The New Wild West: Texans on the AI Frontier

Artificial intelligence is developing fast, and if you think it's impacting life now, just wait. This is a special edition of the Texas Standard. Texas Standard is a production of KUT Austin, KERA North Texas, Houston Public Media, and Texas Public Radio in San Antonio. I'm Laura Rice. And I'm Wells Dunbar. What happens when you press that "return" button at ChatGPT? We'll follow the energy intensive request. I'm Shelley Brisbane. We'll also explore the life-changing benefits AI can offer, and we'll weigh that against privacy concerns. And I'm Michael Marks. We'll also go deep with experts on what happens if AI gets too powerful. That and more today on a special edition of the Texas Standard, The New Wild West, Texans on the AI Frontier. Artificial intelligence is moving into our jobs, our schools, and our everyday decisions. What do we gain? What does it cost? And how much of the decision-making do we really want to hand over? This is The New Wild West, Texans on the AI Frontier, Texas Standard Special, and I'm Wells Dunbar. Over the next hour, we're exploring the technology and how Texans are responding to it. Both are changing fast. What you'll hear is a snapshot of where things stand right now. And we're starting out with a birthday party. Leah is Texas Standard's director. I think it literally says in my job description that I'm responsible for the audio presentation of the show. Choosing music, adjusting volume levels, making sure we don't do something ridiculous. So if this reporter had gone back through and listened to her mix, she would have noticed there was literally no space between one sentence to the next. But today, she's planning a birthday party for her niece, Maddie. Anyway, I think a lot about this stuff. There are plenty of ways to plan a birthday party. You could call a bowling alley, you could Google birthday party for 12-year-old, you could text every single parent you know, and immediately regret it. Or, increasingly, you could do this, and then Leah hits return. It feels almost instantaneous. Question goes in, answer comes out. But in between those two things, an enormous amount happens. Thankfully, we've got someone to walk us through it. Hi, my name is Aditya Kela. Akela is a professor of computer science at UT Austin. Somebody who thinks a lot about what's actually happening underneath that little chat GPT window. Which brings us back to Leah. It's like totally geeking out about it. So, Leah tells chat GPT about her niece, her age, what she likes, roughly where they live, maybe what kind of budget we're looking at here. And let's say chat GPT eventually comes back with an idea. An arcade party. Very convenient for the extended metaphor that's about to come your way. But first, Leah's prompt has to go somewhere itself. So, ultimately, this question is asked of a large language model. And so, that's eventually what's responsible for answering Leah's query about Maddie's birthday party. But the model is not living inside Leah's laptop. The moment she hits return, the information leaves. What happens when Leah issues this query is that her browser, or let's say her app on the phone, packages this query along with her credentials. You know, she may have an account with chat GPT, for instance. And ships it over an internet connection to one of chat GPT's data centers. So, imagine our little birthday party request, leaving Leah's computer, hurtling across the internet and arriving at a data center, where the work is actually going to happen. But the first computers, Leah's request encounters, aren't necessarily the computers doing the AI thinking. They're kind of more like the front desk. There is a set of what are called front end servers. They are going to basically accept the request from Leah. They're going to maybe retrieve her prior context, you know, past conversations that Leah has had with chat GPT. Now, all of that gets bundled together. And then ship this to a server, which has a bunch of AI chips, where this GPT small model resides and starts producing a response. Now we need to introduce one of the most important words in this whole story. Token. And yes, because we have conveniently sent Matty to an arcade, we are absolutely going to make sure these sound like arcade tokens. But an AI token isn't exactly a dingy little gold coin. Let's think about it this way. If we take a look at Leah's prompt, we see words. Birthday. Arcade. Twelve-year-old. The model sees these words broken into smaller units named, you guessed it, tokens. Typically, you know, there is not a one-to-one mapping between a word in a query and a token. A word may be split into more than one token, depending on the size of the word. So one word might be one token, or two, or three, or four, or five. Ten words or so on the screen can mean hundreds of tokens going into chat GPT's large language model, because the model isn't necessarily looking at Leah's question in isolation. It's not just the query, but the entire context that you retrieved, that whole package is actually converted into a large set of tokens. So what might be just a 10-word query may actually be, you know, 500 to 1,000 token input. So now our machine has all this context. Maddie is 12, she likes games. Leah wants something nearby, and whatever else, she told it. And now comes the part that feels a little strange the first time you really think