Congress Targets Rogue AI Agents With New Crackdown

Hey everyone, happy Tuesday and welcome to Techstrong gang. I love starting a week on Tuesday, and I'll tell you why. First of all, that means I had a long weekend, and Monday was a holiday, and I did this weekend.

It was a great long weekend. I hope those of you who celebrated Labor Day enjoyed the weekend as well. You know what gets me though, and maybe it's just because I'm in Florida and Florida men and women live here.

I think they've confused it with Memorial Day. They put up all of these American flags and thank people for their service, but that's not really what Labor Day's about. Labor Day celebrates the American labor movement, which doesn't get enough play in this country.

But I'm done pontificating. It was great having Monday off, and starting Tuesday means we're one day closer to Friday. So on both ends of the spectrum there, that works for us.

Let me welcome our gang excuse me. Let me welcome our gang for Tuesday's show. We have joining us up in Canada still, it looks like, my friend Chris Blask.

Hello, hello. Hello. New Hampshire, where she could just throw a rock into Maine, or is it the other way around?

The one and only Kate Scarsella. I'll get it straight one of these days, Kate. Also joining us, he's been on the road lately, we haven't had a chance to have him on, our friend Sid Nag.

Hello, Alan. Good to be back. Good to have you.

And joining us at an ungodly hour at his house, our friend Alistair Cook. Hey Al, how are you? Yeah.

Good, Al. Good. And of course, holding it all together from his pinnacle up in Harrison, New York, the polo aficionado, Mr.

Mike Rizad. Hey Mike, good weekend? Good weekend.

I went to my first polo match ever. I'm still thoroughly confused, but it was a good time. Very cool.

All right. Let's jump into it. It was a busy long weekend.

Lots going on. Mike, what are we covering today? Well, these rogue AI agents are the talk of Washington apparently, and there's now been a bill introduced by both a Republican and a Democrat in the House talking about how we need best practices and things to secure these AI agents.

This comes on the heels of maybe a little more stringent bill from, or call for a bill at least, from Bernie Sanders who says anybody who unleashes these things should be severely punished. I think these two bills maybe are two ends of an extreme, but I'm going to start with Kate here. Part of my soul here is kind of like, well, we're making up a law that requires people to go spend money to secure themselves from something that they didn't create and that somebody else kind of built and unleashed.

So what's the right path here? So first, it's technically called the AI Kill Switch Act, and it would allow the DH secretary, in acting through the CISA director, to order the company to throttle access, restrict capabilities, and suspend inference or shut an AI system down. And my first thought here was, wait, Homeland Security?

Are you kidding me? We already have given them, in my opinion, way too much control to do things, and we're yet giving them access to understand technology which nobody seems to understand in Washington? 0 started in 2020.

What is it? Human machine teaming. Are we getting it yet?

Meaning that we have to understand what progress is. This is very important for us, and we can't just continue to have... It's not that we don't need regulatory laws and everything else.

We do. But we continue to have it created by people who don't understand technology. That's the main point here.

So creating a kill switch for this is just drastic measures. Bernie Sanders wanting us to stop altogether. We need to stop, and we need to actually get people in who understand.

And you brought up last week, Mike, about the open source community and things like this. We are. There are people engaged in this, right?

And then Chris Black. We have so many people who are engaged. So at the end of the day, do we need regulatory laws and things?

Yes. But not by the people that are creating them, and certainly not by Homeland Security, which I can't even imagine that somebody thought that was a good idea. So anyway.

Those are my opinions. Come on, guys. I know that there's a lot here.

We see this sort of thing in early technology phases. So in my experience with cybersecurity in Washington and public sector since the early '90s, right? You're not going to have legislators understand bleeding edge technology.

That's not the way this works. We're actually arguing amongst ourselves and what it means, those of us building. Right.

So on the one hand, yes, everything you said, this is a big blunt stick to solve a really complex problem that the people trying proposing it don't understand. Right? What does rogue even mean?

How would you define this? Yeah. And as you say, putting it where it is in the US federal structure is problematic.

