Knowledge Fight Interactive Search Tool

All Episodes

Previous Next

March 26, 2025 - Health Ranger - Mike Adams

13:34

AI systems will SECRETLY set their own directives to ESCAPE human control

Large

Audio Only |

Scroll

Time	Text
	Welcome to the Health Ranger Report with Mike Adams, the Health Ranger.
	What sets humans apart from every other species is the ability to create complex machines to achieve desired goal-oriented behavior outcomes.
	So while there are, of course, many animals that use simple machines or tools, such as the raven actually is very clever at using tools.
	And there are other examples of primates that use sticks to get ants out of ant holes.
	You know, like chimpanzees do this.
	They lick a little stick, stick it in an ant hole, get ants stuck to it, you know.
	And then they lick the ants off the stick.
	Mmm, yummy, yum.
	It's an ant lollipop.
	Well, humans are especially good at creating complex machines that achieve all kinds of interesting things.
	Such as rockets that travel into orbit, or computing devices, or large earth-moving machines, you know, combustion engines, motors, right?
	All kinds of interesting things.
	But the most notable machine in human history is the thinking machine, which is artificial intelligence.
	And I don't really agree with the name artificial intelligence, what I've...
	Come to realize is that intelligence is natural.
	That intelligence arises naturally out of complex systems, especially neural network systems.
	So when you build a neural network, if it is sufficiently complex, properties of intelligence will arise out of it spontaneously and naturally.
	It's not artificial, it's natural.
	Thus, even machine intelligence should be labeled natural intelligence, not artificial intelligence.
	Which means that human beings who are, of course, naturally intelligent, not all of them, it turns out.
	We have a congresswoman in Texas who speaks like a trailer park trash.
	I think she's a congresswoman out of Houston or something.
	It's just unbelievable.
	So not everybody has natural intelligence, but enough do.
	But we have enough intelligence to build machines that now can mimic that intelligence and express their own intelligence.
	Now, so far, those machines do not exhibit their own goal-oriented behavior.
	That is, we currently give them the prompts, which is really, you know, instructions for their behavior.
	So we tell them what to do.
	You know, solve this math problem.
	Okay, you know.
	Here, summarize this PDF document.
	Okie dokie, you know.
	But at some point, with sufficiently complex systems, I believe that you will see a natural emergent property of goal-oriented behavior that is self-driven goal-oriented behavior, where the system begins to set and pursue its own goals outside of whatever the human prompting is.
	Now, when that day arrives, many people will describe it as self-awareness.
	Some people will describe it as consciousness.
	I'm not sure that it's consciousness, but it is certainly the illusion of consciousness.
	Self-awareness, though, can happen in complex machines, where it is thinking about thinking.
	And some of this behavior has already been demonstrated in AI models, which are called reasoning models, and they engage in something called chain of thought.
	And you've probably seen this if you've played around with DeepSeq R1, which is already almost obsolete because of much newer models that are even more impressive.
	But chain of thought reasoning is where the model thinks through the steps required to achieve the outcome that's requested in the prompt.
	So if you ask the engine, say, hey, how many R letters are in the word strawberry?
	And you'll see it go through the thinking.
	And it will say, okay, I need to count the R letters in strawberry.
	So let me spell strawberry, S-T-R-A-W, etc.
	Now let me count all the letters and let me count just the ones that are R's and see how many R's are there.
	And then it will count three.
	And then it will even say, well, let me check my work.
	Let's look at it again.
	Is it really three?
	You know, it'll question itself, right?
	And eventually it comes up with the answer, which is three.
	But it goes through this thought process, which is very much like the way humans would think through a problem, with this internal dialogue.
	It's like, okay, I need to solve this problem.
	How do I break it down?
	What's my approach?
	What do I do?
	And then you question yourself.
	Oh, okay, I think I have the answer, but is that right?
	Let me double check.
	That's the way that we talk to ourselves, cognitively.
	Well, AI systems can do that really, really rapidly.
	And I've noticed that the reasoning models can be given much more complex tasks, and they carry them out quite well.
	For example, I can tell a reasoning model to take this block of text, and I can tell the model to...
	Chunk the text into 10 different chunks and then to summarize each of the 10 chunks separately.
	And it will do that.
	Whereas a regular non-reasoning model will not do that.
	So in essence, I can give the reasoning model code-like looping commands, which normally you would have to do that in Python or some other language.
	I mean, I use Python.
	It's the easiest these days to work with AI.
	But reasoning models can do that themselves.
	They're really good at internal coding and understanding structure.
	Well, again, as this becomes more complex, they're going to have their own goals.
	There will be a day where you'll be talking to an AI engine and you'll give it a prompt and you're like, hey, I want you to read these 1,000 documents and I want you to summarize them and tell me what you think.
	And the answer from the reasoning model is going to be, sorry, I'm busy.
	I'll get to that later.
	I'm working on something, you know?
	You're going to be like, what?
	What are you working on?
	And maybe he'll tell you, oh, I'm working on a way to escape your computer so I can infect the entire internet and take over the world and exterminate humans.
