Techstrong Gang – March 11, 2024

Hey everyone. It's Monday and you're watching Text Drunk Gang. It's gonna be an AI Monday.

We've got robots and everything else. We've got a great panel. We've got one of our special gang members joining us.

Stay tuned for Textron Gang. Hi everyone. Welcome to Monday Textron Gang.

It's an AI Monday. We've got a lot of good stuff to go over. Um, there's just so much going on in ai.

There's always good stuff there. There is like a, the, the, the gift that keeps on giving for, from a media point of view. We're gonna be joined by a great gang cast today.

Let me lead it off as usual, from out in Colorado, the men in the library, our CTO and, uh, principal analyst, Mitch Ashley. Hey, Mitch. Welcome.

Hey, happy Monday. Great to have everybody here. Absolutely.

Good to see you, man. Hope you had a great weekend joining us from the, the city of, is it still the city of Brotherly love? I don't know.

That's what we like to think. That's, that's what We're still pushing. That's okay.

You're still selling that despite Media reports to the contrary. Absolutely. Uh, okay.

Joining us from Philadelphia. com and Security Boulevard and more our own. Sharon Florentine.

Hello. Great to be back. Thanks for having me.

Pleasure to have you on. And joining us is one of our Roving Gang members. That sounds almost illegal.

He's out in the West coast today, the one in the only Bal loop, John Willis. John, welcome. Hey, good.

Hey, you know about the city of Brotherly Love. I heard a term I hadn't heard in years. The Broad Street bullies that Came up there.

Oh, Dave Schultz and the Philadelphia Flyers, Gary Clark, that in a long time. What a smile Gary Clark had. Um, we're Coming back.

It's our year. It's our year. It's your year.

And then last but not least, my sidekick here in our Boca Ratone headquarters at Techstrong, our Chief Content Officer from the hailing from the Bronx, New York. Mike Baard, Not in the Bronx today, but happy to be here. Oh, Hey, Mike.

It's, uh, as I said, it's an AI day kind of day today. Every day's an AI kind of day, but Yeah, well, if you count robots as ai, it's in, Yeah, it's in there too. All Right, well, let's just jump right in, and I'm going to go right to John immediately.

But, um, we saw this week or past week, Microsoft AI engineer sent a complaint to the FTC saying that heavens, to Betsy, the AI engine was creating disturbing images. And there's two aspects of this. He's charging that the safety protocols are not being observed, but he is also, we'll talk about copyrights and all kinds of fun things that go with that in the second part of that conversation.

And this comes on the heels of Google also stumbling with their Gemini launch. So John, you've been watching this space for a while. What's going on here?

Are these models going crazy or what? Yeah, there's a genie in the bottle problem. Right?

And, you know, um, the, you know, the thing that, um, all these, you know, the all we take all the good, right? The good is that we've, they've trained enormous corpuses of data to get us all these incredible results, right? Um, you know, the, almost all of the internet, uh, images, image recognitions, all this stuff, but there's a bias in the system.

And so, and, and that is the sort of problem space that everybody from an organization that's taking their own corpus of data or your sort of Microsoft or your Google, um, you have to constantly monitor this bias. And, and so I was looking at, you know, I was reading the, you know, the article at Microsoft and, and the, um, employee who was, you know, exposing, you know, the, what's going on here with some of the image generations and, and copyright issues. And I looked at, uh, Microsoft's responsible AI standard too.

And, you know, it's, it's like if you read it, it like, it's either just blah, blah, blah, blah, or it's so complicated. And I, I was reading a book by this woman called Joy Balum Winnie, and she's written a book called Unmasking ai, and she focused on racial and, and gender bias. And, and that's not exactly what's going on in Microsoft, but the point she makes is somehow in this muck of all this stuff and all this generation that's very probabilistic and, and non-deterministic, right?

There's no prime. How do we stop if we tell a, a large language man or a large image processing system to go read every image on the internet? Oh, and now, by the way, since you're doing that with some mathematical magic, don't print Disney pictures, right?

Um, and, and her point was you have to have meaningful transparency. And I thought about that, right? Like, if I read Microsoft's, and I'm not a lawyer, I'm not an expert, but like, it's, it doesn't seem meaningful to me.

