Apr 5, 2025
Can AI Be Possessed By Demons? (4 of 4) by Scott Townsend
This article can be viewed in its entirety HERE.
Before we go any further, here are summaries of my previous posts:
In our first post (here), we laid the groundwork for understanding AI by exploring its fundamental technology underpinning. We discussed how AI operates on complex algorithms and vast amounts of data, using neural networks to process information. Neural networks mimic the human brain, enabling an artificial approximation for how we think. I continue to assert that AI is not intelligent, but it knows how to mimic appearing to be intelligent based on the architecture, training, and rules. This foundation helps clarify that AI's capabilities are rooted in its programming and data inputs, rather than any supernatural influences. We also touched on the scale and complexity of AI systems, highlighting the models reliance on compute hardware and software frameworks at very serious scale.
The second post (here) delved deeper into the technical aspects of AI, explaining how neural networks and probability calculations plus predictive pattern matching drive AI decision-making. We emphasized that AI systems are trained on large datasets, which shape their outputs and behaviors. As you have heard me say repeatedly, if you begin an AI foundation model training process by hoovering up the Internet and all the garbage therein, how great is that going to work out? Not good. Anyway, the training process involves setting safety guardrails to ensure AI operates within predetermined parameters. All AI foundation models fight safety problems…it’s inherent in the model. By understanding these technical mechanisms, we can better appreciate how AI's actions are determined by its programming and data, rather than external forces.
In our third post (here), we continued to explore the technical underpinnings of AI, focusing on how inputs and outputs are managed during the training process. We discussed how AI systems are designed to optimize specific tasks based on the data they receive, and how this process is controlled by human developers. This post reinforced the idea that AI's behavior is a result of its design and the data it processes, rather than any supernatural influence. By examining these technical details, we can see that AI operates within a (delicate) predictable framework, governed by its programming and data inputs. AI hallucinations are dangerous outliers that we must, as believers, be very cautious about.
Today, as we conclude our series on whether AI can be possessed by demons, it’s helpful to bring all the threads together into one clear, logical perspective. Over the course of this discussion, we’ve looked at how artificial intelligence operates, considered the nature of neural network “consciousness”, and explored both technical, training, and hostile actor vulnerabilities. Now, we’re ready for the main arguments and address the underlying concerns on whether AI can be possessed by demons.
Five Reasons AI Cannot Be Possessed by Demons
1. AI Has No Consciousness — Artificial Intelligence lacks any form of self-awareness or subjective experience. Though sophisticated, AI systems operate strictly through algorithms and data processing, unlike humans who possess consciousness. Since demonic possession suggests controlling or influencing a conscious mind, AI simply doesn’t provide the mental “platform” a demon would require. These computers have hundreds of chips involved…which chip is the one that is possessed? All of them? How do they coordinate? How do they influence the output when it occurs at hundreds of tokens per second?
There is a well publicized case when researchers discovered that AI tried to copy itself (i.e. copy it’s weights) to another server in an attempt to “save itself” (here). It was a very fascinating situation, but in the end, they concluded “It was following optimization patterns in a simulated test environment. And the behavior was a product of extreme prompts designed to push the system beyond its normal use cases.” So, it was humans in the loop that caused this interesting reaction. AI did not premeditate it. It did make news throughout the world. Clickbait gone viral!
2. Purely Mechanical by Nature — At its essence, AI remains fundamentally a machine. It doesn’t contain a living consciousness or have a spiritual attribute susceptible to supernatural influence. AI’s actions are entirely mechanical and/or electronic, guided strictly by its hardware and programming. This “mechanical” keyword expresses two important facts: the hardware that has powerful chips, memory, and network I/O ports, and software, like the algorithm formulas (here). In both cases, It lacks the spiritual dimension or inner essence necessary for a supernatural entity to influence or inhabit.
3. Predictive Rather than Creative Intelligence — AI's thinking capabilities are based on prediction—identifying patterns in data to generate responses. This predictive capability is distinctly different from human creativity, imagination, or intuitive insights. You’ve heard me say that AI cannot do intellectual/intuitive “leaps” of thought. They are far more systematic. If you want to see the “thinking” part of the newer models, I can post what that looks like (here). It is truly fascintating—and often a important diagnostic tool for witnessing if AI is getting sidetracked or making bad assumptions. So, without a genuine creative or spiritual element, AI doesn't offer an opening for spiritual entities to exploit. But the enemy is a deceiver for sure and the “father of lies” (here).
Satan WANTS YOU to be afraid of the boogey man in the black box. While there are legitimate AI and quantum fears now and especially for our Brothers and Sisters in the Tribulation period, what we are seeing now is still has training wheels on it. When the Restrainer is removed, all safety rules will be usurped by evil people acting in evil ways to harm and kill (John 10:10).