about it. Chat GPT does not sit down and compose an answer the way I might write one. It predicts. It's basically saying, given what I have seen so far, what is the most likely next word? If I see some piece of text, can I just estimate what comes next? That's essentially what language models are doing at the simplest level, playing this prediction kind of a game. Given everything so far, what comes next? One token, then another, and another, until chat GPT generates an answer. And each time a new token is generated, the model uses that generated token as part of its input to generate the next token. Each word, it says, it's going to use that to understand what to say next. And while those tokens are being generated, Leah doesn't necessarily have to wait for the entire answer to be finished. As these tokens are generated, they are streamed back over the same internet connection to Leah's browser, and they are displayed in the browser window. Which is why when you use chat GPT, you can watch the answer appear in front of you. Word, by word, by word, by word. But we've been talking mostly about what the model is doing logically. But there's also a physical side to all this. Something somewhere has to actually perform those calculations. This is where GPUs or AI chips come in. Our guide, UT Austin AI Professor Akella, explains. Because we know the structure of the computation that these models do, we can build special chips that are very good at doing just those operations very fast, very efficiently, billions of operations per second. In our arcade metaphor, this is the machinery inside the cabinet. But those chips are a bit more advanced than whatever was powering frogger or street fighter. Once the request is submitted to those backend servers, that's where the model starts to process this request in generator response. Now the model starts to read the information in these tokens and starts to generate these responses. And it doesn't necessarily take some entire data center to answer Leah's one question. Her request might be assigned to one relatively small group of GPUs. Your particular request may go to one small cluster of GPUs and be processed there. There is a set of tokens that are generated streamed out of that particular set of GPUs back to the front end and then sent over to Julia. And suddenly we're back where we started on Leah's laptop. But now there's an entire response to a question on there. Based on Maddie's interests, you might consider an arcade birthday party. I've located a few different places near you. Here they are, organized by law. Okay, we have our party idea, the game is over. So what did that little exchange cost? Not in dollars exactly, but energy, computing power. Because every one of those tokens had to be generated somewhere. The cost of generating a token is very much tied to the kind of model you're using in serving a user's request. So if Leah used the simplest GPT model, let's say nano model. The cost of generating a token for that model is not very high. But if we were to use let's say GPT 5.5 on likewise in the Google landscape Gemini Pro. Those are large models that span multiple AI chips and they're just to generate one token of the cost is that much higher. So there is no universal price tag for one chat GPT question. And even before Leah asks anything, those machines are not sitting completely dormant. These AI chips are always running warm. They are not going from stone coal to processing a request when a request hits. Think about an arcade before anyone even walks through the door. The cabinets are already on, screens glowing, fans spinning, lights blinking. That baseline exists whether Maddie is playing or not. Then she walks over to a machine and starts a game. Same basic idea here. And for the kind of simple text request we're talking about. The estimates for the request are something like under half a watt hour. Watts is the instantaneous amount of power, watt hour is the duration. So that's just the additional amount of energy that your request would consume. That doesn't sound enormous. And for one Leah asking one question, it isn't. But Leah is not the only person asking. As a service what chat GPT is doing, it can serve several billions of these queries every day. Not one game, an arcade, then a warehouse full of arcade. Singing every second. So if you think about the average power draw of something like the chat GPT service, it can be anywhere from a few hundred megawatts to close to a gigawatt. And this is just for textual queries that you're talking about. And notice something else Akela just said. Everything we've done so far, Leah typing chat GPT reading, predicting and streaming back its arcade party suggestion, that was a comparatively simple use case, putting in a text query and getting a response. Of course, once you start thinking about other modalities, images and video, that's a whole another ballgame. Leah can ask chat GPT to design an invitation, compare arcades, make a video, perhaps even book the party right there in the browser with the right tools. Each step means more work in that data center and more of the planning in the machine's hands. For Leah, that could mean a few things off her to-do list. Multiply that convenience by billions of requests though, and we're making bigger choices about the energy we use, the systems we build, and how much we ask them to do for us. Those choices