaccounted to multiple testing hypotheses and there is always a chance that it’s a coincidence — like everything else in our world.

But, I want to believe, that it is not.

Code for analysis — link.