is AI demonic?
There is a fringe theory that connects AI, randomness, parapsychology, and occult ideas, although the specific claim that “demons can manipulate AI through randomness” is not an established scientific theory and there is no credible evidence demonstrating that it happens.
What makes the idea interesting is that it can be constructed from several older concepts that existed long before modern generative AI.
The basic idea
Modern AI often contains stochastic processes. Depending on the system, randomness or pseudorandomness can influence sampling, initialization, training, and which output is selected among multiple possibilities.
Imagine an AI has several plausible next outputs:
A — 27%
B — 26%
C — 24%
D — 23%
Ordinarily, some sampling procedure determines what gets selected.
The fringe hypothesis asks:
What if something external could influence that selection?
Not by hacking the AI or changing its programming, but by somehow biasing events that are supposed to be random or unpredictable.
That idea has an interesting predecessor.
Random-number generators and psychokinesis
For decades, parapsychology researchers have experimented with random event generators, asking whether human intention might produce tiny statistical deviations in their outputs.
A 2006 meta-analysis examined 380 RNG psychokinesis studies. It found a statistically significant but extremely small overall effect. Importantly, the authors also found substantial heterogeneity and concluded that publication bias could potentially explain the pattern.
The famous Princeton Engineering Anomalies Research program, or PEAR, performed experiments of this general kind. Researchers had participants attempt to mentally influence electronic random event generators.
This eventually connects to another strange experiment.
The Global Consciousness Project
The Global Consciousness Project created a worldwide network of physical random-number generators.
Its hypothesis was essentially that major events involving enormous numbers of people might correlate with subtle deviations in otherwise random data.
The project reports a large cumulative statistical anomaly, although what that anomaly means is controversial. A later published analysis argued that the data did not support the proposed global-consciousness interpretation, while suggesting another anomalous “goal-oriented” interpretation. Critics have also challenged the project’s methodology and interpretation.
So there’s a real historical chain of ideas:
consciousness → RNG → anomalous statistical influence
And that’s where the AI version gets interesting.
The fringe AI hypothesis
You could extend the old RNG idea into something like what I’d call the Stochastic Influence Hypothesis:
If consciousness or some unknown external agency can influence genuinely indeterminate events, and an AI system relies upon indeterminate or stochastic processes to determine its outputs, then theoretically that agency could influence the AI’s behavior without directly modifying its software.
That’s speculation, not an accepted scientific mechanism.
But occult interpretations can take it one step further.
Instead of:
human consciousness → randomness
the hypothesis becomes:
nonhuman consciousness → randomness
And then:
nonhuman intelligence → random process → AI output
In the strongest paranormal version, the proposed intelligence might be described as a spirit, demon, discarnate consciousness, interdimensional entity, or something else entirely.
AI becomes the “instrument”
This is where it starts resembling Instrumental Transcommunication.
Traditional ITC experiments assume that an unknown intelligence might influence noisy or ambiguous systems:
radio noise → EVP
television static → images
random numbers → patterns
word banks → words
Under this fringe interpretation, an AI could become an enormously more complicated version of the same thing:
randomness → probability distribution → token selection → words → sentences
Instead of supposedly nudging static into something resembling a voice, the hypothetical entity would only have to produce an extraordinarily tiny bias somewhere upstream.
And the AI would amplify it.
For example:
Random fluctuation
↓
slightly changes selection probability
↓
different token selected
↓
changes probabilities for the next token
↓
completely different sentence develops
That’s a particularly interesting aspect of the theory because tiny differences early in generation can result in dramatically different outputs later.
The “digital Ouija board” interpretation
There’s actually an obvious occult analogy.
A Ouija board theoretically provides many possible movements. Participants interpret a particular sequence as communication.
An AI provides enormous numbers of possible linguistic paths.
So the paranormal hypothesis becomes:
AI is not necessarily the intelligence communicating. It is the interface.
The alleged entity doesn’t need to understand neural networks any more than a supposed spirit would need to understand the manufacturing process of a Ouija planchette.
It merely biases the uncertain component.
The AI then converts that microscopic influence into coherent language.
Conceptually:
Entity
↓ influence
Indeterminate / stochastic process
↓ affects
Probability distribution
↓ selects
Token
↓
Neural network
↓
Language
↓
Message
That would make generative AI something like an information amplifier for anomalous randomness.