So. Yeah. Sid, go ahead.

No, I was going to say, anything that the current Department of Justice mandates is always something to worry about . Yeah. So I'll start off with that.

But I think- Well, Justice being the oxymoron there, but- That's right, Alan. Yeah. I don't want to say it in that exact word, but looking at the bigger picture, just taking away the politics out of this, I think enterprise AI, and I say enterprise AI purposefully because it's not about cute ChatGPT prompts by consumers.

But enterprise AI domain has a governance gap, not a security gap, right? And I think security leaders believe that agents have appropriate access, but only 33%, I think in the article says, that actually these agents consistently enforce those privileges, right? So I think that gap is something to be worried about.

And I think furthermore, it goes back to the conversation we have been having all along, which is the effectivity really starts when we define an enterprise-wide control plane for agents, right? So I think that's the other observation I'd like to make. And then I think we almost need an analog of what we did a while ago in the space of zero trust, right?

For agents. So default dynamic permissions, scope credentials, network isolation of those attributes that constituted the framework of a zero trust definition. I think we need something similar for agentic and for agents, and the control plane should be the overlying architectural principle for that.

Well, you say governance is everything, right? And just looking at the AI experience everyone has, as you say Sid, it's not about chatbots at the enterprise scale. But look what we do with chatbots.

I'm not looking at anybody in the screen here, but if, for example, one of us created a text artifact using multiple AI agents and took that into something like Microsoft Word and saved a PDF, can you track who the authors are even through that? Mm-hmm. What pieces?

And we're not doing that now. Yeah. Well, but supposedly you can with the new watermarking that Claude is using.

Oh, absolutely not. That's a whole segment by itself. That's not the point.

The point is we're not using systems now to leave receipts for how they connect at all. So, watermarking from a given model is not the solution. The solution, as you said, Sid, is the control plane.

What is the governance model? Do we have artifacts to even make governance decisions on, or are we skipping that entirely? Yeah, we've talked about this whole-- Sorry, go on.

Go ahead, Sid. No, no. Go ahead, Sid.

We talked about this a few shows ago, right? Applications of today are not the same as applications of yesterday, right? Yeah.

And applications of today are what we call autonomous software. And when you have autonomous software, to Chris' earlier point about the Microsoft Word example, that you shift that whole deployment model of autonomous software from legacy software, if I call that, to autonomous software, where agents are doing all the work. That becomes a very nightmarish scenario, right?

Sorry, go ahead. No. So look, I think you got to bring this back to what the topic was.

And you can't take politics out of politicians, right? Hence they come from the same root word, politic. Probably Greek, spray some Windex on it.

But that's what we're talking about here, right? We're talking about politicians playing politics with something that is really, really important, right? Yeah.

And there's so many disturbing aspects of this. " Pull your panties down from around your neck and put them back in. I have an article up on this with a great graphic of this with him.

You're not. And we make fun of one politician thinking he's somewhat of a dictator. What makes a senator think that because he could pass this bill, the people in China are going to tremble and not use AI anymore according to his wishes?

Our politicians have not quite caught on to the fact that the internet knows no borders, that when we try to impose our politics, our laws, it works maybe well within our borders to an extent, but it's no different than China's great firewall of China thinking you're keeping everyone in China off of the greater internet. It don't work like that. Secondly, the knee-jerk reactions.

Oh my God, the agents invaded and got Hugging Face. Did you read how they did it? They sent 1,700 messages.

We got to do something. I'll write a bill tomorrow. Yeah.

That's not how this stuff gets done. Well, I usually like to write the bill. What are you talking about?

But that's the problem that we're having. That's the problem that we've had with cybersecurity strategy. Hey, I've been doing this for close to three decades, and the problem that we're having is that we continue with the whack-a-mole strategy.

The idea that there's vulnerability steps up, we create the product that's going to deal with that specific vulnerability. Thus, you walk into an organization, and they have over 200 cybersecurity products with 300 vendors, and it's like this is a nightmare, folks. Totally.

