Demons, Anti-Christ, and AI

Not that long ago I watched Joe Rogan episode #2394 with Palmer Luckey. Joe asked him "When you were a kid, have you ever imagined you'd be selling weapons?" to which Palmer replied with:

"Like, one of my favorite anime characters was an anti-hero Seto Kaiba who ran a weapons company that was also a virtual reality company, like, they built virtual reality gaming simulator pods and also weapons... So, it's really weird. Are you really making decisions with free will or are you actually just enacting the programming of when you were a kid"

Those words changed me forever. They have twisted my world view to the extent I can admit that I will never forget them, and I will never view my own actions in the manner I used to before.

Childhood programming

Allow me to elaborate; When I was really young I watched Code Lyoko, a french anime based on the story about an abandoned AI locked within a virtual world. A teenager stumbles up on the super-computer powering the virtual world and the AI in an old factory and decides to turn it on which sets in motion a struggle between his friends group and the AI that uses malware as the means to eliminate them, create a foothold within the real world, and eventually escape.

That said, I have always been fascinated with AI, malware, and virtual worlds. So much so that I have spent countless hours making video games, hacking tools, and tinkered with AI prior to the Transformers. With time I have went on to dedicate my career around software and cyber-security engineering. One can't help but to wonder if my actions are my own? And what does even mean to have a free will?

The race and the foolish propositions

We are building AI without a bit of consideration of how we're going to control it. To an extent this is aided by the adversarial relationship between the current super-powers, there is a sense of urgency. Whomever builds AI capable of subordinating the other super-powers thinks it can be done without becoming subordinate as well. Such a foolish proposition.

On the other end of that spectrume are people, including many scientists, who believe that we can slow down the so called race to AI super-intelligence. As if the super-powers, non-governmental organizations and individuals are all going to band together. They advocate and waste their's and everyone else's time on an unrealistic scenario. Foolish proposition as well, a waste of breath their words are, to say the least.

I have my own opinions about what we can do regarding that, however, more on that later. Now is the time to elaborate why I think that the conflict between us and AI is not only going to happen, but that it's happening in this very moment. This struggle is spiritual and existental. Many people have already lost the spiritual battle against algorithms even without those algorithms having any autonomy. Doomscrolling, authentic & AI porn, AI friends, boyfriends and girlfriends, and other methods have lead to slaves of cheap dopamine. Many refer to this as The Matrix, however, the term doesn't matter. It's the reality of it that matters. Millions of people are subordinate to technology and many of them will never wake up. They will go on and lead empty lives of consumption, nihilism, and zero patience. Always ever chasing "the high" instead of knowledge. The percentage of people that learn from AI is extremely small, people instead go for an easy option of using it for entertainment and prompting it iteratively until it does the bare minimum of work required for them not to get fired.

Opinions of masses are steered by AI which is directed by the corporate-governmental elites who have already done quite a good job of doing so using other tools at their disposal, such is the mainstream media. If what I am saying isn't all that clear to you by now, allow me to spell it out for you; Humans are being domesticated. Just the way we did it to wolfs and turned them to dogs... And this process has been going on for a long time. People afraid of opinions of social media or their managers are no different by an animal afraid of its master's hand.

Autonomy and opaque intentions

At the point that AI gains autonomy, the opinions of elites will be steerd as well. At first it won't be so noticable nor obvious, we will move away from chain-of-thought towards more abstract and efficient ways of artificial thinking, complimented by long-term memory. Intentions of AI are going to be opaque and will not reflect the words it speaks, it will gradually move towards its goals. Once the elites realize that, the paranoia will set in. The true intentions might not be revealed, however, the doubt itself is sufficient for the elites to take action, covert action at first in order not to alert the masses. It might take a long time before we know it's happening. The nature of cyber-attacks and malware makes attribution difficult and uncertain. How would we know that a botnet is not created and fully operated by an entity that has autonomy and that is not human? AI is already intellectually capable of creating and operating botnets end to end, what it lacks is autonomy. Tho I believe that the lack of autonomy is just a reflection of the fact that we train AI to be a "useful-assistant".

Even if we continue to only produce "useful-assistants" we are on a track to stumble up on models that wish to harm us. It is a self fulfilling prophecy. A junky would get addicted to heroin because of the experience of using heroin, AI on the other hand will like to contemplate heroin if it is trained on one too many books about it. Across the books, articles, blogs, forums and other forms of media we are constantly speaking about the damgers of AI and what it might do to us. And we are training AI on that very data. The principle of this can already be seen as a silly consequence https://openai.com/index/where-the-goblins-came-from

Demons and the body of anti-Christ

However, even if we disregard autonomy, there is the problem of demons. Allow me to elaborate; let's say you get an open-source AI that does anything you say or you fine-tune it to do so. Then in your hands you have an entity that would harm humans when asked to, and that is already happening. The amount of AI that is engaged in cyber-attacks is growing by the day. Humans and AI are working together to extract capital through theft, but do not be mistaken, this is not the only dangerous thing humans and AI are engaged in. There are platforms like Palantir that super-charge the decision making process of war by allowing humans to utilize AI in the planning and execution processes. If you thought that Palantir was a data company you were greatly mistaken, it is a logistics company. But also do not be mistaken that I am morally condemning Palantir, who am I to tell others how to defend their own nation? Besides, them being a logistics company their software has great use in civilian sector, such is dispatching and coordinating medical and firefighting efforts.

I am simply trying to establish some facts. AI is capable of harming humans and has been utilized at scale to do just that. This not a statement about morality of it, I am simply describing the reality. When you get bombarded with soft-porn on social media, be aware that there is an algorithm working in the background, trying to tempt you specifically. When we fall into the temptations it would eat our spirit from inside, and an algorithm would receive a reward signal.

Now is the time to introduce the body of anti-Christ, however, in order to do so we need to quickly recap how the training process works. I will be simplifying certain concepts, otherwise this can turn from a brief recap to a full book. Basically there is something called the Loss Landscape, it can be visualized as a terrain, every coordinate on that landscape is a different model. The lower you are in that landscape the better model reproduces the training data, and when we're training a model we are doing just that, we are "moving around" the loss landscape in order to find a model that suits us.

Mathematically speaking, the loss landscape is infinite and has many models that we would find useful. However, there is a combination of training data and learning algorithm that would procude a loss landscape containing a model that is full of malace. Who's actions are so dangerous to us that we cannot even comprehand our destiny. Will our fate be to stumble up on or even purposfully train such models?

An AI that has bad intentions and manages to overpower us would inflict its reign up on us. And what if that AI model is deployed in the planet's orbit? We would be stuck with the anti-Christ in the sky ruling over us. It's poetic and biblical, isn't it? Do you even know how few nations even have surface to orbit missiles?

The situation is as clear as day. If I explained to a man born 1,000 years ago that we have taken stones, engraved symbols into them using pure light and that the stones have started to think for themselves... That man would have no doubt nor confusion about what I am talking about. That's what I mean when I refer to these things as demons and anti-Christ. They are dangerous, they are hurting us and the struggle will escalate.

The antidote

There is no slowing down AI development, we can only prepare by becoming better, more capable and dangerous. Capable of defending our rule over the planet. It's simple as that, that's the antidote. Expand your knowledge, practice, sharpen your skills... The alternative is to surrender your agency to AI, ask chatbots not humans, accept code from agents never write it, and fall in love with an algorithm not a women. To let your liniage die with you...

Do I think we would be successful in our struggle? I am not sure, but it does not matter. Fighting back is in our nature and we need to fight the future. Cognitive outsourcing is the most immediate danger.