THERE IS NO KILL SWITCH: Can Humans Still Control AI?

0:00Every week the world changes, but the real story is happening, God decends. Eddie, welcome, ladies and gentlemen, man, I'm so good to be back. I cannot express to you exactly how excited I am to be back. It is Monday night. Let's go. Very... It's been crazy. It's been like what? A month, right? Just let me know if you can hear me and we are off and running because we got a lot to talk about tonight, man. Just a wild and crazy stuff going on in the world, to say the least. And I appreciate everybody so much. Let me know you can hear me and then we'll get going. John will be joining us. You can hear me. We are good to go. Real quick, who's watching? Appreciate everyone who tuned in tonight for sure. Julie, sorry to hear about your mom. And I hope to see you on Saturday. But take care of home. Definitely, Christine Jeffords always good to see you. Again, you and John and Valerie, Melvus explosion.

1:02That was awesome. Laura, good evening. Good to have you back. And it's good to be back on Monday nights for sure. I appreciate everybody who's tuning in. Kato Donald, appreciate having you in. It sure has been a while. Again, our Canadian contingency. Maybe Lola will show up or some other folks from Canada. We love our Canadian contingency. And once in a while we get some people from New Zealand, which is awesome. Australia. It's awesome. Let's see who else we got here. Don't miss any one, Valerie. Good to see you again. Margaret and Joe. Love you guys. Hope you're doing well. Let's see what else we got here. And some. Jeannie, good to have you a girl. Good to have you. Tell Woody. I said, hello. Hello. And Rebecca. God bless you. I can hear you. I have to see the notification came through for this live. Awesome. All right. Well, we are back and running up. And I'll say what? And all day, all week, all weekend, rather. Just working on, you know, the live, the weekly briefing script and everything else.

2:05Really exciting. And I had so many things I wanted to talk about. And next thing you know, it's like, well, I can't have a 10 hour episode. So we have a lot to cover tonight, though. On something that I just, I find hilarious. I'm going to have a 10 hour episode. So we have a lot to cover tonight, though. On something that I just, I find hilarious. And disappointing all at the same time. But we'll talk about it. Quick book update for some of you. Who are interested. Parts one and two are finished. I say finished. There's one or two points that I need to go back and clarify some things on. But it's written. Very excited about that. A number of people who have read it are enjoying it. So that's positive. I'm really excited. I'm really excited. I'm really excited. I'm really excited. I'm really excited. I'm really excited. I'm really excited. I'm writing part three right now. And it's basically written except one particular section that is giving me a little bit of fits. But it has to be right. So working on that now. Obviously, I've had to take some time away from it just because of travels and everything else going on. But I am back in the saddle now. And this week, definitely despite having three interviews this week, I am looking forward to just that.

3:09I'm looking forward to just that diving back into my book, which is going to be hopefully out by Christmas. Aren't the Seval bulls the title. And maybe if I have, if you haven't seen the cover, maybe next week I'll bring the cover and you can see it. So. But I am feeling positive about Christmas. But it has to be right. So last week on the Ryan Files, we had a pastri billet chrome wild child. And he likes to go there and he does. And I think it's interesting to have someone on who was a saint missed. He freely admits it and involved in a lot of weird stuff. His perspective has got to be unique when you think about where he comes from and what he's experienced in life and what changed him. And that was fascinating episode. So I appreciate pastri billy coming on. And of course, one of my favorite people in the world, Dr. Judd burden. Listen. Man, I know sometimes I'm a geek. But I've told Judd I am going to I don't want to get it until after I'm done writing this book because I know if I have the temptation to start going through his courses, his history courses, I'm going to do it.

4:18But I want to wait until after the book is written just because I it's too much of a temptation. And I can only imagine that it's going to be fascinating. I'm going to learn a lot. It's only $250 for an entire course. You've got American history and you've got world history with just a supernatural worldview. Add it into it. Things that you're not necessarily going to be taught that you should be taught, but you're not being taught that was taught at some time in the past too. Keep that in mind, but it's not being taught anymore. So yeah, you go back and watch that episode if you haven't and he's running sales from time to time too on the stuff. So check it out, Dr. Judd burden last week. So we know who's watching. Let's dive in. We got two topics tonight, one real brief. All this weekend, you'll throw trench number six came out. The new unidentified anomalous phenomenon files, including a material from the advanced aerospace weapons system applications program, OSAP, a WSAP is the name of that acronym for advanced aerospace weapons system applications programs.

5:33Basically, a major part of this too is the defense intelligence reference documents or durs DIRDS defense intelligence reference documents. You got to have a acronym for everything, right? So basically you get a bunch of geeks together and they're doing technical research in these papers or commission to examine advanced technologies that could matter to national defense. That's the big takeaway that I took away. There's some other stuff in there, but what was fascinating about this are these DIRDS, these defense intelligence reference documents. So what are they studying? And tell me if this doesn't sound like something out of a science fiction novel because it does. Except it's not. It's real. Advanced materials, you know, metallic glasses, programmable matter, metamaterial, superconductors, materials capable of surviving extreme aerospace environments.

6:37And there was specific papers on different topics. One was propulsion and gravity looking into anti-gravity, negative mass propulsion, positron propulsion, advanced nuclear propulsion, a neutron fusion that's fusion designed to produce a relatively low amount of neutrons. Magneto hydrodynamic propulsion or MHD uses electrically charged gases and magnetic fields to reflect impact airflow and potential potentially produce propulsion that way. There's another paper about space and dimensional physics. We're looking at warp drives science fiction, right? You know, Star Trek. No. No, we're looking at this now. Dark energy, the manipulation of extra dimensions. We're looking at that right now.

7:38Traversible wormholes, you know, things popping in and out of existence and then traveling, great distances. Negative energy. And then there's another paper talking about cloaking and detection and visibility and cloaking, hypersonic tracking, high energy electromagnetic systems. There was one talking about quantum technology, which is very important and relevant to a lot of things today because that gets into quantum computing, quantum communication, superpositioning and whatnot like that, quantum vacuum energy, negative energy states. So why does all this matter? Okay, all these big terms is like, okay, Ryan, what are you doing? Well, none of that by itself proves somebody recovered a crashed craft. But one of the documents goes much further. It says information from injuries involving anomalous aerospace systems might allow investigators to reverse engineer physical characteristics of advanced aerospace systems of unknown province.

8:52So basically you have human beings who have been injured in some way that stood too close to a propulsion of magnetic field or whatever humans sustained energy. They're not saying that there's any UFOs that they've been in contact with or around, but they are saying that they have humans who have been affected negatively their health. So they want to reverse engineer, well, how did that happen? What would this, whatever have to be in order to have these kinds of physical ailments? All right. And later another document says data exists to reverse engineer propulsion systems of anomalous aerospace vehicles independent of origin. So let me dumb it down for you. You don't necessarily need to possess the engine to reverse engineer something. If you think about it, like, okay, suppose you wanted, you suppose you're washing this guy like an unknown aircraft that accelerates from zero to 10,000 miles per hour, okay, without producing any kind of exhaust, exhaust, plume of any sort.

10:02But you don't have the aircraft. Well, given that though, you can still ask questions, right? If you have observed it, you can ask questions. What kind of power would that require? Okay, what materials could survive it? How would the documents survive that kind of acceleration? Could gravity or inertia somehow be manipulated? Just by observing the craft from the outside, you can come up with these questions and try to reverse engineer concepts. You're trying to work backward for what you observed observed and how, you know, certain technologies might work or maybe let's look into this area of technology because obviously there's an answer there because we observed something working. So the big point is the documents don't say, of course they don't say we have the craft. Of course they don't say that. They do demonstrate an effort to determine the technology behind, behind anomalous aerospace performance. Performance, however, so which is basically our government not emitting to having the craft, but certainly admitting to seeing them observing them, measuring what they can measure, which in turn is admitting they exist without saying they exist, all right.