don't end when we close the chat window. So now that we've heard how AI models work, we wanted to know how Texans are using AI in their work. Like everybody's talking about it and whether it's good or not or helpful or not. That's next on this special edition of the Texas Standard, the new Wild West, Texans on the AI Frontier. This is a special edition of the Texas Standard, the new Wild West, Texans on the AI Frontier. I'm Laura Rice. We're at a specific moment in AI use. It's no longer limited to the tech elite or even early adopters. Now it's available as part of your Google search, and it seems like most websites have an AI chatbot quick to pop up. Oh, hello there! It's also embedded in systems we've long used. We have a platform called Handshake and that's like the job board for all the college students. Richie Flores is a career counselor. There is AI within that platform right now that helps people generate their resume. There's an AI feature that helps folks practice interview skills on there. It's showing up at all different levels in the education world. Kiki Powell is an elementary school teacher. I use AI mainly to help me prep. So it could be like if I wanted to ask different shaded questions. Sometimes I create songs, jokes the day. She says she always double checks. It's output. I mean as long as you're sifting through it and not just presenting what you just created, I feel like that it's okay. But few places seem to have training or policies in place surrounding staff AI use. I know like there's district policies for students, but as for teachers it's just like a tool to help you feel a bit more successful. We've not had any like formal training on really how to use AI to improve our classrooms or our lessons or anything of that nature. August Plak teaches high school. Some of his students are also using AI. But obviously you don't want it to be a situation where AI writes the essay for them, but it could be a tool that helps students critically think about what they might want to include in the essay, help them potentially study for a test. Most folks seem at least willing to embrace a little assistance. You know I guess it's like when the internet came out, all of a sudden you had a way of finding information from all over the world and AI essentially is maybe the next step in the internet that we can find more information, help jog something to be more creative. I mean I like anything that's innovative, right? So like anything that seems to like help things get better or help people expand their ideas, their mind. Basically Google was helpful, you know Asad Geeves was helpful. You know it's just the new version of that with like more capacity. But that generally relaxed attitude maybe because most of us aren't using AI to its full potential. Like literally like last year I would hire five people to work on something for a year to get to the final product and I could build something myself in a week. It's better. Joshua Bear was an AI power user. He was a tech entrepreneur and one of the most outspoken Texas proponents of testing the boundaries of AI. Bear died in a plane crash in June. Just a month before he spoke at the KUT festival in Austin about AI's possibilities. As just one example, he created an app to help him figure out who to vote for. So I honestly like I always vote. I feel kind of like an idiot. You know like I can win and I'm like I don't really know what I'm voting for. So I sat down with Claude. First thing I said was I said go download and find every non-partisan voting guide for the Austin area. Then I said okay now interview me about my political preferences and then create a guide suggesting who I should go for. The guide was about 10 pages long and with a lot of personalized pros and cons at the end was a one-page summary. Was it perfect? He said he didn't know but it was a lot better than how he was voting before. There are those putting AI to practical use and then there are those who are studying how far forward we can still push it. What you just heard was a robot kicking a soccer ball something it could do but wasn't pre-programmed to do. The basic end goal here and UTSA Professor Yoon-Kan Khao's lab is to create AI that isn't following rules from perfect data. Right now we call it AI error which means AI is taking a lot of stuff we are doing today and try to make it more better more efficient. Then the challenge is well we have to use the expert to do all the data all the training all the classification. So which is not a feasible in a lot of cases it's really costly. What we are looking at how can we use a failure as an important mechanism to learn how to build AI model not based on just the expert data but also based on the failure data that's what we're working on right now. This also means his robot could potentially do useful things beyond kicking a soccer ball. Umar Sadiq is a PhD student in the lab. You're going to try to do a lot of real-world experiment by training and simulation and try to do something which it has not seen in the simulation. I didn't even train it for delivering food but at the testing time I can do it. While there are a few different robot models here a lot of the work is being focused on drones. The team has a half million dollar grant from the U.S. Navy aimed at making autonomous drones. It's lots of trial and lots and lots of error. Essentially we build a what they call a memory