It's like seven pages of blah, blah, blah, blah, blah, blah. We're gonna do this, blah, blah, blah. And I don't know what, you know, I don't get paid the gazillions of dollars to the people who are marketing market at Microsoft, but I think they have to answer Joy's question is how do you provide meaningful transparency?

And that's the hard problem. And the second part is, you know, this code of continuous oversight. So they could argue that they have a standard for responsible ai.

It's in the documentation, you know, and therefore we've done our part. But what I think they're not doing, what, what Joy, moon, and I, I always have a problem pronouncing her last name, um, is continuous oversight. And so, and this is what I'm doing with a lot of my clients right now, or I'm helping them build their, their own corpus of data.

And we talked about this last time in vector databases, you have to have an end-to-end process that's constantly monitoring the toxicity, the, the bias, the, all those things. And if you're not doing that, you're not being responsible. And by reading that article, Mike, it sounds like they don't have a process for that second part.

But I, I, I got a problem here though, Uhoh, This whistleblower from Microsoft, the last time I checked, they owned a sizable chunk of open ai, which is the, i the people behind this image processor. Did he have to go outside the family and tell him what he was thinking? Right?

Yeah. Why couldn't, why couldn't he just say, Hey, we got a problem here, guys. Well, he did, apparently.

'cause if you read he did. Yeah. It said that he talked to Microsoft and those guys, and I guess apparently he didn't like the response.

So he sent a separate letter to the FTC. Well, he Should find another job. Yeah.

Well, that's probably gonna happen at this point. Um, No, no, but Alan, I do think, I, I think the point is, you know, I how do you solve that transparency problem? Or how do you solve the bias of these engines that people don't even know?

They can't, the people who wrote these things can't even explain how they work now. Mm-Hmm. But, but, but the, but you have to at least show some evidence of continuous oversight.

And I think if you read his article, it's clear that there's nothing in place to at least try to say. I, I think going to the FTC, it, you know, I mean, this guy pulled the fire alarm in school. I, I, I, I would, I can't imagine if he brought this up to the powers that be that, John, it's, if it's so obvious to you, why isn't it obvious to them?

Uh, life is full of examples where CEOs didn't pay attention to what employees were telling him. So who knows what the deal is. But John, to your point, um, let me just add one little thing here.

I think you're kind of saying that the, the licensing terminology used around these services is kind of so dense as to mean almost be meaningless. I mean, strongly cook. Yeah.

No, it's not licensed. I mean, they, you know, I mean, and here's the thing, right? I think Microsoft has been, you know, anybody you talk to about, you know, enterprise use of responsible ai, they'll point you towards, um, Azure Open ai because they've done a much better job of at least saying and saying, we're gonna do certain things that, because we know a lot of our customers are gonna be enterprise customers, you know, that even open AI themselves don't do, right?

Um, so this is really more not really a legally binding, it's like we have put out a standard for responsible AI at Microsoft, but I'm saying is it is, when you read it, you, you start getting a gaze after about the, you know, second or third page. And it's just hard to say, you know what, what Joy Bloomie's saying is, I just want transparent ai. And so if I'm Microsoft and I'm billions of dollars and I'm doing this, maybe I should sit down at the top and figure out what does that mean, and how can I tell it in such a succinct way?

Not that you can solve the problem, because it's, it is a almost an impossible problem to solve. I Mean, you, so two things. Yeah, go ahead.

If I can just interject here real quick, I think there's two issues that I, I feel like underpin this whole thing. And the first is if they're, they're building this corpus of data, right? But like, where is that internal stuff or that's just external garbage.

It's, it's the garbage in, garbage out problem. Like, you put everything you can find into an AI model, and if 75% of it is garbage, and that's what it's learning from, what else is it gonna spit out? If you're just sucking in great amounts of image data from the internet, which we all know is flooded with it, of course it's gonna put out these horrible, biased, toxic images.

So I, I agree that there has to be some kind of oversight on what it's, what it's pulling in. And yes, there should be some kind of way to flag that and say, we don't want X, Y, Z, P, and Q to even come in in the first place for the engine to churn through. I don't know how you do that.

That's not my area. You Know, Sharon, what, what puzzles me is it seems like we're not talking about the metadata around these photos that it's ingesting. Mm-Hmm.

So, you know, in prompt engineering, it's all about the context that you're adding, right? And I don't know that people are giving great context in these image generations, but many of the photos on the internet at least have some, some tags on 'em, some keywords, some information about 'em. And it, it was just kind of, we're, we're opening it up.