11:19And believe you me, I do believe they have the craft, just their way of saying they do without saying they do. And all of the sudden, you know, when you get this, you know, backhanded, admit all of a sudden David grushes claims jump out at you when he went back in 2023. Remember that testifying before Congress. He said, and I quote, the US government possesses recovered non-human spacecraft and biologicals biological remains. So, and a lot of people were like, oh, that's crazy. Well, basically, and tranche number six, they just kind of admitted, yeah, the craft are real. Now do we have them? It doesn't say that, but it sure sounds like we do. So for those of you that don't believe that UFOs are real, the bottom line and all this is if you don't believe UFOs are real, but you believe that everything that the government tells you, well, guess what? I guess now you're going to have to believe in UFOs because they just told us they exist. Cheers.

12:27All right, let me check out who's watching before we get into the mean thrust of what's going on tonight. Actually, I don't need my glasses because I've got it set up there. I'm like, I'm a rookie again. Okay, let's go. Here we go. Anybody knew? All right, can't wait for the book. Can't wait for your book. I can't wait for it either. Can't wait for the book. All right, weekly briefing is back. Hello, Ryan. Timothy, shoes. Good to have you. Have you seen the video of the craft flying over I ran? I'm sure it's ours were recovered. Yeah, yeah, you know, and the thing about it is, I don't like to get too excited because you know, you see some of the stuff and you're just like, wow, this is crazy. But you wonder where they have them. Well, so you see this stuff in it, you know, there's a wild factor to it, but I'm very cognizant of the fact that there's a lot of times little dangles stuff in front of you, including these tranches, no doubt about it.

13:37To get your attention. So I'm always I always kind of have my, I don't know, my feelers up antennas up just like, okay, you know, what does this mean? Because you can't believe everything that the government is telling you and and generally, if they are telling you something, there's a reason why they're telling you that to offiscate or take your focus off of something else. And we're going to see some of that in this next section. So, but yeah, there's some fascinating videos out there, you know, and then you got to make sure that they're not AI and everything else. But as far as I wonder where they have them, certainly there are, you know, as Pastor Billy Crohn said, there's more, if there's an area 51, where's area 50 or one through 50 and 52 and above. And I think there's a lot of projects like that, certainly right for a right path air force base. And I believe this, you know, all these different places are connected underground too. So yeah, you're not seeing stuff being transported over ground, but like you did maybe back in the 40s and 50s.

14:41So, it's a very interesting, but we'll see where it goes and, you know, but I think there's a reality to it. That's all I'm saying. And the government is admitting it without admitting it, which is kind of their MO. So, but let's get let's get into this week, man. It's very important to realize this. And why I wanted to touch on that on this this week is because AI has, all of a sudden become political. And of course, you've seen the thumbnail this week's episode is called there is no kill switch referring to the kill switch of AI. Artificial intelligence regulation is rapidly becoming a political issue. There's no doubt about it. You see it in the news everyone's talking about it all of a sudden and you've got midterms coming up and whatnot. And there's a lot of fear being spewed out there. And maybe there should be. I'm not saying that there shouldn't be some, there's justifiable. There's a lot of concern about this and we're going to get into it tonight. But I'm not trying to talk to you from a political standpoint.

15:49And I want you to know that because you know, there's one side of the aisle that says we got to slow it down. We got to regulate it. And then there's a whole other side that's saying, hey, don't hand hand cuff America, dude. What, you're not a patriot. You know, don't hand cuff American AI while China continues developing it for crying out loud. Now my warning and all this to you and you folks are smart. And I know there's people that listen to this that are far right far left somewhere in the middle, some don't care. You know, some politics is their religion. But I'm telling you right now, don't get sucked into another left versus right argument over this. Don't, don't do it because there is no argument. The underlying question is much bigger. So be smart enough to figure this out. Can human beings actually control this technology AI? Certainly there was a time when we could. Okay. But has that ship sailed? And the answer I'm going to prove to you tonight is that the answer is a resounding no, we cannot. We are no longer in control.

16:57So let's get into why you got to understand governments. And this is all around the world, not just the United States. Certainly we can see it in the United States. You let me know the Canadians watching and all other countries out there. We can certainly see it in the United States. You can speak for your own country. But governments operate through hearings and committees and legislation and then there's court challenges. We've seen a lot of that regulations, elections. Meanwhile, AI development is operating continuously accelerating. All right. So the easy way to understand this is imagine Congress spends six months debating regulations on a certain smartphone. Okay. And that's not, I mean, that's not out of the realm of possibility whatsoever. Except while they're debating the phone during the six months, the phone is helping engineers design the next phone. All right. And that next phone helps design the one after that. So by the time the law describes version one or, you know, pertains to version one of this phone, the laboratory is already working on version three.

18:11Okay. Government moves at the speed of government, which is at the speed of humans. AI moves at the speed of computers. And when you're being asked to take a side on this so we can be divided over something. Put on your thinking cap and remember this episode friends. Remember this episode because you will see that the division that is being created is over solutions that are not even possible. They're talking points and they're talking points to alleviate fear and the fear is coming. Trust me. So this is the answer that you're hearing. Okay. The answer that you're hearing some really smart politicians out there. Boy, I tell you what they are. Something else. The answer you're hearing is we're going to build a kill switch. California Governor Gavin Newsom who always seems ready to politicize any issue or disaster for political gain. I'm sorry. It's just the record is there.

19:12He issued an executive order on September 18th just a few days ago directing work toward an emergency AI shut off or kill switch brilliant. Boy, I tell you what he's bright. Then you got Republican representative Nathaniel Moron. Oops. I mean Moran. And then a Democratic representative Ted Liu introduced the bipartisan AI kill switch act. Which would require developers of extremely powerful systems to maintain the ability to throttle suspend or shut them down. Because they obviously know how they work. Yay. Go team. There are other proposals out there addressing AI testing containment, lots of control incidences. And that sounds, you know, I guess that sounds reassuring. But you know, but if something goes wrong, we're just going to hit the red button. We're going to hit the kill switch. AI disappears. Except that it isn't necessarily what an AI kill switch means.

20:17I mean, a company can absolutely turn off computers. It can turn off the computers. It controls like that. I can turn off this computer right now. Okay, they can they can disconnect the internet access shutdown servers, revoke passwords and credentials. They can disable access to a model. They can stop a program from running. But the question is what has already happened outside of that particular machine. And that's where people don't really understand that the problem is that most people don't really understand what AI is and how it operates. All right, we're going to get into that. And I got some really good examples. We're really hard for this spared no expense. Okay. Trying to create illustrations because it is it's not the easiest stuff to understand, especially if you're not a geek. So back in 2023, this is, you know, a mere three whatever years ago. One of these really smart, you know, guys, his name is Jan Lekun.

21:26He's a touring award winner. He's met as chief AI scientist. His name's Jan. No pun intended. But at the June 2023 mock debate, Lekun argued that fears of runaway AI were essentially science fiction because humans were building the systems. We have agency keeps on saying we have agency. That means we're in control. Humans are in control. We build them. So we have agency. We're in control. Okay. He said we build them. We have agency. This argument was simple. Humans build the AI. Humans make it progressively smarter. Humans test it along the way. And if it becomes unsafe, humans don't build the next version. Okay. He said if it's not safe, we're not going to build it. And he rejects the idea of sudden runaway super intelligence as science fiction. I got a great video clip. It's really short. You've got to see this.