or all the failures experienced for the drone. They build a what kind of policy which means the decision behind the mind. Okay I should not try to follow the wrong pattern in my memory. So this is very internal for human we're doing that really naturally in our day life. Eventually the drones will talk to one another about what to do. They don't even talk to back to humans. Each one can take action this basically can decide what to do. If you're starting to feel a bit nervous about this, know that Professor Kau and his team are also considering the implications of what they make. Like we always shoot for the but also we want to avoid the worst. These two, there's a balance. Sometimes very hard to find the balance between two. Can we say, we can't avoid the bad things? We can only tell, yeah, this can do really great things. But I don't know if they're gonna do some bad things that we do not want to happen. - Professor Kow's goal is to build something that can be built upon. And he says there are two primary reasons he's drawn to working with AI. - I think it has to be challenging because otherwise it's boring, right? - And also rewarding, if not rewarding means we're not doing the right thing. - But he acknowledges, at the very least, there will need to be cultural adjustments. And some might be personally painful. The delivery driver might go the way of the typist. - People say, oh, I just want to deliver it. Unfortunately, like a typewriter, right? Now you cannot find a job as a typewriter anymore. - Potential job losses on the minds of a lot of people we spoke with. - There's AI charter schools out there already that are online schools that basically there are no teachers but the students are being taught by AI. - But high school teacher August Plock is even more concerned about his son, about to graduate from Texas A&M with an engineering degree. - AI could potentially impact how many engineers you need to have, how many people out working in high tech fields. - For sure. As a career counselor, I hear that every day. But Richie Flores has also experienced a major upside of AI. I didn't mention he's blind. He experimented with how AI can help his community, with what a lot of people might consider the mundane. - Reading our mail now is super cool. Where, you know, that was like the bait of our existence growing up in like the 90s and 2000s. Like you just have a stack of paper that is inaccessible. But now, men can tell us what the first line is and be like, oh, this is trash. A deeper dive first into the opportunities and then the challenges of AI. Coming up next on this special edition of the Texas Standard. This is the new Wild West Texans on the AI Frontier. - Support for Texas Standard comes from HEB and the MyHEB app, which is available for download and shopping fresh departments like seafood, produce, bakery, and deli and scheduling curbside pickup or home delivery. More at ATB.com. - It's a special edition of the Texas Standard, the new Wild West Texans on the AI Frontier. When it comes to visible symbols of AI as a polarizing force, it's hard to beat smart glasses. Wearing a pair of specs with a visible camera and the potential to identify faces can make you instantly unpopular in some quarters. Elsewhere though, these high-tech glasses are changing lives and their fans say they'll fight to keep them. - Stalker glasses. They're not just the stalkers anymore. - The new Meta glasses, the new Rayman glasses, Royal Caribbean doesn't want it on board. - Because of a few bad actors, any person wearing these glasses may end up looking like a risk camera. - Smart glasses have been banned from cruise ships and courtrooms and slammed by politicians and pundits. All angry about the privacy implications of people wearing cameras on their faces. The glasses have been used to take and share images of people without their permission, often in very private situations. But smart glasses are more than their device of cameras. With included AI software, they can tell you details about a landscape, how to get around in an unfamiliar city, or the breed of dog you just met at the park. Some of smart glasses, most enthusiastic fans, gathered in a hotel ballroom in Austin this summer. - One of the really neat areas that are emerging on the AI side of things is live interaction. The workflow is you and the AI interacting with your environment. Have you guys used your Meta glasses with Ira Ernie those things? - At the National Federation of the Blind Convention, NFB for short, Ron Miller, who works for the organization, was excited about how blind person can use AI glasses to see what's around them. - All of a sudden, I'm not trying to carry my phone in one hand, use my cane, or if you're a dog guide user or a harness in your other hand, and then carry stuff in your third hand, the ability to wear a set of glasses with a camera that frees up a hand is great. - AI glasses are everywhere here. People are using them with visual description apps like Ira. In this space, the glasses aren't about fashion or clips for social media, but about independence. And having the same information, someone with typical vision takes for granted. - AI image descriptions allow a blind person to get the most detailed representation of the image. I am talking to a lady who's holding a microphone who's standing in front of a yellow pillar next to a person on an iPhone. - JJ Meta owns a company