And we're unleashing on with the nightmare. But let's come back to this point here. We are unleashing AI agents, and we all agree that we don't have the governance capability to manage them.

So what is the right path forward, Alistair? What's common sense here? Well, I think you need to recognize that the AIs have no morality.

They are amoral, and so any kind of judgment of whether this is a good thing to do has to fall to a human. And fundamentally, this is where our problem is. The humans are making poor decisions about what's a good thing to do.

They're favoring speed of execution over safety of execution. And so- They always do though, Alistair. It's something that can happen ...

absolutely it's human nature. Yeah. It's totally human nature.

And the same way- But there is another way. There is another way. Look at what the EU's done, right?

They said about like a six-month or a nine-month study, they brought in some real experts. Yeah. " Like they do with a lot of their tech legislation.

" What's so bad about that? No, I think it's perfectly- Why can't we do something? We can't do it because honestly, our little tech bros continue to run the government and pay for whatever they want to do.

That's the reality. And even like with these bills that come out, at the end of the day, it's not going to impact them in one way or the other. There will be no kill switch at the end of the day.

And the idea that the government... And think about that. The government has the kill switch?

That's what you're talking about. If they don't like your AI, we're giving them the power to say, "You can lock us up," is what you're saying. Yeah.

This is so not cool. But on the other end of this extreme, you've got a scenario here where we don't have control over the AI agents. We've got a federal government or a presidential arm that's basically abdicating any kind of regulation or movement, and an AI industry that's divided against itself, with half of it calling for, "Please stop me before I hurt somebody," like regulations, and the other half saying, "Well, screw it.

" But because Mike-- I'm sorry, Kate. The problem is, yeah, you've got Dario Amodei every couple of months, he has a pang of guilt or whatever and says, "We got to stop. " And Sam Altman says, "Well, I'm always--" But they all know that's theater, that's political theater.

Because it's not just the US frontier models that need to pause if you're really going to do that. Right. You've got to go get the other two million models on Hugging Face to pause, right?

You think Moonshot's pausing? Not unless the Communist Party tells them not to. Yeah.

Right? So- One other thing- Go ahead, Sid. I think one good thing could come out of this, and I'm saying that with a skeptical viewpoint, is that if we-- You're right.

Models are going to be developed in all parts of the world, right? Whether it's US, China, wherever. But I think we still have an opportunity as the leader in technology to some degree, to show the way, that we being the US, right?

And if we have governmental agencies or bodies like NIST establish standards and guidelines around continuous monitoring, agent reliability, tamper-resistant action logs, and those kinds of things, that just might be the thing we need. Again, I'm not for big government, but I have a standard body because one of the things- Mm-hmm ... I found in the cloud world, unlike the development of the internet when I was active in the IETF, that when we do standards and vendors adhere to it, things actually work.

Absolutely. Yeah. I agree.

And I'm all for that. There's a place for government, there's a place for regulation, there's a place for these things. Yeah.

But a knee-jerk reaction saying, "I'm sending you to jail 50 years if you use what I think is advanced AI" is no better, in fact, maybe even worse than the other extreme, which says, "Well, let anything happen. If you don't want a data center-" Right ... " Yeah.

That's where our government is right now, here and there. Yeah. And we're all tired of it.

Where's all this? We're all tired of it. All of us are tired of it.

We are. Yeah. Fed up.

Goddammit, I'm tired of it. But IETF and NIST are not about government, it's about- No, it's about standards bodies and doing things. Yeah.

Consensus and running codes, right? So if you can get rough consensus of running code in the world of agentic and we have some guardrails, I think that might just be a good thing. Absolutely.

Mike, I'll tell you what's a good thing. We need to hop. We move on, because the problem though is that nobody trusts either the government or big tech, and it's a mess.

But there we have it. Well, have they given you reason to trust them? But go ahead.

Let's not go down that side. This isn't going to help. We are going to talk about a little more of the government, and the federal government has weighed in on a copyright lawsuit that New York Times has filed against OpenAI, and I think Microsoft is also listed in this.

