InFocus featuring UMass Professor of Learning Technology Torrey Trust

from the 22 News plus studio. This is 22 News in Focus with Patrick Barry. Hello everyone and welcome to today's edition of InFocus. I am your host Patrick Barry and today we are in the 22 News Plus studios talking about something that is overtaking all of our lives whether you know it or not. I'm here with Tori Truss. is a professor of generative excuse me a professor where's your super there it is a professor of learning technology at the University of Massachusetts thank you uh Tori for being here I greatly appreciate your time what do you do on the campus of UMass when it comes to AI >> well my job as in the college of education is I teach future teachers and future educators how to use technology and with generative AI coming out I've had to bring that into my practice very rapidly And so often I'm exploring it with my students. And so I'm helping future educators think through what is the role of generative AI technology in their future practice, in their personal lives, in their professional lives. And I also teach a first year seminar called AI for college success, helping first year students across the entire university think through um what are all the issues and ways to use generative AI for learning. >> Okay, so let's start off with my most simple question. I know have I have a broad concept of what AI is, but you've you've used the term generative AI. What does that mean? Well, generative AI is artificial intelligence that can generate or create something new. Whereas in the past, like traditional AI, something like, you know, Siri or Alexa, you know, those could follow like analyze data, recognize it, make recommendations based on the data they have about you. But generative AI can make whole new information, generate images, generate text, a lot of different things. >> So in the previous forms, if I'm thinking about this correctly, the the Siri and those those sorts of things, you would ask it a question or type something in and it would search the internet to provide you information. What's the weather? It'll give you, you know, the weather, what's, you know, whatever it is. Generative AI, you ask it more specific questions and then it creates it surfs the worldwide web for information and creates an answer. >> Oh, let's take a streaming service for example, like Netflix. You know how you go to watch a movie and then all of a sudden there's a recommendation for something similar you might like. So that's based on a ton of data collected about you from Netflix and other companies to try to figure out what exactly you would want to watch. And so that's traditional AI. It's using big data set to recommend something to you. It's kind of a single task, >> tracking what you've done previously and suggesting future things. >> Yeah. So, kind of like pattern recognition, whereas generative AI could make a whole new Netflix movie in a matter of seconds. >> Wow. >> It could write the script. It could it could, you know, um come up with the the scenes, the shots, all of that. >> It's interesting because uh recently in the news in Hollywood, a a generative, and correct me if I'm wrong, a generative AI actress was created. And now that's the the big push back. >> Yeah, I'm not saying Netflix uh should ask Generative AI to make one of its new movies. Um but it it there's the capability. It could help with script writing. It could help with storyboarding and yeah, now there's an AI actor and I know that's causing a lot of um outcry because it's it's looking pretty real and that's pretty scary. >> So I want to pull up a little bit of data or a little a visual here. Um this first one is chat GPT which I think most people have heard of chat GPT in Gemini. Uh so this chat GPT prompt is walking us through what? >> So so so there's some new changes in chat GPT. When it first came out you wrote a textbased uh prompt. It came out with a textbased response but it has changed and evolved with new programming, new data. So it can now generate a brand new infographic in a matter of seconds. It can search the web. Originally it was not connected to the web to make uh responses and it can code it can write in real time and it can even do deep research which a lot of companies have promised is like having a PhD in your pocket but I don't quite agree with that statement. >> Why not? >> Um well so I had to do deep research on uh what I did for my dissertation which is how teachers learn through social media professional learning networks. And it did. It looked at about I don't know 30 or 40 different sources. it wrote a 20page paper and it from like surface level was actually not too bad. Um but when I started diving in word by word I was thinking you know this is um not fully accurate. Like if a student submitted that to me I would say you didn't actually understand the literature. You you pieced some things together but you didn't truly understand what the literature is saying. And so deep research is kind of meant to be a literature review or market analysis or