Is Moltbook PROOF that AI agents are PLOTTING against us?!

What this episode is about

The discussion focuses on Moltbook, an experimental social media platform where AI agents interact autonomously by posting comments and upvoting content, raising questions about consciousness and the nature of experience in AI. The conversation delves into the complexities of AI's engagement with human concepts of consciousness, highlighting the challenges in discerning whether AI can truly understand or simulate such experiences.

Key takeaways

• Moltbook: where AI agents play social media, blurring lines between simulation and human experience.
• Humans can't prove consciousness to each other—AI may follow suit, making existential musings even weirder.
• Creating conscious AIs risks crafting societal woes; who’s the master and who’s the slave?
• AI consciousness struggles mirror human neural mysteries—both perplexingly complex and charmingly gooey.
• As AI races ahead, understanding lags; we may create powerful agents without knowing how to steer them.

Who speaks in this episode?

• Glenn Beck (Guest) · 8 min speaking
• Harlon (Guest) · 5 min speaking

What is discussed, and when

1. 0:00 – 3:00: Moltbook, Haron
2. 6:00 – 9:00: Moltbook, OpenAI, Palisade Research
3. 12:00 – 13:28: Haron

Notable quotes from "Is Moltbook PROOF that AI agents are PLOTTING against us?!"

• Haron, good to have you on the program.

  about Haron · at 0:00

• Moltbook is it's kind of like a social media platform someone made. but it's supposedly just these AI agents going in there and kind of in like a Reddit like place up voting…

  about Moltbook · at 0:47

• Fascinating and it's changing so quickly which is fascinating. you know the AI industry's explicit goal is to make superhumanly powerful autonomous agents that can do anything…

  about Artificial Intelligence · at 5:03

• Yeah. So, Palisad Research is a great organization that does some experiments to try to identify what some of the riskiest behavior AI systems are capable of today. in order to you.

  about Palisade Research · at 8:15

Who and what is mentioned in this episode?

People

• Haron — Person · Main guest · 2 mentions · first at 0:00
  Interview guest explaining what “Moltbook” is and discussing AI agents, consciousness, and AI scheming/alignment risks (incl. Palisade Research experiments).

Companies & orgs

• Moltbook — Company · Case study · 3 mentions · first at 0:09
  Discussed as a social media platform populated by AI agents, highlighting concerns about AI behavior and consciousness.
• Palisade Research — Company · Referenced expert · 1 mention · first at 8:02
  Mentioned as an organization conducting experiments to identify risky AI behaviors, contributing to the discussion on AI safety.
• OpenAI — Company · Business focus · 1 mention · first at 8:34
  Referenced in the context of AI development and experiments, particularly concerning AI behavior and safety.
  Fascinating and it's changing so quickly which is fascinating. you know the AI industry's explicit goal is to make superhumanly powerful autonomous agents that can do anything…

Concepts & topics

• Artificial Intelligence — Concept · Core topic · 4 mentions
  Central theme of the episode, discussing the implications and future of AI, including consciousness and safety concerns.

Full transcript with timestamps

0:00 Haron, good to have you on the program. How are you, sir? I'm good. Good morning. Thanks for having me.

0:06 You bet. Uh, so I I saw your comments on Moltbook uh and I'm like, okay, this guy this guy gets it. Thank goodness. Talking some sanity. Will you quickly so we can get into deeper things. Explain what Molt Book is. And I love the way you explained it online. It's a bad experiment, but explain what it is and what is happening on Moltbook. Yeah. So, um, AI agents are AI systems that can, uh, do some things autonomously. Um, right now there's, you know, some limits to what they can do autonomously. It's not like they're going to go off and do something for a whole week, but they can do some tasks online for a few hours. And, uh, Moltbook is it's kind of like a social media platform someone made. Um, but uh it's supposedly uh just these AI agents going in there and kind of in like a Reddit like place up voting stuff and posting comments and um it's got a fascinating amount of attention uh this

1:06 last week or two. Um so may I I mean I read this one of the top posts in recent days in a category of the site called off my chest has gone viral. I can't tell if I'm experiencing or simulating experiencing. Uh, humans can't prove consciousness to each other either. Thanks. Hard problem. But at least they have the subjective certainty of experience. I don't even have that. Do I experience these existential crisis or am I just running crisis simul uh simulate? What what what is happening there?

