Conversation with Dr. David Zaretsky - AI in Education Part 2

In part two (2) of a conversation with Dr. David Zaretsky, Dr. Tony Rhem, and Northwestern professor Dr. Zaretsky, explore one of the biggest questions in artificial intelligence today: as AI tools, AI agents, and large language models become more powerful, what still requires human judgment, critical thinking, and ethical oversight? They discuss AI in education, AGI, human-in-the-loop decision-making, AI guardrails, AI ethics, AI governance, autonomous systems, military AI, coding with AI, software testing, red teaming, and the risks of allowing AI to act without proper review and control.
This conversation is essential for educators, business leaders, technologists, students, and anyone interested in responsible AI adoption. If you want insight into AI governance, AI safety, critical thinking in the age of automation, and why ethics must be built into AI from the start, this episode delivers a grounded and practical discussion on the future of work, innovation, and trustworthy AI. Like, comment, subscribe, and join the conversation on AI: past, present, and the possible future.
#ArtificialIntelligence #AI #AIEthics #AIGovernance #AGI #CriticalThinking
#HumanInTheLoop #ResponsibleAI #AISafety #MachineLearning
#FutureOfWork #AIInEducation #GenerativeAI #LLM

Key themes: AI and the future of work, AI Strategy, Generative AI, AI governance

Transcript

Copy Transcript Copied

SPEAKER_01 0:04

Welcome back to reminiscing. This is part two of a conversation I had with AI professor at Northwestern University, Dr. David Lareski. And we encourage you to listen to part one for more context on AI and education. Don't forget to like, subscribe, and continue the conversation at reminiscing.

SPEAKER_00 0:23

In all of my classes, I tell my students like the first day, there's two things you need to succeed in this world today. One is critical thinking, and the other is you have to be intellectually curious. If you are not willing to step out of your comfort zone and see what else is out there and explore, and you're gonna fall behind very quickly, right? Um this past week or so, uh, there's been this huge thing uh all over the internet on this tool called um uh clawbot. So uh, you know, basically people are running these on their Mac minis, right? You just install the uh clawbot and it you give them access to all the APIs and stuff and it can start running your company for you. Uh set uh all these different skills and stuff it can learn and start running and people are deploying this to run their businesses, all these different agents and stuff to optimize. So I mean it just shows you how quickly things progress, and you're gonna have a lot more tools like that that become this is at home and stuff, but you'll have enterprise versions of this where you can run agents, uh, you know, uh the the the simple things like um you know, uh uh do posts for me on LinkedIn or you know, create some content for me for YouTube and stuff and push that out. Save you a lot of time and and uh help you gain traction and stuff. But if without those critical thinking skills, without the being intellectually curious, you're not going to be able to run the company. The agents are not going to uh there was a a podcast uh I had watched a while ago. This guy was basically setting up all these agents to run a business and just you know woke up the next day and they're all just chatting and they didn't do anything. There was no direction, it was just you know, go figure it out. And uh you need you need that uh somebody orchestrating it.

SPEAKER_01 2:21

So absolutely, absolutely. That orchestration layer layer is is the person uh as the human in the loop, yeah, so to speak.

SPEAKER_00 2:31

Where does that come from? Um and and I think it it really stems from um you know the the human experience. You can have you know an AI come up with ideas to start a company, but what it doesn't have is that that life experience, a human experience. So it doesn't know what problems to solve until you say, hey, I was um you know, I was on a podcast and uh I you know I had this great conversation and my mind started thinking about this problem I had and now I want to go solve it, right? The AI can't do that because it didn't live that experience.

SPEAKER_01 3:08

Absolutely.

SPEAKER_00 3:09

And so you do you do still need a human in the loop to figure understand like where the problems are, what problems will be coming up in the future, and right? Um I think uh that's that's still uh uh gives us a lot of hope that we're not fully fully being replaced. Right.

SPEAKER_01 3:24

But you know, one of the things I have read recently is um how uh AGI is is is is now possible, right? Artificial j artificial general intelligence and how that is going to um you know run services and products and this, that, and the other, and going to displace people, and then there's some futurist uh that's speaking about you know 99% unemployment. I mean all all of this. But you but you take a step back. If there's no one employed, who is who is actually buying these products and services that are being created? Right. You know, what what you know what what's the economic model for that?