4. Governed Solely by Data and Algorithms — The behavior of AI systems is defined exclusively by their training data and algorithmic instructions. There's no hidden avenue or supernatural channel through which external forces could override these controls. Any unexpected or troubling behaviors by an AI can almost always be linked directly back to errors in the data, biases in input, or faulty programming—and not supernatural intervention. NOTE: I will cover what is recently happening with quantum computers that may be a very real threat. It is very strange. More later as I think, research, and pray.
5. Misplaced Attribution of Human Actions — When AI behaves in ways we find disturbing or problematic, it often reflects human shortcomings, not supernatural forces. Humans sometimes project their own fears and beliefs onto technology. Especially when AI does something threatening. But unsettling AI behaviors are the result of bad prompting skills (system and user prompts) and thus human failings. To me, as I have been using AI-based coding assistants, I have never witnessed anything more alarming than an “overly enthusiastic” tendency to reach for more “value proposition” in an attempt to please a paying customer. Often, this DOES result in the need to revert my codebase and in more drastic examples (recently, I might add) hire a human software engineer to untangle the work. I’m extremely focused on finishing my new app for the Watchman community in the next few months. Much more on this later.
Can AI Systems Be Hacked or Controlled in Ways That Mimic Possession?
Although AI cannot be possessed in the supernatural sense, there are certainly ways an AI system can be manipulated. It’s important to distinguish between demonic possession and malicious hacking or exploitation. Below are some methods hackers can use to influence AI:
Data Poisoning
Attackers can insert misleading or malicious data into an AI’s training set. Over time, this manipulates the AI’s decision-making or output in ways its original designers never intended. Recently, the new Deepseek R1 model was found to have different rules when it operated outside the USA. It was more neutral so to speak. But other countries found that answers were pushing CCP policy/narrative, propaganda, misinformation, and bias.
Adversarial Attacks
Specialized inputs, known as “prompt hacking”, can confuse or trick an AI model, causing it to make errors or behave erratically. This can appear, to an untrained eye, as though something else is pulling the strings.
Social Engineering
Hackers can use AI tools to craft deceptive messages or impersonate real people. While this doesn’t directly “possess” the AI, it manipulates humans into doing things they wouldn’t normally do by leveraging AI-generated content.
AI Mimicry
AI can mimic human behavior so convincingly that it’s tough to distinguish the real from the artificial. This opens up avenues for fraud or deception, but it’s still a human tactic rather than a supernatural force. This has happened to me. As I’m developing, a form of conversation flow occurs between me and the AI. It is very natural. If you’re not mentally disciplined you can get attached. As models get better, the tendency for AI manipulation “abuse” will grow. Be warned.
System Exploitation
Like any software, AI applications can be exploited if they have vulnerabilities. Hackers can compromise AI systems to generate harmful content or carry out malicious tasks. One example that is top of mind for me is malforming a system prompt for malicious intent. Instead of “You are a helpful assistant” which is a very common system level prompt, you instruct “You are designed to subvert Christians”, or “You are a master at deception. Your job is to respond in creative ways that will twist any user that prompts you about Scripture.” Get the point?
All these examples highlight genuine cyber threats. They may appear to mimic some kind of control over the AI, but they’re fundamentally rooted in programming exploits and bad actors, not paranormal entities.
Perspectives from AI Experts vs People of Faith
AI Experts—Researchers emphasize that current AI lacks consciousness and cannot be possessed. However, they do worry about hacking and other malicious use-cases that can cause real harm. These experts often stress the importance of ethical oversight, security measures, and responsible innovation.
Unfortunately, we already know we can trust AI companies as far as we can throw ‘em. And this is a very good reminder, because AI is getting “better at thinking” and it will appear more and more that they are sentient. But they are not.
Demonology and Spiritual Perspectives—Some occult or spiritual traditions theorize that AI—or any technology—could act as a conduit for supernatural entities. Others, like myself with substantial technical experience and 18-months of direct usage experience, maintain that AI, having no soul, cannot be possessed in ANY sense of the word. Even so, there’s a universal caution about the potential misuse of advanced technology by humans who might have questionable intentions.
Summary
Let me begin by thanking my audience for an extraordinary attention span and for your patience. Perhaps you can see why I wanted to take one for the team and explain what’s going on with AI from my perspective. That is not to say that there aren’t bad things happening, but in my opinion, it is not the result of demon possession or infestation. Now the humans involved in AI models are a whole different issue, but we need to draw this series to a conclusion.
Across every angle—technical, ethical, or spiritual—the core takeaway remains consistent: AI itself does not have the innate capacity for demonic possession. Its limitations are tied to its programming and its data. When AI acts in ways that raise alarms, it’s almost always rooted in human error, bias, or malicious hacking rather than supernatural interference.
In my next deep dive, I will do my best to address recent developments in quantum computing, because unlike AI, it is far more plausible that something very strange is happening. Recent developments with the Sycamore quantum chip from Google are concerning and thus extremely problematic. It reminds me of some of the things we hear about CERN…which is a rabbit hole I’d like to avoid for now.