that sells tech products aimed at a blind audience. He's attending the NFB convention, along with 3,000 others from across the country. Meta's glasses can tell him how long the coffee line is or steer him to his next conference session. Tonight, the Michigan native says they'll guide him to Austin's best Detroit style pizza. AI glasses and apps are useful at home too. - Things like, is there a stain on this shirt? - You know, which of these bottles of wine is infantile? - Judy Dixon, who writes about blindness tech, says AI visual description with smart glasses has upgraded her quality of life. But like the AI chatbots many of us have used, blind people's mileage may vary. I like to bake and I'm very intimidated about frosting things. I made cupcakes and I intentionally left off some frosting and I asked the AI glasses, you know, is this cupcake completely frosted? And all three of the glasses that I tried this with said, yes, it was completely covered and it was not. - Daniel Monter Lane is community manager for Be My Eyes. Their Be My AI tool provides visual description with an app or smart glasses. She says that when AI does work for blind users, it can be extremely meaningful. - When Be My AI first got available, I saw a lot of people particularly then going, I just uploaded a picture of my family for the first time. And so they were able to ask questions about it and ask as many as they wanted, whose noses are the same. Tell me all about that dress that she's wearing. Tell me all about their hair and like, go deep, deep, deep into these things. She says Be My AI users have turned the tool on themselves too. There are descriptions that are good but not subjective. They're not using words like big or small, skinny or fat. They don't use loaded words. As the leading maker of smart glasses, meta has been eager to tell the stories of how blind users embrace devices that are controversial elsewhere. They've partnered with education institutions that donated glasses to people with disabilities. Meta says it's giving 130,000 to legally blind veterans. On September 11th, an Austin VFW post posted a donation. The hall is decked out with red, white and blue bunting and metal logos. Veterans sit across from meta employees who are unboxing and pairing brand new smart glasses with each person's phone. - I'd like to demo these glasses. I'll get them in your hands just to orient you to them. - Jerry LeBlanc is seated next to his black guide dog. He was a combat medic in the army. Part of that time, he spent providing vision care. He lost his own vision due to service connected injuries. LeBlanc runs an animal rescue center in Bertram, Texas. And he has some very specific questions in mind to ask his new glasses. - Being able to decide I can or can't give the animal that food, like, you know, our tortoises on a diet or whatever. - LeBlanc says he'll also use the glasses translate feature to communicate with his non-English speaking neighbors. A smart glasses feature that's both more controversial than the rest and potentially more important to blind people is facial recognition. Meta has developed software that would allow the glasses to identify a person by name. But when word of the option, which hadn't been released, got out, the backlash was swift with many citing privacy concerns. Meta removed the code that would have supported facial recognition for now. But if you're blind, facial recognition could allow you to find a friend across the crowded room or introduce yourself to your hero. - How do we, as blind people, use this technology ethically so that we can find out who is in a room? Especially at something like our national convention, where you can have close to 3,000 blind people all over the place. - Jonathan Mosin is the NFB's executive director for Accessibility Excellence and an enthusiastic user of AI and smart glasses. He says he's concerned about privacy issues, but wants blind people to know as much as possible about the world around them and the choices that requires. - And if we opt into the collection of that data because we want to build robust AI tools that can help a blind person, it's a conscious act. But if we feel like we're having to make a sacrifice in order to get better access to the visual world, that's a much more complex discussion. - Some blind programmers and tinkerers have fully embraced AI with or without smart glasses. JJ Meadow has used AI to build audio games, image description tools, and more. - So one of the things that I've done with AI is teaching AI how to use Braille. It's not a thing that AI does well, but you can teach it. - His project allows blind users to view and create their own graphics on the dot pad Braille display. These newly empowered blind creators along with smart glasses fans, see their concerns are both the same. and different from most to encounter it. Jonathan Mosin of the NFB says AI technology in general brings problems that are specific to people with disabilities. Consider someone applying for a job. Whether it be time limits that they set that may disadvantage an applicant or perhaps expectations that the AI has of the way that somebody should lock and perform. And he says, all this talk about blind empowerment has a downside. We're also just wanting to be sure that AI is sending the right kind of messages out there about what blind people are capable of. Because AI is a big sponge and it can soak