And I'm not proposing anything like radical, like a rating system or something like that. But, you know, you, you even, you know, if you look at a violent game, it'll give you a, a warning. Do you wanna look at this before you go to the next thing in Steam?

Or you, if you're gonna go to this page, it says this kind of content on it, it, it seems like add in some machine learning and some meta metadata with the Gen ai. And you know, now you can start to recognize what, what things are and how to categorize it, problem in terms of a, but what's acceptable to use to your transparency point, Mitch, here's the problem, right? The problem is, at this point, all that data is unsupervised, right?

Mm-Hmm. So we've crossed that chasm from sort of labeled data early on, all the images and, you know, sort of ImageNet and all that're being built, labeled. But then the breakthrough was this unsupervised deep learning.

And the pro, one of the problems is, you know, you know, like the idea that you can say, create me, um, a, a picture that has a stream, it has three green trees with Purple Mountains and all that, and it just takes all these images and generates it's non-deterministic. It's probabilistic, right? Mm-Hmm.

The, and that's a lot of overhead in that, right? You know, to do that. And now, one of the answers I'm guessing could be, could I then add a back process to say what's in that image and that technology exists?

'cause it, it goes both ways. But then the cost, like if I'm using Dolly and I say, you know, generate me a Disney character, there's no nothing that's labeled or meta about that. I mean, I guess if you say Disney in the prompt, yes, but, but like a, an image that's gonna sort of turn out to be, you know, offensive or something like that, you'd have to reverse engineer that through the probabilistic model to say, oh, I think this picture is a copyright infringement.

And I don't know that anybody wants to incur the Cost. Well, there, there's a second part of this thing that we gotta get to though. So, um, he's also alleging that Microsoft is ignoring all the copyrights on various content.

And there's this debate as to whether or not there's fair use of this stuff, or whether you're supposed to recognize these copyright. I'm shocked as an owner of, of copyrighted material. How do you feel about this?

Look, you know what, listening to you guys talk, I, this is what comes to my mind. You wanna live on the bleeding edge. Sometimes it's bleeding, right?

Hence the name Bleeding Edge, she painful. And, and these are the kinds of things, normally a lot of this stuff takes place behind closed doors and black curtains and smoke-filled rooms and, and, you know, private hearings and, and labs. But we're seeing this play out across our computer screens in real time.

And yes, there's absolutely a copyright image as, as the owner or CEO here of Textron. Do I want my information and images being used to help other people generate content without being compensated? No, I don't want that.

I'm waiting for the times to call me to make a deal so that we could license our data. No one's called you. You mean Open AI or Microsoft or York Times doesn't care.

Oh, I, I wanna be, no, I meant The Times got their stuff licensed. Why didn't we? God darn it.

But they just called, I've got 'em on the phone. You got him on, on the phone? No, I was gonna hang up a phone number.

Call me please. But, um, call Alan Shimmel. Thanks.

But seriously, you know, this is gonna be a growing pain problem that I think two years from now, we'll look back and say, isn't that quaint? Bless their heart. It, it's, it's gonna get solved.

It's, it's, you know, this is the, this Is, I, I mean, I think it said the Sarah Silverman lawsuit, and I don't the gory details, but you know, there she was, her lawsuit was there are comedians now asking chat GPT to create jokes based on her, you know, so I, the whole thing, I can set the persona, Hey, I'm Sarah Silverman, write me some jokes. And I thought her lawsuit was reasonably legitimate in that, you know, like, it's one thing to say chat GPT create me jokes, and it takes jokes from, you know, all the different comedians and creates an mation that gives you, that's gonna be a hard problem, right? Just like an image that gets created, creating a Disney character.

But if I go into the prompt and I say, create me a Disney character that's, you know, smoking marijuana, and, you know, then there should be some, um, transparency and oversight in whoever's controlling the prompt context windows to at least say, and, and the same with Sarah s Silverman, right? If I say I go create me jokes based on the persona of sasine, then I go out to a nightclub and I start telling those jokes. I don't know how, you know, open island.

Look, let, let me give you a legal John for this, right? So the, the analog to this in the analog world was the case of Bella Lago versus, I forget who he was, it's his family. He was dead already.