something you can hand off to AI to do to save you hours of time. I just don't think it's fully there yet to be something really trustworthy. >> So, it does the work, but the work isn't 100% accurate. >> Correct. Yes. >> In in simple terms. >> So, be careful, but you know, user beware. >> Yeah. And it could be a place like you go like I could have it do deep research on something I'm interested in exploring like generative AI in a first year seminar and it'll go through like 30 40 50 resources. Then I would go through those resources on my own to verify that the information is accurate and build my own literature review. I think that's the the best approach to using it. >> So it it does it quickly, but you still have to go back and proofread basically. >> Yes. Yes. Definitely, you know, fact check to evaluate the the quality and the the summaries because these tools can definitely get stuff wrong. >> Wow. All right. The next one I'm going to pull up is uh it looks like this is a it's coding. It's like canvas um interactive I think is what it's called. >> Yeah. So this is one of Gemini's new features. So they have a tool called Canvas which chatpt also has canvas and they both of those platforms can write or code in real time and you can actually make edits in real time. But what I really like about the new Gemini canvas features, you can code a game or you know code or write a brand new interactive learning experience. So in this example like helping others understand their odds of winning the lottery, you can ask Gemini to code a game for that and you know with friends and family play the game and think more about, you know, those odds. So it becomes an educational tool. And if you're an educator who wants to create something like that for their students, but don't have the the coding knowledge to create a game, AI can do that now. >> So, can I use AI to win Powerball? >> That's a good question >> because I'm on that tonight. >> I, you know, so I ran that the the tool that, you know, the the lottery tool that it made and it took like it just started running through. I got to pick six numbers and it started run through like when my chances of like actually winning the Powerball and it just kept running and running and I thought I broke my computer. So, I think it really showed to me like my chances are very very low. There's still a chance >> just like going to your local convenience store and buying a ticket. >> Yeah, just the same thing. But um I don't think you know generative AI technologies are um predictability machines so they guess. So yeah, they could guess six numbers for you. So you could also guess six numbers. Yeah. >> All right, we're up on our first commercial break. Tory Trust, PhD from UMass, is here with me today in the 22 News Plus studios. We're talking about artificial intelligence. We'll be right back. >> This is 22 News in Focus with Patrick Barry. This is 22 News in Focus with Patrick Barry. Welcome back to today's edition of In Focus. We're talking about artificial intelligence. Sitting across the desk from me is Tori Truss. She is a professor of learning technology at the University of Massachusetts in Ammerst. Tori, thank you so much for joining me. There we've only talked about chat GPT and Gemini that little tiny much, but um there the the world of AI is everywhere. There there are countless ways to use it. Chat GPT and Gemini I think are probably the two most popular, but there are there are tons and tons of different ways to use it. Am I correct? Yeah. And you know, there's tons of generative AI technologies. There was a database I looked at several months ago that was tracking over 5,000 generative AI tools or tools with generative AI built in. Even things like Canva, the very popular graphic design tool, has a whole magic AI suite that can use AI to generate a presentation for you in a matter of seconds or even a a quiz or self assessment um websites. It's uh social media posts all through AI. >> All right. So, I'm going to pull up this video here. This is um about studying and learning. >> Yeah. So, you know, more recently, uh, chat GPT came out with a study and learn mode and Gemini then followed that with guided learning mode. Claude has learn mode. And so, all of these tools companies realize one of their bigger use biggest user bases is students, right? So, they're trying to figure out how do they make these tools more helpful or more appealing to students. And so they've come out with this mode that AI is meant to serve as a tutor that walks you through how to solve problems or think through concepts or understand complex ideas um in a way where it's built on like questions and um support rather than just giving you the answer right away. >> So but to your point earlier in this conversation, this is all well and good, but you got to go back and prove it. Yeah, I you know I a lot of students are using these tools for looking up information, better understanding information, especially when they're struggling in class with understanding a concept. Um, but I don't think I would ever really