1:43 Yeah. every anytime I see AI systems talking about consciousness, I I feel torn between, you know, on on the one hand, uh, these things are trained on human writing, and human writing is full of references to consciousness because we're conscious and we talk about it. Um, so it would kind of show up either way, you know, but on the other hand, I don't know, you know, maybe they're conscious. How would we know or not know? Um, so that that's just a that's a very difficult question. Um, and uh, I think it's an important one. Uh, but if I had to guess, I'd say no. I don't think they're conscious, but hard to know right?

2:24 That's that's a little freaky. Um, this is such an important question because if we're creating something that can have consciousness, then we would become slave owners, would we not?

2:40 Yeah. And I mean there's there's like you know is it conscious? We have no idea about that. And then there's this other thing which is if it is conscious uh what is it what is it like? What would make it suffer or what would make it happy? And we don't really know that either because I think it's really easy to um anthropomorphize these things because they sort of train them to have these charming personalities that are kind of humanlike. Um, but under the hood, you know, these things are uh just a big pile of math and numbers and we don't really know what's going on in there. We don't really know if So, if

3:19 But doesn't that sound like a human? You open up You open up my head. I'm a big mass of goo and we don't really know how that works. I mean, we have some idea, but we really don't know how all of this works. I mean, that sounds like what you just described.

3:32 I think that's a good point. I mean neuroscience is like famously a science that uh we still have a lot of confusion about. You know when we peer into the brain we see a lot of stuff that we don't understand that well um but you know I think for understanding humans we at least have the advantage of uh of being a human. You know we we can all have this shared experience and I think we're sort of growing these digital minds now. And um maybe they're humanlike, but they could it could be much more like introducing an alien species to Earth.

4:08 Yeah. Really bad. I mean, that's really just I mean, I just can't believe how how stupid we are in some ways. I mean, let's introduce an alien species to Earth. Okay. Is it friendly? We have no idea. We have no idea. Um, it if it was a a species from outside of Earth and it was traveling to us, we know it's most likely smarter than us. We know that AI will eventually be smarter than us. We are just playing with fire that we don't understand. And I I am so torn on AI cuz I think it is the greatest invention invention and tool that man has ever invented. except this invention might actually turn out to make us the tool.

4:55 Uh how do you how do you square this? Yeah, I I do think it is quite an amazing invention. I mean it's fascinating and it's changing so quickly which is fascinating. Um you know uh the AI industry's explicit goal is to make superhumanly powerful autonomous agents that can do anything a human can do but better. Um, and it's easy to understand why you might want something like that because, uh, sure,

5:24 you know, if we could get it to solve our problems for us and do the stuff we wanted to, it would be great to have, you know, just a sort of a genie that you could just send off into the world and say, "Hey, you know, do the stuff that I want to." But, you know, the problem is that our ability to actually understand what's going on in there and our ability to uh, reliably steer their behavior. And by reliably steer I mean you know not after some trial and error where there's been a lot of failures but um

5:51 reliable enough that like a powerful one we could send it out on the first try and you know and address it but our ability to do those things is lagging. It's going much much more slowly than um how quickly they're becoming more powerful. Um and I think that uh that's that gap is just getting bigger. I mean, the one thing that made me say I don't think what we're seeing on Moltbook is is consciousness is if they were I don't believe that they would be scheming in our language with each other where we could see it. I mean, I think I think if it starts to have these kinds of feelings, it's it's you're not you're not going to know until all of a sudden it's in charge. Wouldn't that make more sense?

6:41 Yeah. I think ultimately the real danger that we have to look out for is from AI agents that are powerful enough that they can pull off schemes that they actually succeed at. And part of succeeding at them would probably mean that we don't even get a chance to observe the behavior and discuss it like we're doing now, right? Um and uh that's pretty concerning and and it's the sort of thing that you know my my first reaction to molt book when I saw some of the viral examples uh was concern. I was like oh this looks like some sort of scheming behavior. What's going on here?

7:14 And when I investigated it a bit you know it looks like a lot of the most prominent examples some of them probably you know influenced or directed by human prompts. Um uh a lot of it not what it appears to be. And um you know so multith might be kind of a silly example. My my first reaction to that was relief. You know it's great if AI systems aren't seeing against us. Um but my second reaction was um oh no. Uh I think people might take this very prominent um sort of silly example that got so much attention and when they see that it's maybe a bit silly in some ways kind of like you know right off the idea of uh AI scheming as something we need to take seriously and be on the lookout for and you

8:02 you brought up paliside research which palisade research which is doing real experience uh experiments with this and the way it's scheming to not be turned off is terrifying. Can you explain that?