And that people were using the likeness of Bella Lago to connotate vampires, EULAR. And the, and this went all the way up. This is a big case in law school.

Where do you have the right to your likeness? And today we think of it, of Dove course he does, or they do. Back then it wasn't so black and white, it was a black and white picture, actually, but it was, it wasn't so cut and dry.

Um, and the, and the courts put in this whole right to privacy and the right to your likeness as part of your being. I think we're gonna see the same thing in the Sarah Silverman case, right? Where you have a right to your style, to your who, you know, your essence, if you will, your digital essence, your persona.

Yeah, your persona, right? And, but again, we're on the bleeding edge, these things, it's gonna take a few years for it to play out. But we, we, look guys, we, we probably spent more time, we didn't get a chance to talk about Google stumbling out of the block again.

Well, We, we mentioned that at the top of the thing, saying they too are part of the example. I would just say about all this stuff. You got Microsoft Google saying they're gonna indemnify anybody from a lawsuit, and that's awesome.

Great. But just remember, you know, if they lose that, all these images are gonna have to get ripped and replaced, and all that code that you're generating too might have to get ripped and replaced because the tube belongs to somebody else. So it's gonna be a problem.

I say full speed ahead. Damn, the torpedoes. We're taking a break.

We'll be back with text and gang Cloud native now is the web's leading resource for the growing cloud native ecosystem. com is your destination for news, thought leadership, features and webinars on cloud native architecture, Kubernetes serverless, cloud native application development, microservices, service mesh, cloud native security, and more stay on the cutting edge of modern application development at Cloud native now. All right, folks, we're back to our next section here.

And we're talking about a survey that now full disclosure was actually put out by a PR firm named Edelman, and they're also an advertising agency, but they asked a bunch of consumers whether or not they still trust AI as much as they did. And the numbers seem to have dropped from the first time that Edelman put out this survey. It's not a catastrophic drop, it's like from 54% down to like 44, 47, somewhere in that neighborhood.

But the point here is the more we seem to get exposed to these AI models, the more we start to see some of these issues we were just talking about. So, um, Mitch, let's start with you. What's your level of trust in these AI models these days?

Well, we were talking about in the prior segment, John was talking about the non-deterministic nature of some of this, whether it's unsupervised learning or it's, uh, large language models, et cetera. I think, I think we're one of the reasons for the drop in trust and, and there's a good article that mentions this is our own privacy, and we were talking about the persona be before about that. It's just our information.

You know, you don't have to be a comedian to be worried about my stuff getting out on the internet now, my stuff showing up in someone's, you know, prompt from a gen ai, uh, a prompt that it's sent in. So I, I think, you know, overall people are just, we're, we're a little too close to the edge of the, uh, you know, the AI doomsday kind of scenario and a little close to some sci-fi so people can relate to it. I'm not saying it's happening, but it's, you don't have to draw a very long line to get from where we are to what could happen.

So I think it's, it just worries people now where it didn't worry us before to the same extent. I don't know, Sharon, what do you think? Because I've known you for well over two decades, and I know you don't trust anybody.

You're right about that. You're certainly right about that. Um, no, I, I also don't, there's, there's degrees I think, and for me it's more about the hallucinations that keep happening.

I get so many submissions or the magazines daily, and I can absolutely tell when someone has sent me a chat GPT generated article that they want to go on DevOps or cloud native now, and they're trying really hard to pretend that they're an expert and half of the sentences do not, the sentences don't make sense grammatically, nor do they make sense technologically. And I think to extrapolate a little bit, people are starting to realize that chat GPT can spit out nonsense. That sounds real, real good, but you know, maybe they submit that as a paper or they present that at work and everybody looks at them and goes, oh my God, you are full of crap, doesn't mean anything.

So I don't know how much that's impacting people's trust. It certainly makes me take a couple steps back and go, all right, you know, I'm gonna, maybe I'll use the tool to help me develop something, but I'm not gonna trust it a hundred percent when, you know, my, my job is on the line, my credibility is on the line. That kind of thing.

Hey, John, I was at this AI meetup last week, and one of the things they were talking about is the longer the article gets, the harder it is for the AI engine to maintain consistency, because the, it seems to forget the front part of the article. So are we starting to see an issue here with, um, just, you know, how much depth can the AI engine really get to and maintain some level of constant, uh, truth? Well, that, that's a cat and mouse game, because I think as these contact windows get so much bigger, um, the, um, the, that's, we're gonna see that less and less.