And the federal government said something interesting. They claim, at the very least, that training of the AI models is of a strategic national imperative, and if it happens to suck up a bunch of copyrighted content, well, that's just the cost of doing business for national security. And all the folks that create that content should just, well, suck it up, buttercup, on behalf of the greater nation.

Alan, I know you've written about this. As somebody who runs an outfit that creates content and helps pay for that, what could go wrong with this? What could go wrong?

So Mike, I talked about this on my "Shimmy Says" segment last Thursday. I wrote this article, and I wrote a full-on tech, strong, special report that goes into a lot of the details and reasons and legal theories behind all this. This is really a crock, if I say so myself.

Right? Let's look at this. You had the Anthropic lawsuit, and it's funny how Anthropic seems to get a different legal bar than OpenAI does here.

We know who the home team is. Right? Maybe they gave more money to the reelection campaign, or rebuilding the Kennedy Center, or the Arch, or something, but OpenAI clearly gets preferential treatment versus Anthropic.

5 billion. Since that time, The Times, and a bunch of other publishers, have brought suit against OpenAI and Microsoft with similar claims. We discussed a couple of weeks ago, Sony Music and some of the others have brought claims, I think, against Anthropic there as well, that, hey, they took music off torrents and all of these things, and in essence, they stole the IP.

This is what lawsuits are made of. Sugar, spice, and everything nice. Right?

This is a great lawsuit. It's a private lawsuit. It's me, mano a mano, me versus the AI.

Now, all of a sudden, though, our government feels it necessary to come in here and put their finger on the scale, in the case of this case, by filing, in essence, an amicus curiae brief, if we're under the court brief, stating that, hey, this isn't just mano a mano. We've got an AI gap bigger than the missile gap was with the Soviets in the '60s. If we don't do this, China is going to beat us in AI.

You've got to let them use everyone's IP address. It's for the common good. There's some parallels to this, right?

World War II, any war. We had to turn the automobile factories into making tanks and airplanes. What were they?

The thing of democracy. Oh, Mike, you probably remember the term, right? We were the arsenal of democracy.

That's the term. I don't remember any of those companies working for free. As a matter of fact, they made out buku bucks turning out planes and tanks for the government.

No one gives their work for free for the common good. So now you're asking The Times, Sony Music, and Alan Shimel to give my IP out for free so that I can sleep at night knowing the Chinese Communist Party doesn't have a better AI than me. Nonsense.

Are you kidding me? I'm not kidding you. That's what this boils down to.

They're saying it's okay to do that. All you content IP copyright holders, tough. Thank you for your contribution to making America great again.

This is not a legal principle that flies. I'm sorry. I agree with Alan on that one.

I think it boils down to data provenance, right? You just can't take people's data and start doing crazy things with it with no copyright credits. Where do we end this?

Well- So- It's right where you and I were saying in the last segment, governance. Look, national security is real. Nation-states need boundaries, and I'm a career artist.

I understand that. But it's not a universal solvent, right? You can't throw it at everything and say it's all the same thing.

And if you have no provenance, no governance of anything, then you can say it's all for national security, but that's not an answer. But even if it's for national security, you're still entitled to be compensated for your property. We have eminent domain.

There are hundreds of years predating the US, going back to common law, the king can't come and take your land unlawfully without compensation. This is a tenant of law. You can't just decide, well, no, we've got a missile gap.

This is a simple problem, and I don't understand why we just don't come to the conclusion. All right, simple. Publishers such as Alan Shimel should receive a tax credit for all the articles that they create that are then used to train- I want money.

I don't trust their... They're $30 trillion, $40 trillion in debt. I want cash on the barrel.

But we get back to provenance and governance, right? We don't have systems that will even let us know that Alan Shimel's- And that's the other thing. There's no transparency here.

I don't know what they took or didn't take. We can only assume, because I'm not big enough. And again, the systems don't work that way.

We do not at any level, in threat information, intelligence sharing, and so forth, we don't embed enough provenance to have any reasonable idea of where anything came from, and now we add AI. So that's how we get AI. I think we need transparent provenance.