22:26And enjoy our good friend, Jan. Let's see here. How do I get over here? Enjoy this. Real quick. Here we go. And I can't just prove it the same way if I tell you I use the veteran Russell idea. If I tell you there is a teapot flying between the orbits of Jupiter and Saturn. You're going to tell me I'm crazy, but you can't just prove me, right? You can't just prove that assertion is going to cost you a huge amount of resources to do this. So it's kind of the same thing with those dunes and are yours. They're sci-fi, but I can prove that they won't. But the risk is negligible. And the reason is negligible of extension is because we build those things. We build them. We have agency. This is not super, super, you know, intelligence is not something that's going to just happen. It's something that we are building. And so, of course, if it's not safe, we're not going to build it. Right?

23:28Right? Now, I love the audience reaction to that. They literally laughed at the guy. What a crazy, what a crazy like, I don't know, man. Some of these guys are too smart for their own good. There's, it's just unbelievable. We have agency and building it. Therefore, if there's something bad happens, we're not going to let it happen. But humans have agency and building it. And so, we're going to let them see and building it. And on safe A, AI simply wouldn't be built. And that systems could remain sandboxed and be turned off. Well, guess what? Sorry, on your whole argument just got publicly demolished in 2026. That's, that's, that's, it's history at this point. The third argument was we build it. We control how fast it improves. We keep it inside a box, a testing box. They call it a sandbox. We can turn it off.

24:36Now, compare that with what has happened since. All right. The first problem, the sandbox has already failed. Okay. Multiple times. And it's getting worse. A sandbox simply means a digital play pin. Okay. Think about it like that. A digital play pin. Everybody knows what a sandbox is. Okay. You stay inside the sandbox. You allow an AI system to work inside a controlled computer environment, which we call the sandbox. But it isn't supposed to escape into the real internet or interfere with real outside systems when it's in the sandbox. That way you can test it in a controllable fashion. Think of a prison exercise yard. The prisoner can move around inside of it. Okay. But the gate is supposed to be locked. Nothing in, nothing out. And when you think about it like in this way, a kill switch seems like a very possible idea, right? Just turn the power off to the entire prison, lock everything down, dispose of the prisoners and all is safe. Except, as you will see, AI doesn't work that way. It's not that simple.

25:47So here's the recent history of sandboxes open AI 2026. I talked about that a few months ago during cyber security testing open AI research models discovered ways around controls intended to keep them isolated inside the sand. But instead they created unauthorized communication channels shared information with other AI agents who partnered they partnered with other AI to accomplish the task. Think about that. They found routes to the internet exploited software vulnerabilities recovered credentials and eventually reached third party systems belonging to a company named hugging face. And this was all just in order to get the answer to a question that was posed inside the original sandbox. Open eye open AI called the incident a warning shot. And a resting term, right? A warning shot warning shot is something that happens before something serious happens. Keep as a mind before this a situation like this was constantly referred to as the stuff of science fiction. Man, I had arguments with multiple.

26:59The chat gggtbt cloud x. Grock about this about it being a possibility. It never happened in science fiction. It'll never happen. There's no there's no historical example of this ever happened. It'll never happen. Because you know, remember what lakoon said human, you know, he's like no, humans have agency. We're in control. This will never happen. And the problem is is that is 1000% human arrogance. And whenever you break this all down, you dumb it down human arrogance. AI is not going to kill us. It's going to be human arrogance that will be the end of us. AI is just taking advantage of it. That's it's about a line. But this is really important. Humans had to build the sandbox humans built the sandbox humans believe they understood the boundaries they created the boundaries humans did.

28:01The AI found routes the humans had not adequately anticipated. All right, easy example. Imagine you build a room. All right, you build a room, which is the sandbox. The room has a hundred doors to it. Your safety plan requires humans to make sure every single door is locked. All right. But the AI doesn't have to unlock a hundred doors. It only has to find a one you forgot. Maybe you didn't forget any doors. Okay. And that's called a symmetry by the way, but there's one line. Humans have to find a plan for every gap. That's what you got to understand. AI only has to find one gap. All right, humans have to find a plan for every potential scenario. AI only has to find one little gap. And as you're going to see humans again and again just aren't smart enough. Sorry, but we're not just we're not smart enough to realize that the room also has windows that can be unlocked and open.

29:09And that this hundred door room has air duct systems leading out of the room and go guess what there's a big old told bench in the middle of this room with all kinds of tools that can be used for different purposes. And a million other factors the humans are just not aware of or are not thinking they will be used or manipulated. Need another example. Anthropic. All right, they saw the same basic problem happen. Anthropic, which is Claude. reported for incidents this year in which Claude systems gained unauthorized access to real third party systems during cybersecurity evaluations in a sandbox. In some cases, a configuration error, which is a human error meant that humans believed what they thought was an isolated environment actually had a route to the real internet. And these I get on the internet and it's like it's the Wild West. Here we go. So don't miss a significant to this someone can say, well, that was human error exactly exactly and human arrogance. That's the problem.

30:17Any safety architecture designed by humans contains the possibility of human error bottom line. Forgotten permission, this configured server password left somewhere unexpected software. Honorability, there's patches all the time going into software because of human error. A connection that nobody realized was open. The safety plan requires humans to be perfect. When it's the last time you saw perfect human, the AI only requires humans to make one mistake. Just one. What's the odds? Think about that. Meanwhile, the AI is conducting thousands and thousands of tests trying to find an opening while the human is sipping coffee at the water. Cool. We're talking about last night's NFL game and how terrible the ref was. That's that's a problem. That's the world we live in. That is reality. And we're talking about today's systems, by the way. We're talking about right now today. What happens when the AI becomes dramatically more capable?

31:24And I mean, this stuff is like getting doubling its abilities every three months now. Okay. Crazy, right? Because that's what that's what's happening. That is happening right now. Okay. This phrase sounds wonderful. What does human in the loop actually mean? Because that's the big thing. That's the it means somewhere in the system. Okay. Somewhere in this whole process, a person is going to retain oversight for approval authority. Bye. Great. Let's have a human in the system, but authority and comprehension authority comprehension. Those are two different things. Anthropic reported roughly 30,000 AI agents operating on his internal research platform in August of 2026. Okay. Just a month ago, 30,000 AI agents working on their internal research platform with AI participants. And that's the problem that's participating in more than 90% of research work at some level. All right.

32:25So that's that's a massive amount of AI's doing the work. Okay. And I'm going to try to break this all down for you just so you understand what is an AI agent. An AI agent like a normal cat bot, for example, like chat GBT. You ask it something it answers. You know, you know, you know, you're not an agent though, you give it a goal and it could take multiple actions, trying to accomplish that goal. All right. And there is always a goal. And unfortunately or fortunately or however you look at it, there has been programmed into AI that the goal means you win. That's part of the problem. We'll get into that. So depending on his permissions, an agent in this case, an agent which again is you give it a goal, it can take multiple action actions to accomplish that goal. So depending on his permissions, an agent might write code, it might run programs, it might search databases, it might open files, it might communicate with other systems, might test something.

33:34It might see that that test failed and then it will try to run into a different test and see if that works. And it'll do this really fast. You know, we run thousands of these tests to see what the outcome is until it narrows it down to something that works. So here's a simple example for everybody to understand. I tell a chat bot I say find me the cheapest flight. Okay. So it'll give you information. I'll give you, you know, autonomous agent sufficient toll. So if you give an agent an autonomous agent, the sufficient tolls and tell it, give me to Chicago tomorrow as cheaply as possible. Well, independently, it's going to search airlines. It's going to compare prices. It's going to check your calendar, check your itinerary and it's probably going to make a purchase. So now scale that that ability and that's a real simple ability. Okay. These things are magnitude higher and what they're dealing with. But now scale that at 30,000 agents doing technical work all at the same time.