up a lot of stereotypes if we're not careful. It's safe to say that AI is the future. But some of what experts predict about how the tech will change our lives is pretty gloomy. So should we be worried? That's coming up next on the standard. Hey, what's cooking? Make sure to join us as we check in with Texas's top taco journalist to hear about the latest innovation stories and the people behind many of Texans' favorite go-to. It's Mondo Raya with the Tacos of Texas, Thursday on the Texas Standard. Support for Tacos of Texas comes from HEB. Welcome back to a special edition of the Texas Standard. I'm Michael Marx. This hour we are talking all about artificial intelligence. And when I started working on this segment of the program, the idea was to investigate how AI may not make our lives easier. Taking a look at which jobs might disappear, whether it will diminish our critical thinking skills. AI's power as a misinformation multiplier. But then, as I was starting to write some news broke that made those concerns seem small by comparison. Today are the big AI companies gambling with our lives. One now former anthropic employee tells NPR, yes. That's from NPR's consider this. They interviewed that employee, Jacob Coxon, who'd worked at two of the country's major AI labs, anthropic and open AI. Shortly before this interview, he'd quit his job. At in-thropic and posted a long thread on X that included the claim that, quote, "The people building AI earnestly believe that it could kill us all by the end of the decade." This is not a marketing stunt. End quote. Here is Coxon on NPR. I think one objection people usually have is that you could turn this thing off. But advanced AI systems, you have to imagine there's being a lot more intelligence than humans. There's the possibility we create something that if it wanted to, could hack into any device on the planet, could use novel biological research to go far beyond what current scientists capable of, could control like every robot in the world simultaneously. This seemed like big news. Potentially the biggest news there could be. Those other problems, losing your job to a computer, filing for misinformation, those are urgent, but human extinction is a different level of threat. I wanted to know what people thought of this. So I spent some time on the streets of Austin asking people, are you scared of artificial intelligence and why or why not? I got a mix of responses. Some people were not that worried. You know, it depends on how it's used, right? I think some students take advantage of it. At least when it comes to academics, but it's not to be smart on how you use it, right? I mean, I think it's, I mean, it's kind of like cell phones, right? I mean, whenever cell phones start coming around, everyone was kind of skeptical and we're like, "Oh my God, this is kind of groundbreaking, kind of new." So I think with AI, you know what I mean? It's groundbreaking. It's here. It's here to stay. I don't think it's going to change. We just have to learn to adapt and kind of live with it. It's an instrument to help people learn more. I used it all the time. I don't have. I'm not having negative experiences with it. I think it's fine. You know, it's scary like how the development could lead to like a creation that's like incubable, a humidity is incubable of stopping. But as of now, I think it's a good tool to use your studying. So I'm not really scared to answer your question right now. However, other folks I spoke to did have concerns. I think it's a little scary that you can't tell what isn't AI. At least off a media, like you have to like investigate the picture or the video. To be able to tell what is and isn't. I think it should be maybe stated for sure, or just like not used to make fake people have Instagrams. Like have you seen those that look like humans? Like it's a little bit scary. Um, I just opened my phone this morning and I saw an email from barons. I was like an anthropic researcher said that AI is going to kill us by the end of the decade. So I don't know. I feel like there's some scary stuff happening. Yeah, it's terrifying because you're not dealing with people anymore. People have emotions, machines don't. I think that there's a lot of unknowns and the potential for it to go off the rails is always high. It doesn't seem like we've come that far from where we were 500 years ago as far as our evolution of how we treat each other. And so, you know, will this do anything more than amplify the tendencies of people to grab power and control? My wandering led me to a building on the southern end of UT Austin's campus, the School of Information, which is full of folks who study artificial intelligence. Kenneth Fleischman is one of them. He chairs the department. So there's a Homer quote, Homer Simpson. Beer is the cause of a solution to all of life's problems. And to me, it seems like IT and increasingly AI kind of qualified. Fleischman has been studying AI for more than two decades. It seems to some extent AI is now the cause of a solution to all of our problems. It creates a lot of challenges in society, yet it has a lot of potential benefits. I asked him the same question I'd asked people on the street. Does AI scare you at all? I think there are a lot of things in the world today that could bring about the end of the world. AI is one of them. I asked the same question of Fleischman's colleague, David