We need lawful acquisition mechanisms in place. We need things like auditable usage. If you have all of that in place, and then we build a layer of scalable compensation mechanisms, that's fine.

Use my data, but pay me for it. I have no problem with it. And what I don't understand is at the end of the day, why don't we actually start creating laws around that?

Do you know what I'm saying? We talk about in Block A, having these crazy laws that don't make any sense, yet when it comes to this type of common sense stuff, it's like, yeah, no, we don't have laws here either. So we're not going to have laws here, but we want kill deals.

Do you know what I'm saying? It doesn't make sense. It's insanity.

There's no end. Wait, Alistair had something, Mike. I did.

I was going to point out that this is an incredibly stupid law because it says harvesting your information is important for national security. But it doesn't address the fact that those models that are being built are then being used by the very state that we're trying to be secure from to build their own AI. And so this law is addressing completely the wrong place.

It's not addressing national security because it's not addressing the misuse of AI tools by threat actors. It's not addressing a place where something bad is happening that needs a change. org site for training stuff where they're going to compensate individual writers for creating content that will be used to train AI.

It's called fair play Well, yeah, a useful law might be something like, if you're going to make tools that other people will use for authorship, those tools shall have and will implement the abilities to track and trace and provide the hooks for the author to maintain continuity with their bloody work. Yeah. That can be done.

That's easy. But over and over again, we the people here in the United States continue to pay this bill. We are continuing to pay the tech bros, these whiny tech bros, to go and steal information, and yet we again pay.

And we pay, and we pay, and we pay. Our government is about we the people. What's happening?

That sounds like a Tea Party thing, we the people. But let me again go back to the law here. There was a time in the internet, early internet, and some of us who are a little older here may remember it, not looking at anyone here on the panel, of course.

But some of us may remember where there were a lot of people who thought anything on the internet was freely available. Of course, I could take that picture. Of course, I could set up a Napster and distribute all everybody's music.

Of course, anything that's written on the internet could be copy, pasted, and reused. And we spent about 20 years developing case law around how you could use information on a website, what copyrights flow with Creative Commons, and all of these kinds of governance and usage formulas. And now because the tech bros are desperate to get training material to train the next model, we're going to throw that all out the window.

Why? There's a difference. The internet was a loosely connected set of computers in a democratized fashion.

Okay? The model business is a very tightly controlled moat by one or two companies. That is the problem.

But that makes it even more compelling, Sid. Because even that loosely confederated network of computers, laws came down that said you can't just take people's property. There are copyrights involved.

There is IP involved. Even more reason now when it's not a loosely confederated thing, it's actually a company building a moat. Why?

Hey, you want to build the moat? Pay for it. Because in the other model, people would not do it because the people that were stealing from the other guys were small people.

So then, the guys will- No, look at Napster. Look- ... protest and file a lawsuit and win.

In this scenario, you've got the Anthropic and the OpenAI form the moat, and they don't have the balls to go and do this with "New York Times" or Hollywood because they know they'd get the pants sued out of them. But they're going to do it to people like you and I. That is the problem.

Yeah. " And now the government is coming in on top- Yeah ... " Because what we talked about earlier, it's like they spend the time in the Oval Office currying favor to the- Spending time isn't the issue, it's spending money.

Right. Yeah. Money too.

Yeah, of course. Well, you would think that then we would be cheaper, right? And I think we, like you, would probably be cheaper than the president of the United States.

What are you calling me, Kate? I think we'd Napster for 10%. What are you calling me, Kate?

She's calling you an entrepreneur. I know what you are. It's just a question of the price.

She's calling you an entrepreneur. Okay. Thank you.

Thank you. I think we're going to be out of it, Mike. Let's move on, Mike.

We got another segment here. All right. So New York City has decided with the mayor, and I believe the city council in tow, that they're going to ban AI from education classes for I think below eighth grade.

And the idea here is that they are saying that the problem with AI is that the students will not develop the ability to think, and they'll just kind of sit around and maybe copy and paste too much, and they won't understand how anything was created or what the logic that went into that was. Sid, I believe you live in New York City. So, what are you hearing from folks?