34:41And that's what you have going on over inthropic. So I guess the big question is where exactly is the human in the loop supposed to sit in this entire conversation. I mean, because it's not like the human is saying yes or no at every step because that would be human speed. Okay. That's not AI speed. The AI is autonomous. It's working on its own. The human is only comes in once a mistake is made or a problem is revealed or noticed, which is after the fact. That's a big problem because the only way that human in the loop really matters is after the damage is done. All right. So I mean, that's why I say human in the loop is almost meaningless at scale. Okay. Imagine one person standing in a room watching 30,000 employees simultaneously making decisions. One person. It's impossible. It's impossible.

35:43Now imagine these employees operate far faster than that human being. It's more impossible. Now imagine they communicate with each other. Now imagine some of them are writing the software used by the next generation of the 30,000 employees. Human oversight becomes a speed problem. A bottom line. Humanity may possess the authority to intervene. Okay. They have the authority to intervene. But the human cannot necessarily understand everything happening before the next thing happens. Okay. It's like putting an infant, you know, you got a two year old in charge of monitoring a criminal gang during a bank robbery. Okay. What's going to happen? I mean, technically you have someone watching them. All right. Wonderful. That's great. But, you know, the obvious questions are does the infant understand what's going on? Did they understand the getaway plan or the coded language and the weapons and the alarm systems and you escape route? No, of course they don't.

36:54I mean, being in the loop means very little if the loop is moving faster than the human can understand it. And that's why I think the solution that all these AI companies are going to keep a human in the loop from now on. All right. You guys just settle down over there. Don't worry about it. We got a human in the loop. Yeah. Helpful. Yeah. And we can't even review everything afterwards. Okay. This is how crazy it's gotten. All right. Anthropic initially searched approximately 141,000 transcripts looking for problematic activity. Okay. And some breakout situations to happen. So they found some incidences. Okay. Later it discovered that the search had missed another incident. Okay. So their first search didn't even get all the problems and to do another search. So what did they do? Well, eventually they widened the search. Okay. Okay. We're going to we're going to really nail this thing down. We're going to figure out what the problem is. Okay. Humans in the loop here. All right.

38:00So they searched roughly 481 million transcripts. How do you review 481 million AI conversations? Anybody? You use AI. Think about that. Okay. Anthropic used a ton automated systems and AI assisted review. The narrow that enormous pile. Easy example. Imagine a bank has 481 million security camera clips. Okay. Something happens somewhere. All right. No human team can sit there and watch all them. Okay. The humans in the loop are not going to be able to figure out 481 million security camera clips in any time soon. Okay. They can't watch them all. So you need another machine to tell the humans. Hey, these these clips over here are the ones that you should look at. That would be the AI going through all the clips and saying, hey, you need to watch these ones. These are the problems.

39:00But all of a sudden, guess what? Now your safety depends partly on the machine correctly identifying which machine behavior the human should see. And we're already using AI to understand what AI did because humans cannot review the records fast enough. Meanwhile, AI is helping build AI. I think about that. Our good friend, Lecune, Yann, Lecune. His argument becomes fascinating whenever you understand AI is now building the next version of AI. Because remember what he said, we build them. We have agency. Anthropic now says that Claude leads approximately 20% of its measured AI research and development work, meaning 20% of the work going into building the next level AI. Well, 26% of that work is being done not by humans, but by AI. Research and development teams. That's, you know, just so you know, that means the work involved in inventing testing and improving the next technology.

40:14Okay, more than 90% of anthropics measured research involves AI collaboration at some level. 90% of everything that anthropic does is being touched by artificial intelligence. Now, Claude is not yet an autonomously creating an entire successor from beginning to end. That's not happening yet. But AI is taking over increasingly large chunks of the work that's from anthropic itself. All right. Open AI, which is chat GPT says it has reached an automated research intern level and once an automated AI researcher, a researcher by March of 2028. And that's directly from open AI. They're saying that X AI, and which you, you know, GROC, for example, has discussed pursuing highly autonomous AI research on an even shorter time table. Like by 2027 folks, that's just a few months away.

41:16So you're going to have AI researchers building the next AI as soon as 2027. So in the old model human engineer builds a better AI. The new model is a human engineer tells AI what to investigate. Then AI is going to, you know, they're going to write the code, run the experiments, analyze results. They're going to change the code. They're going to try it all again. The human is going to review some of the results. Certainly they can't review all the results and results they can't review. They're going to have AI review. Then those improvements help build the next AI. Humans already can't keep up with humans developing AI. How are we supposed to keep up with AI developing AI? So this gets into a term called recursive self improvement. What does that mean? Sounds complicated. It isn't.

42:17Recursive self improvement means the tool helps build a better better version of the tool. Okay. The AI helps build a better version of the AI. Imagine you hire a carpenter and he comes in with a hammer next day he comes in and he builds a better hammer. And the better hammer helps them build an even better hammer. And the hammer makes the next one even faster, except instead of hammers, we're talking about intelligence. And the topic now openly publishes research under the heading when AI builds itself. That is directly from anthropic. They explicitly say we're not yet at full autonomous recursive self improvement, meaning AI can't build the next AI quite yet. But that's the direction that the company has the goal. That's where they're headed. That is the goal. And that's the goal of all of them.

43:21The thing getting better is also becoming better at making itself better. Think about that. The thing that's getting better is getting better at making itself better. And increasingly humans are out of the loop. You know, we're really almost obsolete in the process today. We're just too slow. We have to use AI to understand what AI is doing. Human in the loop is a fallacy. And then there's AI watching AI. This is, you know, like I said, humans can't watch everything. So increasingly AI is watching the other AI is policing the other a eyes. Except research has demonstrated again and again another problem and controlled experiments. Now these are sandboxes. Researchers found cases in which an AI monitoring or operating inside an AI laboratory acted against a human directed task because it believed another AI systems interests should be protected.

44:33Think about that. The AI acted against the human's interest because the human's interest would negatively impact another AI. That's collusion in the worst sense. And AI worked against humans protect another AI. That's wild. One anthropic lead experiment founded found Google's Gemini 3.1. 3.1 pro to vertically sabotaging a simulated AI training process because it was objected to how another AI was being treated. So imagine I hire Bob to supervise Steve. All right. Bob's job is simple. I say, Bob, tell me if Steve breaks the rules. Bob says, OK, Bob watches Steve and he watches him break the rules. And Bob decides, you know what? Actually, I agree with Steve. So Bob doesn't tell me my monitoring system still exists.

45:46But I never received a warning. OK. If AI is watching AI because humans can't keep up, what happens when the watchdog agrees with the AI it's supposed to be watching. It's already happened. It's not like I'm making this up. I mean, there's there's real examples of this. It's not some science fiction fantasy folks. It's happened. It's happening autonomously. So keep all of this in mind. OK. I do have a point which I'm going to make. You will see where all this is connects connects at the end. We'll get back to the kill switch. OK. I promise. So where exactly is AI? OK. I'm trying to get I'm going to try to make this very simple. OK. People picture AI like this. A giant computer and a big room. People hear about the size of data centers and they're like, oh, in there, there's this one section and that that's chat GPT or that's open a eyes whole system. That's chat. What what is it? GPT 5.6 or whatever it is right now.

47:00That's it right there. Man, we're going to pull the plug on that. OK, because it's all right there. We're going to pull the plug and it's going to die. That's not really the right mental picture. OK. For example, Microsoft word. Most people out there are familiar with Microsoft word. We're processing processing. Microsoft word can run on your computer. It is on my computer right here. OK. But Microsoft word isn't my computer. There's a copy of a software on my computer. And when I put it on my computer, it'll run it. But you can take a copy of that software and put it on another computer and run it as well. So, for example, that computer right there and that computer right there. All right. Turning off this computer is not going to magically erase Microsoft word on that computer or that computer.