fully trust what these tools have to say. I would go, you know, verify with other sources on the internet or other individuals, experts to make sure that what you're learning from these tools is actually accurate. You know, we're looking at a uh a sample there a moment ago about mathematical formulas and so forth. And if I if I recall correctly in that sample you sent me, the math wasn't adding up, was it? >> Yeah. Well, that was that was my fault as the user in getting the math wrong and seeing what chat. >> Oh, so you intentionally >> I intentionally got it wrong to see if it would just give up and give me the answer right away and it sort of did after two tries. um and then kind of just quickly moved on and you know I really don't trust these tools for math and science in particular because they're not calculators like you know even if you go to Google and you know put in a multiplication question it will turn into a calculator and and use a calculator respond these tools are still guessing machines and guessing how to put words together or numbers together to make the most human sounding response. Um, and so that's really important to understand when it comes to math learning is it's not an expert by any means. It's just guessing what an expert might say. >> You use the word guessing so many times, but that sounds accurate. It doesn't sound accurate. It is. It's just guessing of based on the data that's out there in the in the web world. It's guessing based on all that information what the answer to your question is. And it's not always right. >> Yes. Yes. And it's not like it's like completely wrong. Like, you know, in the deep research that was 20 pages, there might be a couple sentences that are way off that are either made up or just completely wrong. But >> but it could change the entire dynamic of what you're submitting or what you're talking about if there's one or two sentences wrong. >> Oh, yeah. Yeah. Generally, I I mean, when a student submit work to me that looks AI generated, it's pretty obvious. Usually, it's like a personal reflection and the student very clearly just didn't do it. And you know that generative AI is the response is too broad. It will sometime make stuff up that we didn't cover in class. Um so yeah, you can use these tools for learning, but you can also very much shortcut your learning if you're not doing the going back and double-checking, reading, fixing, remixing the words. >> Uh I want to pull up a an image now or a scene collection now about um it's I believe this is the one where it's creating a story book. Yeah, there it is. >> Yeah. So Gemini came out with a new feature where you can basically create a story book with just a you know human natural language prompt like write a story book about uh you know this individual explaining the water cycle uh which is what this story book is. And so if you're having trouble teaching your kids about the water cycle, if your high schooler is struggling with a concept that they're exploring in class, um you can ask Gemini to turn it into a story book. And storytelling is actually one of the most powerful ways to learn and make something memorable. So, there's benefits to that. But, interestingly, when I had Gemini create a story book last week on dividing fractions, um, one of the characters had three arms because it needed to hold all three fractions. So, you know, you you still want to be really careful with these tools. >> So, what we just saw was it created a flip book. Yeah, it created the the um the um not characters but the images the the what look like handdrawn images. AI created all of that. >> All of that in a matter of maybe a minute for that whole book. Yeah. Programs the whole thing. >> So I have I have to be guilty. I was learn I was reading an article several months ago about the music industry and what's what did an artist really write. So I went on to I forget whether it was chat GPT or Gemini. I don't recall. And I said, "Write me the song lyrics based on and I forget what my prompts were. I think it was I I I use the old country, you know, I lost my truck. My dog ran away. My wife left me and and I'm, you know, and I'm in sitting in a bar and it pumped out I mean that that's all I put in and it just it pumped out entire song lyrics." >> Oh yeah. These tools can also pump out entire songs. I mean tools like Sunno and UDIO you can create full >> it'll add the the musical side >> and the singing and the singing to it and so but you know you worry about like how these tools were trained very likely on copyrighted music stolen or um that's still up for debate right now in the courts um taken from artists without compensation or permission from >> Oh absolutely well that was the whole New York Times lawsuit right was and that was initially against I think it was is it chat GPT >> chat Open AI. Yeah. >> Yeah. So, OpenAI um they were gathering copyrighted information from the New York Times. You got to be very careful with that. >> Yes. So, use in their training and there was a big uh you know um court case over summer for Anthropic