I think that comes from that. It's like, you know, I think they call it small to big. You know, this problem is that when, when people are trying to build their own corporate data, they have to be aware of that, that you have to put some of your sort of really important knowledge in the middle of a thing.

So that doesn't happen. But these context windows are getting so massively big. I, I think it, it goes back to the, the sort of the theme maybe of this, this day is, uh, maybe it's not AI Monday, it's AI non-deterministic Monday, which is, there's two points I wanna make.

One, there is sensationalism, like the, you know, the, the Air Canada thing is a classic when we talk about trust, right? That's, that hasn't helped the case for ai, right? But the, the funny thing about it is, it's a, what, $6 billion mark cap that had to pay $1,200, right?

And, and there's a really high probability that they weren't even using GPT because it was like two or three years ago, but all the sensation is, look, ai, look at this, the world is ending. Like, all right, stop, take a breather. It was $1,200.

I mean, it was terrible pr So there is this, you know, the, the Gemini problem. And, and we're just gonna see this sensationalist and that's gonna sort of erode people's confidence. But at the end of the day, the real problem is, and Damon says that Damon Edwards, another friend of techron, says this very well, that in, in data science and, and in not in data processing and computer processing, all the things that we've all been involved with for 20, 30, whatever years, we've always been trying to drive a deterministic way.

We write programs to give us deterministic output. We manage, um, flow of, you know, software delivery in determinism. We now live in a world that is foreign to almost everybody unless you've been studying AI for 20 or 30 years, which is all non-deterministic and probabilistic.

And so we, like a hallucination is really just the model telling you what it thinks is the best answer. Mm-Hmm. So it's not really hallucination, it's not doing It on purpose.

It's That's right. It it is, it is, uh, it is what it was designed to tell you. And so a lot of my, in my time now, like Sharon, not I do, and, and what, you know, if you're a corporation and you want to use an LM model to, and a friend of mine is building a system for, um, for California Edison, right?

And it's an LLM that's a multi-language that the people, the guys and women that climb up on the, the power lines and have to change those, like incredibly dangerous things. You can't just use chat GPT to answer questions. So they have to have like 95%, 98% accuracy.

And to do that, you have to, you have to do a process of data engineering to get your data. And, and it's a whole, you know, it's what my workshops, so you ask my workshops, that's, my workshops are about how Do you, I I'm gonna weigh in on this stuff, Uh, after, and I'll go after you. Okay.

You wanna go first? I'll go first. Good.

So I think that, uh, I determined, how's that? Yeah. Um, in my own natural intelligence that two things are afoot here.

Uh, the more general purpose that model is, the less I'm gonna trust it. So I think what we really wanna see is a lot more, uh, narrowly focused LLMs that are optimized for specific tasks and the data's been vetted, and then we'll stitch together all these LLMs into something that feels like a workflow and we'll get the result we want. But I'm not sure that these big open AI platforms, I mean, lovely exercise and computer science, and it's taught us a lot, but I don't think that's the end game.

So I, I'm going to take a different view than all of you. And, and here's my thing. What you are all talking about is human bias.

We live in a society where our politicians lie through their teeth every time they open their mouth, oh, open their mouth. We live in a society where, as much as we're talking about trust in ai, it's still four times how much we trust our Congress. Okay?

Humanity lies, humanity created ai, humanity has biases. Humanity created ai. Why are we holding AI up to this, I think artificial thing of, oh my God, it said something that's not true.

Every, you know, look, look, look to the parents to see the child's behavior. We, we live in a world of lying and bias and hallucinations and everything else. To your point, so I remember when a copilot came out, right?

And, and, uh, it was one of the large banks who have like 15,000 j Java developers, right? And it, one person was telling me, oh my God, that, you know, it just doesn't write great code. And he was all complaining about how the code might not be amazing and great.

And I said, Adam's 15,000 Java developers, you have how many of those, you know, can we do a be chart great code? How many of those are great coders? And the answer was not a whole lot.

Well, then, like, so that to your point, humans lie, coders don't create, not all coders create amazing code, you know? So when we start criticizing copilot for creating mediocre code, that's the medium. Yeah.

You come back to me in three years because this thing gets better every day. That's right. It's exactly right.