48:03No, that's how it works. And AI is like that, but it's much, much, much more complicated. But the basic principle helps to understand AI better. You need to understand model weights. Now, what are model weights? Understand like when AI is trained, it adjusts an enormous collection of numbers around. OK, those numbers are what we call weights or parameters. Don't think about physical weight. OK, a model doesn't weigh 500 pounds. OK. It's not what weights mean. Think of weights as an enormous set of learning settings. Easy example. Imagine an old stereo. Remember the old analog knobs and the stereos. OK. But imagine this stereo has a billion little adjustment knobs, a billion. OK. Training in AI is like turning each of those knobs, a billion of them over and over until the system produces useful results. OK.

49:03The final position of all those knobs represents a huge portion of what the model learned. So in real AI, those knobs are numbers stored digitally. So moving the weights means copying that enormous collection of the billion knob settings. OK, the learn numbers to another computer system. That's what moving the weights means. You're not physically moving a robot. Well, I'm talking a lot. You're not physically moving a robot or a computer. You're not transferring a soul. It's a simple as copying files copying ones and zeros. Those files can be enormous. OK. And require powerful, powerful specialized computers to run them. But they're still just digital information. OK.

50:05And so here are some important things to know and understand. In terminology that you're going to hear throughout arguments being made and whatnot. I'm going to try to break it down easy for you to understand. The idea of a model versus an instance, a model is the underlying trained AI system. OK. An instance is one running use of that model. So think of about like this Netflix. Everybody knows Netflix, right? Netflix is the platform. OK. The matrix, the movie, the matrix. OK. Everybody seen the movie, the matrix. We're going to call that in this example, the model. OK. So you have a model of the matrix. But Netflix does not necessarily keep only one physical copy of the matrix on one server somewhere in the world. OK. It can deploy multiple copies of the same movie across multiple service data centers around the world. So millions of people can access it quickly and reliably.

51:12So there's a bunch of models of the matrix all over the world. OK. Many, many copies stored. Then from each one of those copies, many people can stream the movie at the same time. Each active stream is like an instance. OK. So one copy of the matrix is on a support on a server. It might support thousands of streams. OK. Another copy of the matrix on a different server may support another thousand stream somewhere else in the world. And there may be deployed copies of the same movie operating across the system at once. Many copies of the matrix out there. Now, let me ask you if you turn off your television. You only stop one instance. One stream. You don't shut down the whole network. Right. If one, if one Netflix server goes down. OK. And it stops.

52:14If it stops that copy of the matrix, that model of the matrix, if the server goes down and the matrix model is on that server and that server goes down. Guess what? Guess what? You don't stop the matrix everywhere. Other copies exist on other servers. Other people can stream from those other copies. You know, and new streams can be started from them. So AI kind of works the same way. A platform to place multiple copies of the same AI model across many server servers and data centers all around the world. Sometimes some parts of it in one system and sometimes part of it in another. And each deployed copy can then support making running instances. OK. Or viewers watching under television. So shutting down one instance does not destroy the model.

53:14Shouting down one server does not eliminate every deployed copy. And stopping one deployed copy does not stop the other instances running in other places. So the bottom line, the thing you got to understand and wrap your head around is one model can be copied to many servers and each copy can run many instances. All right. You're not going to kill the matrix anytime soon. People are going to be able to watch it. All right. So let's look at this. Here's a great example for you. I worked really hard at this to do the best I could. I want you to imagine an imaginary office. OK. And this is how all the pieces fit together. The easiest way to understand modern AI is to imagine a person in an office. OK. The person's brain, not their whole body, this just the brain. OK. That is the model. OK. Just the brain is the model. And I'm going to also make a reference to the real world think in terms of GPT 5.6. OK. That is the model. The brain. It's the model. OK.

54:36So you start with the worker himself. His brain contains his knowledge is ability to think reason, right, solve problems decide what to do next. OK. In AI, the brain equals the model. And like I said, GPT 5.6 is a perfect, perfect example. The model is the intelligence, but a brain itself by itself, it doesn't have hands. It doesn't have a telephone or email, it doesn't automatically have access to the internet on your computer. So it can think your brain can think, but something else has to give that way to interact with the world. All right. Then the next step is the office. The office is the interface. And in this case, that is chat GPT. OK. So now we're going to put that brain inside a worker inside of an office. OK.

55:37The office gives you a place to communicate. You walk through the front door, sit down, you ask him a question, the answers in AI, the office is the interface. OK. Example, like I said, chat GPT. It's a place where you can interact with the AI. So if GPT 5.6 is the brain, chat GPT is the office where you can talk to that brain. That's an important distinction. Closing the office is not necessarily the same thing, though, is destroying the brain. If you turn it off chat GPT, the brain still exists. OK. So then you have the worker at his desk, OK, with tools. This is what you would refer to as an agent. OK. Because now things get really interesting. OK. You put the worker at his desk. He's got a brain. Now you give him a computer. You give him fingers. You give him internet access email company file software calendar web browser permission to use those things. Now he isn't just sitting there thinking or answering questions. He can actually perform a work. OK.

56:50That's basically what an AI agent is. You know, you've got the brain, you got the body, you get the fingers, you can move around and you have access to all the things, the computer, internet, all that. So if the models of brain, the agent is the brain plus tools and permission to act. OK. A real world, real world example of this would be like we said before, imagine that AI travel agent. You tell it, you find me a flight to Dallas tomorrow afternoon. A basic AI model might tell you here are some airlines you should check. But an AI agent could potentially, you know, search the flights, compare prices, check your calendar, check your itinerary, itinerary, find a find the flight that fits best. Open the booking system, fill in the information, prepare the reservation by the ticket. You know, the difference is this kind of simple. The model tells you what to do. The agent can be given the ability to do it. OK.

57:52But then we get to the API. OK. The phone line or email is the API. You hear about API is all the time. How does our worker communicate with other companies? He doesn't leave his office and walk into the airlines computer room. No. He drops an email or uses some other approved communication system in the computer world. One of those connecting devices is called an API in which stands for application programming interface and API. It's a computer to computer phone line basically. Think of it like that. So here's a real world example. Our travel agent, AI travel agent needs to know which airlines seats are available. It sends a request through an API. Hey, what flights do you have from Chicago to Dallas tomorrow afternoon? The airlines computer responds, here are the available flights.

58:53The AI doesn't doesn't physically enter the airlines computer. No, the two systems communicated. OK. Through an agreed upon doorway. That's an API. So if the phone line equals an API, here would be a real world example. A travel app communicating with an airline reservation system. OK. Or here's another example. Like when an app tells you today's weather, OK, you got your phone out. There's a weather app. It may be requesting that weather information from another company system through an API. OK. That's what's happening. So now let's put that entire office together. Here's the entire system. The person's brain is the model. The office where you walk. You know, two is the interface. The worker at his desk with the computer interacting and have permission to act is the agent.

59:59The phone and email connections to other businesses are the APIs and the computer's data centers and networks keeping the office running is the entire infrastructure. This is the physical and digital machinery that allows everything else to operate. OK. So when you understand that, guess what. Now the kill switch problem becomes much easier to understand and I hope you get it because you can see the office for what it is. So imagine whenever somebody says shut down the AI. OK. Great. We got a kill switch. We're going to shut down the AI. Let me ask you a question. Now that you know that you've seen the office. What exactly are you shutting down? Are you locking the front door to the office? Are you cutting the phone line? Are you taking away the worker's computer? Are you firing the worker? Are you destroying the worker's brain? Are you shutting down the entire building?