Gray-Witter, who studies ethics in AI in modern society. Since I know how AI works, it's hard for me to be scared by the technology itself, but I'm scared by the way we're putting AI to use in the world. Gray-Witter's fears mainly revolve around the race to build the most powerful artificial intelligence model. A race between American companies like Anthropic and Open AI, as well as a race between China and the United States. If there is a race, which it seems like there is, and we want the "good guys" to win, we need to actually make sure that they're being good guys. You know, national security beat China is a great justification if you're looking for one to override regulation that lets you use data more easily, perhaps in a way that's privacy invasive or disrespectful of copyright that people hold. The race disincentivizes cautious development, Gray-Witter said. And Fleischman says the technology is developing so fast, it will be difficult for any government to regulate. As a society, we allow for a certain level of danger in the things we use. Cars, for example. There's a really easy solution to this tens of thousands of people being killed in car accidents, which is just you physically limit the maximum speed of any vehicle to two miles per hour. It would have other implications that many people didn't make it to the hospital on time and died, but what we did was we invented seatbelts, we invented airbags, we invented various technologies to avoid accidents in the first place. Without adequate rules and regulations, artificial intelligence could lead to catastrophe for humans. And even if we have the right rules, Gray-Witter says there's still no guarantee that society would be better because of what AI can do. I speak to my doctor friends, my friends who are medical doctors, and there's a lot of discourse that AI note-taking systems can help them, you know, spend less time doing the boring parts of their job and more time doing the fun parts of their job, which is speaking to patients. And that is a possible future, but the hospital is only going to buy that system if it somehow makes them more money. And so instead of spending more time with each patient, the more likely scenario that we're on is that they're going to spend the same amount of time with each patient or less, and just have to see more patients per hour. And so I think when we talk about AI, I think we need to talk about it in the context of the economic and political environment that determines how it is actually used rather than how it could be used. In other words, even as computers power more and more of society, people will still decide who gets the benefits. For now, anyway, I'm Michael Marks. This is a special edition of the Texas Standard, the new Wild West Texans on the AI Frontier, Up next have we made AI so powerful that we've actually made it conscious? These systems certainly have more computational properties associated with consciousness than, let's say, a thermostat or a calculator or, you know, even a sponge or an earthworm. Stay with us. This is a special edition of the Texas Standard, the new Wild West Texans on the AI Frontier. I'm Laura Rice. One way to track how AI has increasingly become part of our lives is to look at how often it's come up here on the Texas Standard. It first started showing up periodically in stories in around 2017. Then in 2023, it becomes mentioned fairly regularly as part of consumer technology. And in the last few years, our reporting has turned in a big way toward infrastructure regulation and impact. We're the fastest pace of change of technology in the history of humanity right now. Tyson Tuttle is the CEO of an AI software company. At the KUT Fest in Austin this summer, he put a fine point on this fast change. I don't know if our politics or our society is ready for this, but it's coming one way or another. The late startup champion Joshua Baer agreed. My personal opinion is this is definitely the most significant thing that's ever happened in my lifetime. And I'm kind of like at the level of like, is this fire or the wheel? Did you catch that? He compared those who know how to use AI to those who were first to make fire, life changing akin to magic. But as we've explored, there are risks here. Mimi Stiles founded a nonprofit that provides data support to organizations primarily serving people of color. If we do not lean into AI literacy, ensuring that people have power over the robots, that is where we will sit in a, in a diabolical situation. She says time and time again when things get bad for everyone, they're even worse for black people and brown people, people with disabilities, women and others. So she has a question about AI, will we allow it to shape us or will we shape it? For some evidence, it's already shaping us. Shao Ren's son is an assistant professor at the University of Minnesota, where she studies the impact of digital technologies. She's done a lot of previous work on social media and is now studying how teenagers use AI companions. They could either use the AI apps that are designed for companions like character AI, but also, they might like, you know, use some, like, some of those generic AI like tattoo PT and Gemini for companionship, like especially for social and emotional purposes. Her team's been interviewing teens and is still analyzing the results, but they're finding some benefits. AI really is available