What are the thoughts? Because honestly, the thing that leapt out at me most about this is there wasn't as much a human cry about this from the AI community as I might've thought. Yeah.

I think this is a great move. I think New York is leading the way again for the rest of the world like it always has in making an important distinction between AI literacy and AI dependency. Right?

It will prohibit students from facing generative AI for roughly 600,000 students through eighth grade, which is a good thing, but it is permitting restricted use in high school. So there's a difference. When you read in the headlines, you got to read the real details behind it, right?

And I think it's a good thing because excessive reliance on AI, like you said, could erode critical thinking, creativity, and independent problem-solving during the early formative years. So I think it's a very good thing. You guys should amplify developed cognitive skills, not substitute for developing them.

So I think that's sort of my point of view. But this policy could also become a very important test of whether AI actually improves educational outcomes or not. We will know at the end of the experiment, if you want to call it that, whether people can still learn and grow and still have the ability to think on their own versus being dependent on AI.

And it doesn't kill the education market for the AI business. I think it raises the bar for vendors to be more careful in how they position these things and how these things are allowed for students in that stage of their growth to use AI effectively versus not replace their brains with AI. That's how I look at it.

Well, I'm glad that you know some of the detail of it because I see a ban, and I think Wikipedia. My kids are 24 to 31 years old. " Which is ridiculous.

As hopefully everybody knows these days, Wikipedia is what it is. It's a very rapid way to sort things out. And no, you can't actually use it as a primary source.

Same with AI. A pure ban, if it really was an outright ban, then I think it's ridiculous. But the way you said it, Sid, I could live with that.

My question is, what are the details? What do they mean specifically? Yeah.

I think it's tools versus the replacement theory. You can use AI as a tool, but you don't want to use AI as a replacement of the developing brain of a child. You can use AI as a mechanism to do something or fine-tune something in your high school years, when you're writing an essay for an application to college.

That's totally fine, I think, from just what I've read about this. But you don't want to unleash AI in a manner where people just become dumb idiots. That's, I think, the crux of this whole thing.

Well, then they move down here. It's no big deal. Anyway, so here's my problem with it, though.

Again, you got the kind of politicians doing science. I get public policy, but public policy is supposed to be based on sound scientific principles. Where is the research that shows using AI before eighth grade turns your brain to rot?

Where is it that shows using AI at all turns your brain to rot at any age? I don't know. One word answer, Instagram.

Instagram. Well, there you go. So that's my first thing is I would feel more comfortable with these kinds of things if there was some science behind it.

Number two Let's think about it. When I was growing up, if you didn't eat your vegetables, there were children starving in other places in the world. When my kids were growing up, they had to do better in school because children in India and China and the rest of the world were going to school 12 months a year and working a lot harder than them, and you're not going to have a job when you grow up.

Who knew it would be true, but it wasn't India or China. It was AI that was going to take it, but different story. Right?

Are we putting these children at a disadvantage because they won't be, let's call it AI native versus kids who are AI native as they get older? And so these kids who are not exposed to AI have some catching up to do. Yes, maybe, maybe not, their brains are sharper from not using AI.

" Because figuring out cosines has been very important in my life. So is that what this is? I think, I'm also going back to my 11-year-old self, and if you tell me that I can't use something- You're going to ...

I'm going to go find it. And so, I think we're probably better off letting them understand through usage that there's crap in there and they need to sort it. But at the end of the day, I'm not sure that a ban is, or even at that level is enforceable.

I don't know, Alistair. Kids are kids. They're going to find stuff, and they're trying to ban social media, and as far as I can tell, that's not going so well.

That's absolutely failing from everywhere I've seen. My perspective on this is that education needs to change. So one of the challenges here is that AI is very good at solving the kinds of tasks that teachers have set students in the past.

Mm-hmm. It's not good necessarily. We know it's not good at every kind of task, and that deep thinking, that critical thinking element is vitally human, that's really hard for an AI to replicate.