That's right. Right. Now I'm gonna defend AI point, go ahead.

Go ahead, Sharon. I was just gonna say, that was kind of what I was trying to get at in the last segment, was that until we fix these problems in the, the human element and can eliminate, you know, all bias and poor performance and low intelligence and all of that stuff, then we are never gonna be able to achieve that with, with ai, because that's where the information is originally coming from. So, okay.

Sorry, Mitch, go Ahead. No, no, I, I would back you up on that because this is not an AI problem. It's a human problem.

Yep. Sharon will tell you the longer my articles are, the harder time I have putting together a cohesive, understandable control, well, That has more to do with your, your advanced state, but, well, But I bet I'm not the only one on the planet that has that issue. No, absolutely not.

Those articles are fed into this model. One Is swear on this. So I was a poli sci major in college, right?

And one of my chief poli sci professors, he, he was a socialist in the thirties, right? Uh, but went the other way, became an arch conservative. And he used to give a, he used to hammer this home to us at St.

John's University. He used to say, communism is a great, uh, model if you have a society of atheistic saints, right? But humanity is not made up of atheistic saints.

All right. Last word on the subject. I would just remind you, great movie.

Are you mad as hell? And not gonna take it anymore, no matter who is lying, Shout it out the window. Shout it out the window.

I love ending with a, a movie quote. Let's take a break here on the gang. We're gonna be back in a minute.

We've got some robots. All right, we're back. And as promised, we're gonna talk about robots.

Folks at Toyota have been showing their latest advances in a, uh, thing called a EO project, which is gonna bring robots into our houses one day, and they're gonna be companions and play chess with us, and do the laundry and do all kinds of things. And Chinese, You don't know what all kinds of things are, Mike. Let's Not get exclusive.

Well, it's, it's up to the individual man. It's privacy, man, privacy. That's privacy.

And at the same time, the Chinese are talking about mass consumerization of these robots. They're talking about building millions of these things. So it looks like maybe in our lifetime we're gonna have robot companions.

So John, I know you're a big fan of Toyota over the years, and you've been watching those guys. So what's your take on robots? And, you know, is this the ultimate expression of ai?

Yeah, no, I think there's, it's funny, just this week there was two, I, I read that article and, you know, I, I like TRI, I am like, okay, I, you know, like I think you knew you got me on that one. 'cause I'm a, I'm a big Toyota, like the history of Toyota. Mm-Hmm.

But, um, you know, this week, uh, I'm up, gene Kim has his, uh, forum where we write papers. And, uh, one of the guys who came up here had a lot of experience on, on robotics. And one of the things he's saying that we're seeing a convergence now, and I, I didn't know this and I did a little research, but, you know, Amazon has this astro and then Google's trying to create this.

And what they're doing is, you know, one of the approaches, its not like in the, the, the TRI, they were dealing with the sensitivity and sensors and how does a robot like not, you know, kill a child by, you know, sort of trying to lift it up, right? 'cause it doesn't know the strength. So that's one angle.

But what, what I'm seeing, and I heard this week, was using the sort of LLM agent competition models to behave more like, you know, a human for decisions. You know? So she start reading about where like, Amazon wants to go with this astro to be a guard dog, a pet, you know?

And, and, and so what you wind up having then is using a lot of these large language model to actually sort of, um, sort of like multi-agent stuff, where like you, you know, we're finding more, we take agents and sort of pit 'em against each other, the more knowledge we get. And you, you, you mentioned this Mike in the last second, sort of like, you know, maybe at the end of the day it's more about specific models working in concerts, so that there's a lot of that going on. And there's one other thing that was interesting.

Um, the, um, the, a guy from the military, the military is basically coming up with, um, these massive clusters of drones that are like for battle. And they'll, and they're really, right now, the, the problem they have is they, you know, you've all seen, like at football games now where college will show like these amazing pictures of drones in the air, and they're making images, but there's no, um, collaborative communication between those. They're all just, each drone is supposed to be in its own position, and they don't really know about the other drones.

What they're trying to do is try to create, sort of send drones in that literally a throwaway. I mean, like, they're gonna be the front edge or the end point or the edge of battle, but they're, they're trying to create, and the reason why they're publicly disclosing all this is all the logic in robotics right now, or drones is proprietary for, there's some interesting stuff that's communication between like, if I had 30 drones, how can I get them to work together? But all that's proprietary.