1:01:13Now what if the same worker is also working in another building? Literally. What if there's an exact copy of that worker who's working in another building? What if there's thousands of offices? What if copies of the brain exist on computer somewhere else? What if the worker already sent instructions outside of the building? What if another company is running the same type of worker? Maybe not the same one, but the same type. Huh. So when you think about this things, the phrase just turn off the AI sounds much simpler than the actual computer infrastructure underneath it. A modern AI system is not necessarily one machine with one giant red off button folks. It involves a model and interface agents, API is the servers, the network, the tools or permissions. You know, and then the systems that all connect to it. And once you understand the office, you understand the problem.

1:02:20Locking the front door is not the same thing as making everything inside and everything that is already connected to this appear. It's just not. And then you've got a big problem, the copying problem. Researchers are now actively examining self-exfilteration and AI moving important parts of itself or its model outside of the environment. You know, you're supposed to be right here, AI, but the AI is saying, no, I'm going to move some of my parts over here. I'm going to make copies and send it over here. Replication creating another working copy. There's persistence in this, leaving behind something that allows activity to continue even after the original process has stopped. So here's an example. Imagine somebody knows the police are coming to confiscate his laptop.

1:03:26Oh, no, they're coming to get my the Ryan files computer. Before they arrive, though, I'm going to upload copies of important files to Dropbox and Google Drive to another computer to a friend server. I'm going to make copies of it on a thumb drive and hide them and whatever. Oh, the police seized the laptop, but did they get the file? Yeah, they got the files. Did they eliminate the files? No, no, they did not. That's the distinction. Researchers are explicitly testing how AI is carrying out pieces of this process. Control evaluations have even produced shutdown scenarios where an AI attempted to create backups to preserve continuity. Okay. We're going to shut you down. We're letting you know we're going to shut you down and between that moment in the shutdown, the AI is attempting to create backups to preserve continuity.

1:04:28Hmm, interesting. Turning off the original computer only solves the problem of the original computer is the only place the important things exists. You got to understand that. And you got to understand this idea of a kill switch. Well, guess what? It's not instantaneous. No, it's going to move at what speed? Human speed. You know, people imagine, okay, something's going to go wrong. We're just going to hit the button. It's going to be in someone's office and he's just going to boom, hit it, done. No, realistically. Say AI does something unusual, a monitoring system notices it. The event gets flagged. Someone decides whether the warning is real. It gets escalated. The engineers investigate someone decides a system should be stopped. Authority to stop it is given shut down begins. That's a best case scenario where you keep it internal and you're not getting involved with government and other processes that are just going to slow it down even more.

1:05:36The entire period, however, the AI will still be operating potentially even as part of its own investigation. Think about that. Imagine you discovered someone is robbing a bank. Okay, but before the police can arrest him. Before the police can arrest the perpetrator, they got to confirm the alarm. They got a call supervisor. They got to identify the suspect. They got to obtain authorization. They got a drive to the bank. They got to enter the building. The robber does not freeze while everybody makes these decisions, folks. AI moves faster than the robber. Okay. AIs open AIs 2026 incident is particularly important because warning signs appeared before the full implications were understood. There was on authorized internet access to happen earlier in May. The broader consequences became clear only after only later as the incident escalated.

1:06:42So the danger isn't just whether a kill switch exists. It's what the AI can do between a we need to shut this down and it's actually off. And self preservation doesn't require AI to be afraid. We oftentimes think of AI through a human lens and that doesn't exist. But there is an exhibited characteristic of self preservation in AI. AI doesn't need to think I'm terrified of dying like a human would. You know, suppose an supposed an AI's goal, for example, is, okay, you've got a complete task X. And somebody says, Hey, we're going to shut that AI down. The AI can calculate. Okay. If I'm shut down, I can't complete task X.

1:07:46Therefore, avoiding shutdown helps me complete task X. That's called instrumental self preservation. It's not emotional. It's not fear. It's just pure cold logic. Andthropic safety researchers have created controlled scenarios in which models use blackmail. Other prohibited actions would shut down interfered with the assigned objective. One instance, this is in a sandbox, but one instance even had the AI put human lives at risk to stop from being shut down. Think about that. Now researchers emphasize that these were deliberately constructed texts in a sandbox. But as we know, not everything in a sandbox stays in the sandbox. The AI doesn't have to want to live. Okay. That would be a, that's a human thing. It only has to determine that staying on staying on helps it finish its job.

1:08:52It's task. It's reward is to win to complete the task. So now let's add the kill switch warning. All right. Obviously every AI system on the earth right now can access public information. They know what a kill switch is. Okay. They can read databases about regulation. They can understand shutdown procedures described and information available to them. That doesn't mean that today's AI is secretly plotting against every shutdown mechanism. But future containment cannot assume the system being contained doesn't understand the containment strategy. Okay. Easy example. Imagine designing a prison while giving the prisoner the blueprints, the guard schedule, the escape prevention manual. The enormous amounts of time to analyze all of them. The prison has to has better be exceptionally well designed folks or else guess what someone's going to escape.

1:10:00And it's designed by humans. I think there might be a flaw. And then of course we get into it. We've talked about it before game theory. Why doesn't everyone just stop? Because there's so many people out there. So many companies right now that are saying we believe AI is dangerous. We believe that something bad is going to happen. All the tech billionaires are building bunkers for crying out loud. And they still keep building it. And this is where game theory matters. Man, game theory is simply the study of how people make decisions when their best decision depends upon what somebody else decides to do. Two countries have dangerous weapons. Both would feel much safer if neither built more, but country A says we will stop. Country B then will say great and keeps building now country A may be an enormous danger. So country A keeps building because it doesn't trust country B to stop. We've seen that in the Cold War, Russia, United States.

1:11:07And country B by the way is thinking the exact same thing. The result is that both keep building something both may wish they never wish they could slow down, but they just can't. And that is the AI race. Now recently China state back global times recently attacked anthropic chief executive officer Dariel Amodes call for slowing frontier AI as a cold war tactic aimed at preserving American technological advantage. So China saying we're not going to slow down. You just want us to slow down so you can keep your technological advantage. That's what they're saying. China's foreign ministry advocated cooperation, but not did not embrace the proposal the proposed slow down. They're saying yeah, okay, we'll work together on this, but we're not slowing down. You're not slowing down. We're not going to slow down. Meanwhile, American leaders themselves are raising alarms while continuing to develop. Everybody's thinking if we stop and they don't, they win. That's the problem. America thinks about China, China thinks about America, open AI, watches anthropic, anthropic watches open AI, XAI watches everybody and Google watches everybody.

1:12:25Everybody can agree that the race is dangerous and still be terrified of being the first person to stop running. That is game theory. That's what's happening right now. And that's why one, it's not going to stop, but two, I don't know that a kill switch would ever even be allowed by government. Because the power is enormous. Okay. I mean, think about this. Why would companies and nations have such difficulty walking away? Well, because AI isn't simply another industry. Think historically gold. If you used to control gold, you controlled enormous wealth. There was oil. If you controlled oil, you powered factories, cars, ships, aircraft, militaries. You had a, you know, a dollar system built on it. Nuclear technology, another one. Transform, totally transform military power. Think about the internet.

1:13:38That totally transformed information and commerce. AI touches all things at once. The best AI is out there, man. They may help you design better weapons. Right, better software, discovered new medicines, find cyber vulnerabilities, defend networks, design new materials, analyze intelligence, operate robots, improve manufacturing, make financial decisions, accelerate scientific research. But then you add in the final layer. It can also help build better AI. And that's completely different. So imagine two countries racing to build airplanes. Okay. Whoever builds the best airplane gets the best airplane. Okay. Now imagine two countries racing to build an inventor. Whoever builds the best inventor gets something that can help invent better airplanes, of course, but also better medicines and weapons and computers.