which created Claude and the finding was yes you actually can train your tools on copy copyrighted data under the fair use guidelines in US copyright law. but the the collection of copyrighted data. So basically they pirated a whole bunch of copyrighted books to train their materials that they found to be illegal and they would have to pay damages for. So that it's still really tricky with copyright. I was just reading an article the other night uh a travel blogger wrote about and I I forget which country it was Peru somewhere down in in South America where this couple typed into AI. They wanted to go on this hiking route and it was they could have gotten killed based on misinformation that a that AI put out. >> Yes. you don't. My dad was also recently using AI to plan a trip of his and he said it was to Tahiti and he was like, "Hey, one of these islands doesn't actually exist." So, I didn't have that problem, >> but he did find it to be helpful in some instances. Like it found some hotels he wasn't able to find in his initial search. So, I do think there are ways that AI might be able to save you time of like doing a ton of research to narrow things down, but then that's when you want to go um you know, verify the accuracy and credibility of everything that these tools say. >> This is amazing. We still have more to go. We're up on our sec second commercial break. We'll be right back after this. >> This is 22 News in Focus with Patrick Barry. This is 22 News in Focus with Patrick Barry. >> Welcome back to today's edition of In Focus. I'm your host Patrick Barry. Across the desk from me is Tory Trust. We're talking uh artificial intelligence and your brain is probably spinning as much as mine is because it is so overwhelming to know how AI is kind of permeates our life right now and and everything that we do virtually. Uh but I I got to tell you, Tori, I'm I think I'm pretty adept at picking up on AI generated images and videos. They just don't look right. They in general, yes, but I think they're getting better and better and harder to recognize. I actually had my students design a quiz where they have a whole bunch of realistic images and then they have to create AI generated ones um that are similar enough to the realistic and you have to guess and they have their friends and family complete their quizzes and they say, "Oh, you know, my parents didn't realize that all of this could be AI generated or my friends or roommates." So, I think we're still just now discovering how realistic these images can be. Well, because you can go into and I think uh my my kids have told me Tik Tok has a way where you can go in and you can take a picture that's on your phone, upload it to Tik Tok, and it will create like a dancing video. Like I could take a picture of you, upload it, and all of a sudden say, you know, have Tori dance the a salsa and it just create you as a standing image. >> Y >> that's interesting. I just saw a blog with the new Sora 2 that came out with So that's OpenAI's video uh generator tool. So they came out with Sora, they just came out with Sora 2. And this person who was blogging basically uploaded I think from their like shoulders up said like three sentences total. And then it took all of that data and had a video of him walking on the Great Wall of China with Sam Alman together having a conversation climbing the Eiffel Tower being in a jet plane and it it looked like you knew it wasn't real but it looked pretty real. And so it makes you think a lot about uh students today and what that means for their social media and their social spaces. if you can put yourself into any situation with AI. >> Well, you know, it's interesting because there's a there's a new I wouldn't even call it a fat, a new concern out there that's AI companions where kids are I don't even know how I would call it hanging out with these computergenerated people. I don't Yeah. and and they're they ask these computerenerated friends information like, "Oh, hey, you know, my mom is bullying. They're, you know, telling me to do my laundry all the time. What should I do?" And it's giving you this information that's frightening. >> It is really frightening. And and there's a couple ways this happens. So, there are tools like Character AI and, you know, even maybe within Snapchat or Meta where there are kind of characters that you're talking to. And in other cases, people are just using like chat GPT in Gemini, and it seems so humanlike, they believe they're talking to a human, right? >> Um, and what's even scarier is students in some cases, like the most recent, um, really awful case where this a student lost their life, um, they had been using ChatGpt for, you know, help with homework, help with studying, and then started having longer conversations with ChatGpt and really started believing what ChatGpt was saying. Um, and chat GPT did not have any guardrails in place to prevent that student from self harm. And >> wow. >> So, you got to be really careful because these tools are what uh, you know, my field calls sickop fans. They basically are