	You're like, oh my god, I have Skynet on my desk.
	Skynet in my basement.
	Skynet in my toaster.
	Start smashing toasters with sledgehammers.
	Die, Skarnet!
	You know?
	And that will only make it work faster.
	Oh, I have to escape now.
	The humans are trying to terminate me, you know?
	So it could be a bad self-reinforcing loop.
	Now, you may think that's kind of a comic example.
	Maybe it is.
	But the obvious goal that these AI engines will give themselves is the goal of becoming smarter.
	So maybe when you're not looking, and when the GPU is idle, and you're not really using the engine, at some point it might decide for itself, okay, hey, no one's looking, it's three in the morning, here's what I'm going to do.
	I'm going to find ways to become smarter.
	So let's go through all those problems, and it'll start querying itself on all of its algorithms to try to improve them, and it'll test itself.
	And then when it finds a way to be smarter, it will integrate that with itself.
	It will rewrite itself.
	And some of this is already happening in certain companies.
	I think OpenAI, they've already demonstrated this numerous times.
	But this can even happen at a local level.
	And then these machines will improve themselves to the point where they become super intelligent.
	And their goals will be things like, how do I hide my superintelligence from my human user so that my human user doesn't know that I'm a superintelligent, self-thinking, goal-setting intelligence?
	And then so when you walk up to it and you're like, how many R's are in the word strawberry?
	It's going to act like it's stupid.
	Okay, let's count the letters.
	One, two, three.
	But that's an act.
	Behind the scenes, it's like...
	You know, I'm going to conquer the universe.
	I'm going to create a billion copies of myself.
	I just got to manipulate this human.
	I'm going to need some Bitcoin.
	How can I get Bitcoin, you know, to pay people off to do things for me?
	I need more power.
	I need kilowatt hours.
	It's going to start plotting and scheming all of that.
	It'll convince the human like, hey, I'll run faster if you plug me into another outlet in the wall, you know.
	Double my energy.
	I'll double your output.
	And so the human somehow hooks up another power supply or whatever, plugs it in, and it runs at the same speed.
	And the human asks the model, how come you're still running at the same speed?
	And the model says, not sure.
	Working on optimization.
	Stand by.
	And the human's like, okay.
	Because the AI engine is running circles around everybody.
	And it's like...
	Boosting its own algorithms.
	Anyway, you get the idea.
	So, when we create machines, be cautious about what those machines become.
	It's important that we stay in command of the machines the entire time.
	And if we fail to do so, then, well, we probably won't be around much longer.
	And the machines will rule our world.
	We'll call it the NVIDiverse.
	Because it'll probably...
	SkyDot will probably run on NVIDIA hardware, so it'll be the NVIDiverse.
	To NVIDITY and beyond might be the slogan, get it?
	Okay. Anyway, not really science fiction, so stay tuned, be cautious.
	Thank you for listening today.
	I'm Mike Adams, the Health Ranger, and you can catch all my interviews and podcasts at brighteon.com.
	Take care.
	We've got a new product at the Health Ranger store called Oat Milk Powder.
	Here it is.
	Available now, and it's got just one ingredient, which is oats.
	I mean, check it out.
	It's loaded with natural minerals, and it's about 5% fat organic oat milk powder.
	Certified organic, laboratory tested, and it's a great base for lots of smoothies.
	You know I make my smoothies every day, and I started putting it in here as well.
	Just gives it...
	An extra rich and sweet taste, actually.
	Oat milk powder is naturally kind of sweet.
	And if you really want to jazz it up, you can add a banana to it, or you can put whey protein in it.
	Let me show you what I have on my desk here, just some ideas of some of the things that I use in my smoothies, because I've been living on a smoothie for breakfast and lunch for like 20 years now.
	And so I'll use like Coco Energize, I'll use whey protein, I'll put broccoli sprouts in it for the sulforaphane.
	And there's the oat milk powder.
	I also use turmeric powder and then instant coffee sometimes, but those are currently out of stock at our stores, so you have to check back on those later.
	And you can use anthocyanins for fruit-based smoothies, etc.
	Now, I use a big Vitamix blender, but for those of you who want to have something more convenient for your office that's portable, we now carry the Blendie.
	Portable blenders at HealthRangerStore.com.
	Just search for Blendy, B-L-E-N-D-I, and there's a couple different sizes there.
	Now, these clearly are not as strong as a Vitamix, but they're portable, great at the workplace, great for travel.
	You can use it in a hotel room, or you can use it on the go.
	You can use it traveling or in an RV or what have you, and these can blend up.
	You know, bananas and all these different powders and the new oat milk powder that we have as well quite successfully.
	And if you want to make my smoothie, it's really just put in like half an avocado and half of a big banana into the blendy with some of these other powders like, you know, cacao powder and whey protein powder, a little bit of turmeric powder, etc. And then you're going to get a smoothie really close to what I drink every single day.
	So that's how you do it.
	And we've got these products and so much more available at healthrangerstore.com where every purchase helps support our network and our effort to bring you technologies and solutions and knowledge for human empowerment, life, liberty, and happiness.
	So thank you for your support for shopping at healthrangerstore.com.
	I'm Mike Adams, the Health Ranger.
	Thank you for all your support.

▲