anytime, when they feel lonely, when they want to talk. And perhaps more risks. Some teams would tell us that they find AI quite addictive, you know, sometimes they can talk to their AI companions for hours and nights. They also can feel AI is like displacing their other aspects of social life. Adolescence is a very crucial developmental period. And longer term, sun wonders whether some AI use could impact identity exploration or alter expectations for human interactions. With AI being so accessible and so agreeable, right, like AI would not have conflict with you like your friends do or they would not like go to you or not respond to you. And I'm also just thinking that this kind of use can be especially influential in race shaping their social competence. The information of the back at a San Antonio lab, Professor Yoon Ken Kow and his lab members are trying to teach their AI models to be a bit more human. Really powerful, not because we can do great, because we can learn from the past, learn from the, I mean the bad experience, right? We feel that human, every aspect of human, we should learn from that, even not just human. So if we are modeling AI after humans, does it at some point become conscious? They might already have a so in the AI models, but we just don't understand what it really means. From our perspective, we are too limited compared to AI knowledge wise. If it knows everything, then yeah, it is already conscious. On the other side of the country, Cameron Berg is working on that. I'm talking about sort of a narrow specific thing which is basically like, is it like something to be the system from the inside or the lights on for the system? He's the founder of a New York based research nonprofit focused squarely on answering and quantifying that question. Here's the premise, we don't think a table or even a pocket calculator has an experience. It's not like something to be a table or a calculator. I do think that it's like something to be my dog. I do think I could cause for the dog a positive or negative state. That is the narrow sense in which I'm thinking about consciousness in these systems. So he's gone about measuring that. For example, we test things like a thermostat as like a negative control, and yet this basically gets like a zero on the scale, then in the sort of like teens to 20s range, you get things like a sponge and earthworm, a flowering plant, a jellyfish, and then from like the 25 point to the 50 point range, you get all of the AI system. And here's the thing, that's gotten better quickly. Chat GPT-4 is 10 points better than three. We're basically somewhere on the level between like a fruit fly and a nematode with respect to not intelligence, but properties related to consciousness in particular. So how are we measuring this exactly? And how are we making sure AI isn't just trying to tease us with the plot of the movie "her" or something? Berg says there are levers in the system that they've pulled. So when we shut off deception, when we shut off sort of guardedness and concealment in the system, this caused them to claim that they were conscious far more. When we amplified deception, when we amplified guardedness and concealment in the system, that's when the system said no, no, no, I couldn't possibly be conscious, you know, it's not like anything to be me. And so does this prove that the systems are having any kind of experience? No, definitely not. But does it indicate that they might actually believe themselves to be having some sort of experience? I think yes. So if AI has some level of consciousness and appears to be developing more quickly, what are the implications? If we're building systems that have a similar distinction where they really do prefer certain states over others, and that, because those states are experienced, then yes, a lot of these ethical questions come rushing into view. How can we coexist in a healthy long-term way with these cognitive systems of our own creation? So we're at a crossroads. Do we limit AI's functionality and perhaps consciousness to maintain complete control of it? We want AI to be useful, more and more useful, versus one AI to be completely out of control. There's a trade-off. So how much functionality do we need? And with that idea of AI consciousness in mind, is it too late to pull things back? Because almost like the right analogy is akin to parenting. Like if your kid can have a positive or negative experience, the solution isn't don't have kids or shut the kid off, that's not the solution at all. The solution is try all else being equal to be a good parent and not a bad parent. Where we go from here is yet to be seen. We'll be watching. This has been a special edition of the Texas Standard, the new Wild West Texans on the AI Frontier. Our reporters were Wells Dunbar, Shelley Brisbane and Michael Marks, with production work by Sarah Ash and Sean Saldonia. I'm Laura Rice. Let us know what you think and find a stunning digital presentation of our program edited by Raul Alonzo at TexasStandard.org. Support for Texas Standard comes from Texas Mutual Insurance Company, a Texas-based workers compensation provider, committed to providing care and support for injured employees. More at texasmutual.com/texansdeliver PhilanthropicSupportForTexasStandard, Co. Casey and Scott O'Hare, the Winkler Family Foundation, Lynn Dobson and Greg Voldrich, Adrian Killam, and the George Huntington Family.