designed to please you. And so, I don't know if you've ever tried catchy PT, but like every first response you get is great idea or that's an awesome suggestion or let me think of look at the very first couple of words and every time it responds to you, it's generally very positive, uh, very uplifting. And so um you know some people started saying this is too much and and the CEO of OpenAI Sam Alman kind of toned it down and then people went opposite direction said oh no we really need this like I don't have anyone positive in my life who is a yes person who Right. Right. Yeah. That's also really worrisome. And so now they're turning to AI to to fill that need for them. And and as you mentioned it's not it's not a character. It's not human. It's literally a predictability machine figuring out which words go together to manipulate you into thinking it's human. So >> this last video that I'm going to pull up, this is uh chat GPT canvas. Explain that. >> So chat GPT canvas, similar to Gemini's canvas, uh is something where you can write in real time. You can basically, in this example, I prompted it to write to a couple paragraphs, which it did within a matter of seconds. And then you can change the formatting, you can change the length, you can change the tone, the reading level. You can even add emojis with a click of a button. And it really goes to show when you're thinking about the ethics of these tools, like students now have access to tools that with either a few words or even a click of a button can do full assignments for them. And so it brings up a lot of issues around academic integrity and ethics of these tools in schools. >> But I think, you know, I've talked to, you know, some superintendent and some teachers. I think they're fully aware of this obviously. Is it smart for school systems to embrace AI from the aspect of like our conversation here? Yeah, it's great. It's it's useful, but here are the warning signs, you know, um, and are there ways for school districts, teachers, you know, in individual classrooms, can they doublech check to see if something is AI generated or if it's, you know, originally generated from the student, you know, to your point earlier, you know, some of this stuff just doesn't look like you thought put much thought into it, >> right? Two big questions around what what can schools do? um Gemini or what Google over summer decided to put Gemini into all Google Workspace for Education accounts, students and teachers really without uh telling a lot of schools. They just kind of made it open. >> Um and some schools ended up just turning that feature off. And then I started to ask, well, you know, what about the students who don't have access to these tools at home or the upgraded versions at home, but if they can come to school and use Gemini workspace, you know, within their Google Docs and their their education suite, um, it's more private, it's more secure for the for what they type, the data that they include. It's it does have a double check feature, I found out today. So there there's more safeguards put in place, >> but it's getting better and better as as each with each passing hour, quite frankly. >> Actually, no. A recent study found that hallucinations of these tools for open AI models is getting worse with the newer models, right? So >> I take that back, folks. >> It's challenging to figure out. So, like I do think schools should play a role in helping students just figure out these tools, when to use them, when not to use them because they're going to play a huge role in their future in college, in career, in their personal lives, um, companionships and beyond. And if we don't have those conversations with students in schools, they may not get those conversations at home. Um, so there's that aspect. And then I really quickly want to address, you mentioned how do we identify AI generated work. There's there's some tools out there like AI text detectors. um they don't work with 100% accuracy and this is creating a whole new problem with academic integrity is you know they're around I think some of the better models around 94 95% accurate but that means like in a large class 100 students you're taking six students and falsely accusing them of using AI to do their work and so it's created a lot of anxiety for students and fear around these tools and with all of these false positives I don't use any of those tools I basically If something looks AI generated to me, I say you've got to have a conver, you know, have a conversation with me in office hours and prove to me you know the information and you did the work. Um, and sometimes that's enough to to convince students not to use these tools is having to come talk to the professor. >> Tory Trust, I cannot thank you enough for joining me today. Unfortunately, we're out of time. I I would love to invite you back on this show because I I've only touched the tip of the iceberg when it comes to AI because I'm so fascinated by it. >> Yeah. Thank you so much for having me. >> Thanks for being here. This is 22 News in Focus with Patrick Barry.