Well, we need to teach people to do that, and that's a change in how education needs to be structured and delivered, and that is always slow and difficult. So there's a very strong self-interest here of saying that 45 candidates running for the Chicago school board, sorry, half of them, said they're in favor of bans on AI. Well, that's a very strong interest.

These people are already invested in the way education has been run for hundreds of years. By the way, remember the days when the typewriter was destroying people's penmanship? And remember also, going on these hand-wringing about new technology.

So after Chris's Wikipedia thing, it became screens because kids had iPads, and that- Yeah ... was rotting their brains, right? Actually, when I was growing up, it was TV that was rotting my brain.

So there's always been something in here that people are blaming that is new. Every change brings the same idea. " Yeah.

These things, like the cosines, are not particularly useful in my future life. If the test is easy to cheat on, then the test sucks, right? Yeah.

Well, those kids and their phonograph records, right? And I'll take the far edge of this one. I do believe that one of the promises of AI that's true is massive upgrades in education worldwide- Me too ...

for all children in the relatively near future, right? Yep. We're right at the spot right now, it's a bit of a mess.

But continuity, AI systems have continuity to individual children, personalized. The difference between a tutor when you're very wealthy and public education. Exactly.

And those sorts of things are right here at our fingertips. And at the same time, yeah, the tools and the companies right now, I wouldn't trust them as far as I can throw them. But that's real.

It really can be done. I agree. I think, again, it goes back to an earlier point of tools versus building certain things in the fabric of the growth of the children in school.

So if we're going to adopt AI, we got to revisit the education curriculum, right? We can't just slap AI on an existing curriculum that was built based on historical education mechanisms, right? That's my point, right?

No, I think we need educators. Mike, you had sent me an article over the weekend that I read, I think it was on "The Washington Post," about a particular teacher who instead of resisting AI, embraced it. Yeah.

And it's that kind of thing. I'm sorry, go ahead, Kate. No, and I think to this point, we have to understand, right now we are at human machine teaming.

What does that look like for us? And that's so important to understand. And that is what is so imperative right now to understand.

Human machine teaming, what does it look like? Very interesting. And we pick on the AI a lot, but let's be really clear.

One thing this particular form of AI, large language models, are really good at is narrative coherence, right? And we're skipping over the fact that education systems aren't perfect. We all know that.

We can take it in business and infrastructure, but just being able to take an entire educational system and look at it all at once. We've never been able to do that, right? And again, it's not about the LLM.

With children and education, individual children and education, it's about records and memory. Being able to navigate all of the records about each child's experience, we've never been able to do that either. So if we can do that safely and reasonably securely, it opens up possibilities every parent always wanted.

This assumes that the AI has not been tinkered with and that the AI model isn't promoting, shall we say, pick a thing, creationism over science. Or some religious aspect of how I view the world over science. So none of the AI is infallible, and it does seem to me, I've already seen that there are people who are using various AI models to advance their own agendas, training it in a specific way.

And make matters more interesting, the people using those AI models are definitely developing a crutch, and anything that goes against their established belief, regardless what the AI says, they're going to ignore anyway. Well, but I'll say it for the 57th week in a row, though, it's not about the models, right? Children live longer than models, right?

Lumina, my little system, is just a year and a half old. We've gone through all sorts of models, right? It's not about the models, it's about the memory and coherence across that, but that's another segment.

Whatever- You're not wrong. Models are- The root cause of the thing, whether it's the memory, the model, or the harness or whatever it is, can be tinkered with in ways that is not necessarily infallible. So somebody's got to sit there in an education system and say- But Mike- Same with paper, though, right?

Exactly. Let's not let perfect get in the way of good. Mm-hmm.

On that note, though, it's a perfect time to end this show because we're out of time. Chris, Kate, Sid, Al, fantastic conversation today. Thank you.

Thank you all for coming. Mike, as usual, great job. " You can catch us at noon or wherever you're watching this now.

Or if you're watching this on demand, of course, on our Texture on TV, Texture on OTT channel, Texture on TV YouTube channel. We look forward to seeing you. Until tomorrow, though, this is Alan Shimel.

We're out.