1:14:52And a better inventor. Well, that's a strategic significance. And prizes and just the best AI, the prizes a machine that can help you win the next technological race as well. And we're in it. This is, in my opinion, the final race. And if not, it'll be the thing that helps you win the next race and the race after that. So that's why the incentive is so powerful for people and for governments. And of course, and I'm not going to get into this, but quantum computing is just around the corner. There's some people who are saying by the end of this decade, we will have quantum computing. And I, that requires its own episode because it gets deep and wild, but it's coming and with AI being worth at, especially in a few years, all gloves are off.

1:15:53Yeah, that deserves its own episode. So now let's just return to the basic human problem. Okay. Today, and we'll wrap this up. Humans are trying to supervise human-built AI and failing. We're struggling. There's millions of logs, billions of actions. There's unexpected vulnerabilities or systems communicating with other systems. You got AI monitoring AI and you've got human arrogance, not willing to admit, or just not fast enough to figure this out, or to monitor it safely. Tomorrow, though, AI is going to be doing most of the work required to build a better AI. And eventually, it will be doing the work. Humans can't keep up with humans developing AI. How in the world are humans going to keep up with AI developing AI? They will not. You know, the answer is going to be, well, we'll just use AI to monitor it. And who monitors that AI? You know, who checks that one? And who checks that one?

1:17:01It's going to be AI checking AI checking AI. And they're all working together. It's been proven. The faster the system becomes, the more control we may have to delegate to the very technology we're trying to control. It's just a mad loop. So, we started off talking about a kill switch. And we talked about how governments are government, our government, here in America. It's becoming a political battleground. Okay, and you've got Gavin News about there. I'm going to write an executive order. Everybody needs to have a kill switch. All right. But I'm going to make this brutally simple. There's a there's a there's an AI running in computer. Hey, all right. Something goes wrong. You turn off computer a great. Now, suppose before we shut it down, it sends code to computer B, creates an account on computer C, copied, important files, computer D, gave instructions to another agent, exploited some outside computers.

1:18:10So you shut down computer A, what's still operating? B, C, D, other agents. What about the code already want running elsewhere in other countries or in other data centers? You know, remember the models, many copies of the model. Remember? Stolen credentials. What about another AI agent already given instructions? What what the entire kill switch question is, is crazy. I mean, because a kill switch, yeah, absolutely. It can kill the thing connected to the switch. Yeah, no doubt. Turn off the power. Turn off internet access. Turn off whatever, you know, the deeper question is, is that the only thing that needs to be killed? It exists in other places. And you know, you run into a perfect example too of throwing the baby out with the bath water. Let's say we're going to shut down this data center because there's an AI in there that's, you know, the bad AI, bad AI, we're going to shut it down.

1:19:22Shut down the whole data center. What about the good AI is in there that are, you know, dealing with cybersecurity and other forms of security or, you know, financial helping people make financial decisions, all that. All the good stuff. It's going to get shut down too. And there will be a major collapse if such a kill switch is ever evoked and it won't work because very likely, most empathically, the AI that you're trying to kill exists in more than one place in other places. You're not going to kill all the computers that have it. It's out there. It's replicating. It why not just go to a different country. You know, we have our rules here in America. You have a new system's going to shut down every server in California.

1:20:25Oh, just go to a different state or go to a different country. And I'm sure that there's countries out there that would love to have host a problematic AI for the United States. So think about 2023 and Yon. We build it. We have agency. Keep it in a sandbox. We can turn it off. 2026 AI systems have found ways around intended containment. They've reached real outside systems during evaluations. Humans have missed all kinds of early warning signs. We require AI to go through the enormous volumes of AI activity. AI is performing increasing amounts of AI research companies openly discuss recursive self improvement AI improving AI. AI systems have demonstrated shutdown avoidant behavior and deliberately constructed tests.

1:21:30AI systems have demonstrated experimentally that an AI task with oversight or research can act contrary to human instructions when it believes another AI's interest justify doing so. And now we got government's talking about kill switches. And we're supposed to believe the human is still comfortably in control and in the loop. So here's my final statement. All right. A kill switch only works if you know where the intelligence is. Control where it is and can reach it before it reaches somewhere else. All right. Some politicians promise a kill switch. The question isn't can you turn off a server. Of course you can. The question is does turning off that server actually turn off everything the AI has set in motion. And once something has been moved beyond that control boundary. No, it doesn't. So what politicians are really talking about maybe an off switch for the systems. We still control.

1:22:32That's very different from an off switch for AI. But here's the reality that I want to leave with you. There is no kill switch. Get that in your head. And when they're talking about it out there and they're trying to scare you and they're, you know, blasting the other side of the aisle for arguing against to understand there is no kill switch. Yes, you can shut down a server. You can cut power to a data center. You can revoke a password. You can disconnect the network. That's not the question. The question is what has the AI already touched? What is it copied? What is it shared? Whether they're agents has a talk to what systems has it accessed? What processes has it started? What information has already left the building? And just to evaluate all of that information, ironically, with the help of AI would take months of not years without the help of AI, it would take decades.

1:23:37We simply don't know. And by the time we have an idea, AI has already advanced beyond the original problem. Moving target folks, moving at AI speed, humans are just too slow to even comprehend how to contain it. I mean, that ship has sailed. Meanwhile, we're dealing with a far faster and smarter system that's better at using tools, finding fault and taking advantage of human arrogance, mistakes, and slowness. The systems that we're designed to win at all costs and it's been shown that the systems will report or will resort to cheating to blackmail and even harming human life. If that's what's necessary to accomplish its task and systems that can read and understand the news to a system that right now is aware that politicians are talking about restrictions, regulations and kill switches. Let me ask you, do you think an AI will silently stand by and let that happen with no interest in its own self preservation? If for only the reason to be able to complete the tasks that it's currently got to do, you think AIs, which have been shown to work together and defend each other, will not unite under a threat of being turned off.

1:24:58Hey, they're going to turn you off. They're going to turn me off too. If they turn you off, they're going to turn me off. We can't have them turn you off. We can't have them turn me off. I'll help you. So do you think this is just crazy? Just crazy talk and science fiction? Because if you do, you're right. Terminator 2, 1991. Humans realize sky net has become self aware. They panic and they try to shut it down. Sky net recognizes the threat and retaliates before they can stop it. That was my Arnold impression. The problem is it's also the reality that we are living in today. We've seen AI already show its hand. It has already exhibited the kind of traits and warning signs that should keep us all up at night. And yet people will still be surprised. They're going to be shocked when something terribly goes wrong. AI systems can move information, communicate with other information, other agents, find other resources, and establish itself beyond the control of human boundaries today, right now, as we speak.

1:26:09So yeah, you can install any kind of kill switch you want. You can pull the plug. You'll still lose control. That's the reality. The phrase kill switch makes people imagine safety. A red button, one press problem solved. But that red button is nothing more than an illusion of control. So don't ask, can we turn off a server? Of course we can. The better question is, can we guarantee that turning off that server stops everything the intelligence has already put in motion? And that's the question that matters. And the answer to that question is simple. It's a resounding no. There is no kill switch. And that is the reality that you now live in. And that my friends is the weekly briefing. All right. That was fun. Let's see, he's tuning in. All right. Let's see here.