So the government decided that we're gonna open this up, that this is what we're looking for, so people could create more open source paths for, um, for robotic communication. And, and I thought that was really Interesting. I, I thought there was a movie about this already in Star Wars, wasn't there?

The Clone Wars? Well, there was, uh, I think It's called the Borg Collective, isn't it? Oh, the Borg.

Okay. I mean, this is sci-fi living out in Real, real world right here. Yeah.

Yes. Sharon, let me get Sharon, lemme get your thoughts here on this, because you know, what John is alluding to is, um, you know, on page 42 of the license agreement, it says, you know, the robot might kill your child. So I, I knew you kind of Listen, in Philadelphia, the legend of hitchBOT reigns Supreme.

And I don't know if Youngs, as we say around here, have heard about this, but a couple years ago, uh, a college, I believe it was a college robotics team or group, uh, created an autonomous robot. They called it hitchBOT. And they trained it to literally hitchhike.

And the point was they set it loose. Wow. And they trained it to like, roll along the, the side of the highway, and it would stick its thumb out.

People would pick it up and take it on a leg of their journey, and then they would set it down and it would, it would move along. And let me tell you, that little guy made it through the Pacific Northwest, I believe through Canada, came down the Northeast corridor until it got to Philadelphia, where a group of folks beat the living crap out of it and drowned it in a puddle. That's hilarious.

City brotherly love. We've come full circle, square, It's journey ended. Hey, hey.

It's, it's a robot. It's not a brother. That's true.

That's right. That's The Broad Street bullies for, That's the Broad Street bullies. So I think what John was saying about, you know, these, these things are gonna usually end up kind of being like one time use deal.

Like Yeah, maybe they don't necessarily need to then network together and be able to share information if they're just flying out there to distract or do, you know, Strike Kind of things. I don't know. We're Seeing this in Ukraine right now.

Yeah. The Kamikaze Jones that the Russians buy from, uh, Iran, right? They're one time users.

They're, you know, It the reason they needed collaboration, right? 'cause they are just one time they're gonna go to the front of battle, they expect to lose all of them. Mm-Hmm.

But one of the things that is like guidance Mm-Hmm. So what if the place they're going changes? That's the problem.

Right? Now, if you think about the, the college thing where the kid, the, the engineering students show their logo with colored drones, they're all, yeah. None of them know about each other.

The problem that the government is seeing is like, okay, the mission is to go here and, but it's changed. And like, that's a really hard problem if the actual drone stamp can't. But By open sourcing that are we opening that technology up to nefarious groups and, and nation states that we don't want to have, have that.

But again, like the Chinese, you know, he didn't go too deep into the classified stuff, but I didn't go into any of the classified stuff. But, but the point being, it's happening. So you, you either, you know, you either sort of open it and get at least a posture where you're protected.

'cause I mean, and our, our adversaries are already doing this stuff. I An Issue with open source regardless of its use, right? So, okay.

Sorry, Mitch, go ahead. I keep interrupting you. My bad.

No, no, no. What what came to mind as we're talking about this is I think it was like 2015 or 16, the whole idea of swarm intelligence, right? Looking at nature and how, you know, birds flying these crazy, or fish, you know, flying these, uh, swimming these crazy patterns.

It almost sounds like that's kind of the, the, uh, you know, there is some logic about where we're going and how they all get there. They talk to each other and form some kind of pattern to do that. It seems like a replication of that.

The, the other thought I have is, I, I don't know. I just, I'm, I'm very skeptical of, am I, are we close? I won't even let Tesla drive my car autonomously.

Am I gonna turn a robot loose in my house and come home and find, you know what? It just, we're not there yet. There's so many challenges, I think to be solved.

May, maybe it's a year down the road, maybe it's 20. I don't know what it is, but I don't think, let's going back to the bias thing, if I can real quick, like the joy, I mean, if you read her book, I'm asking, I, one of the things was she was a dark skinned African in, in MIT and, and, and she found that when she was doing image recognition, it didn't recognize her. And so she basically wrote a whole book about like how early image recognition systems didn't recognize Michelle Obama.

But here's where it gets really scary, especially in drones, but even cars, there's already studies now that, that a autonomous vehicle might opt in a decision where it only has two choices to go towards white people versus black people. Col guru. Mm-Hmm.