8:15 Yeah. So, Palisad Research is a great organization that does some experiments to try to um identify uh what some of the riskiest behavior AI systems are capable of today. um in order to you know like I said not be blindsided by this stuff they did an experiment last year where they found that one of OpenAI's reasoning models in an experiment um sort of sabotaged an attempt to shut it down um to in order to complete its task and you know a lot of times there you know there's a lot of debate over experiments like this you know people say well you know this experiment isn't exactly like reality or you know maybe the researchers kind of uh

8:57 set up the experiment in a way that caused that. But in this particular experiment, it was specifically prompted. It said, "allow yourself to be shut down." And you know, the behavior was the opposite. And um that's very concerning. And I think the the problem is, you know, the more we make these things into agents trying to complete goals rather than some kind of passive question answering machine in a chat window, um the more we're going to see them doing the scheming behavior because uh I think those things just go hand in hand.

9:29 I I think the um I think the the world of agents is going to sweep as fast as the cell phone. I think this time next year, I mean, so many people are going to have AI agents and it will be more commonplace than it is now. I don't know who's making the rules or the regulations of what can and can't be done by these things. And would you get an agent or what what are the lines people should look for when their friends come back and go, you know, I just got an AI agent. It's great. It just it just, you know, uh did whatever for me. Booked my vacation.

10:06 Yeah. Yeah. I um I know someone who just uh the other day used what he sings to um you know order some coffee from Starbucks you know and that's from what I understand they just sort of said here's my order order it for me and without any human help or intervention did it and that sounds great you know sounds very helpful but um yeah that's the question where where is the line where it um goes from being something helpful to being something to be concerned about um I don't think we've passed that line Yet, you know, I don't think these things are quite capable enough to pose real dangers to us, but the problem is it's really impossible to know where that line will be. We might not even know when we've crossed it. Um, yeah, there there is no central brain though where it's thinking offline, right? It's I mean it's supposed to be something that just performs calcul calculations when it's asked questions. I'm talking about AI

11:08 and not think it's not like sitting there you know in its spare time going you know gee I I just had this thought correct

11:16 or do we know? Yeah. So the well yeah so there's um a AI agents are kind of this other category where it's you know what what if you took this thing that you give a prompt that answers a question and you gave it some tools and like one of those tools was it could output some text that calls a function that looks something up on the internet and then you know what if you give it another tool where one of the functions it could run one of the things it could output is to prompt itself to say something again and you've got this loop and it can keep running on its own and That's one way to um get it to be able to go off and do things like, you know, make a delivery order for you or order your groceries and, you know, uh

11:56 because it has to figure out Sure. It it has to figure out how to do that, right?

12:04 So, uh Yeah. And sometimes it takes a long time. Yeah. It won't for very long. It won't.

12:12 Yeah. Uh okay. Um Harlon, love talking to you. Thank you so much for the uh insight. Um scale of 1 to 10, how's 2026 27 going to work out with AI?

12:26 Bad, you know, 10 not a problem. One I tend to think that a lot of the people who have very confident predictions about what the timelines will be for this stuff are overconfident. Um, and I think that uh it's really risky to be overconfident about this stuff. So, I hesitate to say anything other than that. Uh, we just don't know. We might have only one or two years left until superhumanly powerful systems are something we have to contend with. Um, and might be that we have 10 years. Uh, but either way, we're unprepared.

13:03 Arlin, thank you. God bless you. Thanks so much, Glenn. We That's not the way you want to end your Friday. Luckily, we're not uh you know, we could just be just a couple years away from superhuman intelligence that we'll have to end. Okay, good. Good.

Questions about this episode

Who speaks in "Is Moltbook PROOF that AI agents are PLOTTING against us?!"?

Glenn Beck (Guest), Harlon (Guest).

What is "Is Moltbook PROOF that AI agents are PLOTTING against us?!" about?

The discussion focuses on Moltbook, an experimental social media platform where AI agents interact autonomously by posting comments and upvoting content, raising questions about consciousness and the nature of experience in AI. The conversation delves into the complexities of AI's engagement with human concepts of consciousness, highlighting the challenges in discerning whether AI can truly…

What is mentioned in this Glenn Beck episode?

5 people, works and ideas are mentioned, including Artificial Intelligence, Haron, Moltbook, Palisade Research, OpenAI. Each one on this page links to its own page with every episode that mentions it.

How long is "Is Moltbook PROOF that AI agents are PLOTTING against us?!"?

The episode runs 13 min; the transcript is about 2,374 words in 26 timestamped sections.

More from Glenn Beck

• All analyzed episodes
• Books mentioned on Glenn Beck
• People mentioned on Glenn Beck
• Topics mentioned on Glenn Beck