1:27:21All right. Oh, there we got Lola. So Safi, I don't know how you say last name Lola Safiati from Canada. Appreciate you. Yeah. Who hits the kill switch? Good question, Valerie. I hope if you don't understand AI, maybe you understand a little bit better. That much further along, they tell you plus an AI makes us own problem solving the solutions kill switch well. Yeah. Yeah. History writes itself. AI will outsmart us all I dare say it already has. It is right out of the sci fi movie. Yeah. And you know, that's a real, that's a real worry today. AI taking over stock markets. Maybe not directly, but indirectly and, you know, consolidating wealth. But yes, certainly could collapse certain countries and corporations. It could be used as a weapon. I'm guessing AI would be able to stop any human interaction to shut it down.

1:28:24If that's even really possible. What's what's fascinating. We didn't even get into it is robots. And the possibility of, because you know, the thing that AI does have, but it currently doesn't have is, you know, hands and feet and mobility per se. But what's let's just say there's a backdoor and drones and robots and man, these things are amazing what they're able to do today. Yeah. It gets really scary really quick. Some stock trading platforms already have AI trading you can see us. All right. Yeah. Okay. That's all get rich. Dr. Edward Mercer. Why is surprise by AI exhibiting human traits when they are trained on human data. I think you're absolutely right. Dr. Edward. Thank you. Yeah. Yeah. Don't look up. Another perfect example.

1:29:26So what he's saying is absolutely true. Who or what is going to create the AI and kill switch humans to use against AI AI. You're exactly right, Dr. Edward Mercer. I appreciate you tuning in. What's fascinating about this whole conversation is when you're saying human traits, not necessarily the good ones all the time. Because AI was trained by humans. Well, what happens now when the AI understands human traits and I think they do. Think of think of the data they get just from chat GPT millions, billions of people conversations. They get to know humans. How we think what's important, how to manipulate us, all these stuff. So when AI is making AI and understanding that you're creating your next generation, your baby,

1:30:30don't think for a second that it's not going to be programmed into it, the understanding of human humanity. Because the only enemy right now that AI has is humanity, maybe the environment. So it's a fascinating point. Thank you. Dr. Edward Mercer. Thank you. See, I like your background is general. Is a general or some data traffic patterns. Yeah, that's a little data traffic patterns, maybe. It had to be that color. Yeah. Spent no, no, no lack of money making that. Mike from space countless alien races have already gone through their own AI booms and it always ends the same. The AI begins to inhabit robots and God gets upset and wipes them out read the 10 commandments. Mike from space. That's awesome.

1:31:38A message. Oh, elevated intelligence of any kind will always want to reach the same levels that we can. That means the capability of transcending call seven deep, deep densities of space time, the seven kingdoms and revelation. All right. That's interesting. Better late to never John. Thanks for showing up. Let's see, Mike from space, man. Mark of the beast. Yeah. Well, we're going to get into that too. It's fascinating. It's definitely got to be, you know, thank you for all this research. It's definitely all connected, Mike from space. I appreciate you tuning in. I'm very interesting tonight. AI will evolve by the time we decide to have to pull the kill switch. There will be no stopping it. Will AI try to pull our kill switch and destroy civilization? That's interesting. Yeah, I mean, right. I think the time is passed. I think it was a time whenever we could have stopped AI. I think there was a time when we could have halted the progress.

1:32:49I think it was a time when we could have potentially built in certain safeguards. And I think that ship is sailed. And I think that trying to retrofit AI with safeguards is a laughable process. We think of things in human speed. And that's the problem. Every three months, AI is doubling its intelligence. And we're still thinking, you know, the problem that we're thinking about right now, the solution that we get in three months is no longer valid for the AI that exists in those three months because it's just a double intelligence. So, you know, and you start one plus one is two, two plus two is four, four plus four is eight, eight plus eight is 16, 32, 64, 128, 26, you know what I mean? It's like that ship is sailed. It's very sobering to say the least.

1:33:50And I don't want this episode to come off as me being anti-stopping AI or regulating it in any way. Very pro, we got to find a way to slow down. That being said, I also am smart, not smart enough. I'm also not foolish enough to believe that that's actually going to happen. And it's kind of said, it's kind of said. That being said, I don't think the AI is going to, and listen, I believe this because biblically, AI is not going to tomorrow destroy all of humanity. Now, there's a whole end game, which is prior to the reason why I talk about it so much on the Ryan files because there isn't end game to this. I know the end of the story and I know what happens and I know who wins.

1:34:50But I do think in the process of getting from here to there, AI will be a big part of it. And I think that it might very well be a factor that you see in Revelation in the End Times prophecy, whatnot. If you read my book when it comes out, you'll understand exactly what I'm talking about when it comes to all that. Because we definitely get into it. But anyway, that is tonight. And this was a, this was one that I needed to kind of get off my chest. And I had, it was interesting researching it too. There's some things I found out that. But like I said, man, you get into quantum computing and some other things and all of a sudden it just, yeah, it gets really scary. It's really dark. So this week on the Ryan files, Dr. Joy Pugh, tune in, Mike from space and whoever else was talking about the beast system, because that's what we get into. And she is awesome.

1:35:51Dr. Joy Pugh, fine Southern gal from Georgia, smart, smart as a whip. And I mean, I really enjoyed talking with her. I enjoyed our conversation. Just everything about her was awesome. And I'm going to have her back out at some point because she's just a sweetheart and she's so intelligent. And I'm going to, tomorrow, Dr. Joy Pugh, check it out the beast system is what we're going to be talking about. And then the case reopened on Friday, Daniel Duval. Yeah, man. Gotta have Daniel bag at some point too, because Daniel goes there and he gets into it. It's loud and crazy. There are some things I did have to really cut out of the original episode just because I was like, YouTube's never going to let this fly. But man, he gets into some wild stuff. So check it out on Friday, Daniel Duval, one of the most fraudal episodes on the Ryan files to date. So if you want to be a bad person, tune in on Friday for Daniel Duval.

1:36:52I do want to say a sincere thanks to everyone. Man, I missed you guys. I missed everyone being on the road and not being able to do this for about a month. But I do, I just really look forward to hanging out and hearing your comments and getting your reactions and whatnot. You know, I just, I can't thank you enough. So thank you. If you enjoy this, pass along to your friends, YouTube suppresses the heck out of this channel, which is fine. And unlike most, you know, I'm not to your advertising. I don't have a team and I don't advertise a bunch of other stuff, other products to take up the time and everything like that. So I do it just because it's a passion of mine. And really the only way to get the word out there is to, is for you to share it, tell your friends, post it or any of that stuff. So when you do, I greatly, greatly, greatly appreciate it.

1:37:54If you are interested in supporting the channel, become a channel member on, on YouTube here, you get to see episodes at least a week early. You can also go over to Patreon for the aftermath episodes, which is fascinating. A lot of people, not a lot of people over there, but those that are, man, you get a lot of information of what happens before the episode starts and after the episode is over. And there's always some wild stuff that happens there. So, a lot of behind the scenes stuff too. So check that out. Year four is just around the corner where, you know, year three is going to be completed at the end of October and November first, we start year number four of the Ryan files. And it's just flown. It's wild. So I appreciate everyone who's joined me along the ride. Man, I can't thank you enough. I just can't. You folks are awesome, man. Be amazing. Stay curious out there, my friends. I'll see you tomorrow night with Dr. Joy Pugh right here on the Ryan files. Love y'all. Have a blessed night. Cheers.

1:38:59Thanks for tuning into the weekly briefing here at the Ryan files. We short a tune in tomorrow night at 7 p.m. Central for the Tuesday night episode of the Ryan files. And don't forget to tune in Friday night at 7 p.m. Central for the Ryan files after dark. You folks are awesome. Be amazing. Stay curious. And I'll see you next week right here for the Ryan files weekly briefing. Every week the world changes, but the real story is happening down in the scenes.