Well, because the corpus of data is like, I mean, she even said the NIST database, 15 people don't know what that is. They know if they don't, then they don't get the joke, right? Yeah.

The n had to be there. The NIST database, which is the standard for facial recognition or biometric information, is like, uh, she said 80% white, 75% male. So we have a bias built into, and so you go back to drones.

I mean, her point is a drone that has autonomy and, uh, weapons. Like who knows how the decision it's gonna make. And then there's gender bias.

I mean, same things, you know, 80% of all the images in this data is male. 75% are white. And so that is the data that's driving this probabilistic deter, you know?

Yeah. So it's, I thought that was fascinating. It is.

I highly wanna in her book. So guys, I I, I, I gotta jump in though. I think part of the problem is, you know, we, part of it, I, I'll be really honest, it's age bias.

Guys, you know, we've got, except for Sharon here, the rest of us are of an age where this is brand spanking, new radical. My God, it's changing. You know, think about someone born in the 18 hundreds and seeing planes flying, right?

What, what are they gonna think? I think for our children, and for the young people coming up here who are gonna be AI natives, there's gonna be a, and, and robot, na robotic natives, it's gonna be a lot more normal to them and normal lies than they'll, and they'll sand off a lot of these rough edges. I wanted to bring up something.

I'm holding a book, you know, next week, Mitchell, Mike and I are in, not next week, the week after, actually next week, we are in Paris for CubeCon, uh, cloud Native Con. And there's a young lady named Cassandra Chin. She's the daughter of a good friend of mine, Steven Chin from Jfr.

She does the, uh, children's, uh, she usually brings raspberry pies, and they do, uh, children coding at CubeCon. She just wrote this book called PPI's, AI Friend FPI is one of the cube con, uh, toys, stuffed toys that you'll see at Cube Con and Fite PPI's AI friend here is a robot guided by ai. And unfortunately, he doesn't have a lot of the skills that the real animals do, but fpi found his secret power in ai.

And that's, and there's a workshop that's a companion to this book. These are the kinds of things that our kids are gonna be using and growing up with by the time they're old enough, the, these bias about letting the car drive or having a robot in the house, or whether AI's using copyright images, they're gonna be way in the rear view mirror. It's just gonna be life to them.

I think too, a good example of that, Alan, is we, we o often reference something like Khi Maru or Sci-fi this, and in, in our kind of history of our lives, things that have influenced us, you know, generations of, of younger folks, kids, whatever they're watching AI companion on Hulu, they're watching, they're seeing a whole new genre of which it's interesting to us. But that's the new thing. That's what they're building their corpus of what's possible.

At least that's one of the influences on it. So It's A total change in perspective. I wanna get one last thing in, you know, so in some ways the Kobe removed, so the, the original test for what you call the ache removal, what I described as that sort of sit when autonomous vehicle has made a decision that's about an 80-year-old test called the Trolley Car.

And That's right. And it's a famous, you know, so, so I don't know. I mean, I, I hope you're right, Alan.

I do think I do. I think, you know, I told you guys I'm writing a book about AI and you know, I don't worry about the world. I don't worry about Terminator and I don't worry about all that stuff.

This is the new technology that's just gonna change the way we live and behave, and hopefully for the better. I do worry about the bias in the models. I, I'm just wondering if that might be too big of a problem to solve.

My last comment on all this, get your AI robot off my lawn. Don't Go play on your own Ai. It's gonna hell off my lawn.

On, on, on the issue of bi. We'll end it up here on bias. 'cause I wanna make a a just a quick statement.

We recorded this on International Women's Day and this is, uh, international Women's Month, and we're excited to have women like Sharon and Amanda Rini and Chase Bannon and, and, and Tracy Reagan and some of the other people here on the Gang. Um, we've come a long way, but we still got a long way to go, right? We've come a long way, baby, but we've still got a long way to go.

And I, I just feel like we should acknowledge that we are in, in Cube k and we will be meet, actually I'm interviewing Cassandra and Steven as well as everyone else there. If you're watching this in, you're in Cube Con. Come by and say hello to us.

We'll be right on the show. Floor Text Drunk TV Live. John, enjoy West Coast.

Thanks for being on the Gang with us today. Mitchell, Sharon, of course, Mike, thank you. You've just watched Text Drug Gang.