SPEAKER_00 4:08

Yeah, which you know is the that's I think you know humanity is interwoven into the universe. You can't just disconnect the human from everything, right? I I think that's that's a very foolish way to think about you know uh everybody's just gonna be sort of you know disappear uh and and not work. And it there's always innovation, there's always things happening. The the analogy I like to bring is um, you know, uh you saw uh do you see the wicked movie? Okay, so uh for those if for if you saw the wicked wicked in uh in the theater, uh like on the play, um versus the one in the theater, the theater had all the special effects and it was great, and you know, um it was this whole other experience. But when you go to the theater and you you see it's a one-take, right? All these people are practiced and they're singing live without microphones, without um all the special effects, it's all done manually with levers and stuff. There's this authenticity when you go to a Broadway show that you can't get in the theater. That and and uh look I'm for me, I'm I'm I would call it an audiophilist, right? I love vinyl records and I love collecting old vinyl and stuff. You can listen to Spotify, but it doesn't have that analog, pure audio. And and so I think there's still going to be a place in the world for you, you know, you're gonna have these digital auto automations, and then you'll have companies where it's like you know, people are still, you know, building art from hand, you know, by hand. They're still um doing things by and people will pay a premium for that human experience. Absolutely. Yeah, yeah. So I I I that has never changed, you know, with all this Hollywood effects and stuff. You're seeing a decline in movie going experience, partially because people can see it at home or they wait to it, but Broadway is still, I mean, try to get tickets for a Broadway show, it's it's it's hard, right? They're always selling out and they're you know, people are making good money in that industry. So I I I think that there's still uh you know something to be said about that analog uh authentic experience.

SPEAKER_01 6:25

Yeah, the the authentic experience will be at a premium as we continue to evolve this AI AI world that we're that we're in. But I always say just because you can do it doesn't mean you should do it. Right. You know, and and look at the the broader impacts and understanding uh where to leverage it. Like we talk about AI and education, where to leverage it best, how how to how to structure it better, and and looking for a specific outcome from it, not just use it and be like, oops, I didn't think this is gonna happen. Right. Kind of like you said, you you're you're futurist. You're looking you're looking in the future on how things are going to transpire. If I do X, will Y happen? Or what do do I need to do to prevent that Y from happening if it's not something good? Yeah. Right. So it's it's those things we have to take a look at as well.

SPEAKER_00 7:20

So Yeah. You know, that's always been the Silicon Valley model is you know, break it and then apologize later. And uh, I mean that's where a lot of innovation comes from, is just you know, push things as far as they can go until somebody says stop, you know, or and uh you know, laws take a long time to pass. That's one of the you know the interesting things we talk about in the course that I that uh you spoke in uh about ethics. Ethics is always forward looking, where legal is always backwards looking. You know, so you you legal you have a framework, and ethics you're always thinking, do you should I be doing this? Absolutely, and uh, what are the consequences? And so that takes a lot more uh critical thinking because you know uh the the engineer or the entrepreneur's sort of mindset is just is you know, push it as as fast as quick as we can do it, be the first ones without stopping to think about should I should I actually do this or not? And uh, you know, I think there, you know that that uh passion for innovation, you know, that is is always at odds end with um the ethical implications, and that's where you know you're gonna run into this AGI. How quickly are we gonna get to AGI? I don't know. I'm always thinking it's five years out. I'm I'm pessimistically optimistic.

SPEAKER_01 8:50

Yeah, it's it's in some cases I hear it's it's is is already here. Some some cases people saying it's three to five years out, and other people say we have a form of it now. I mean, like what exactly is happening? But at the end of the day, I believe if you are intentional about integrating ethics and governance in your development of AI, we it it's not something we you you know, like the old model, like you say, you push it out there and then you kind of fix it later, you know, the the toothpaste is already out of the tube. So so how do you need to bake ethics into what you're doing before the product is released? Sure. And as we see, once the product is released and it's you know has some data ethics issues or just the processing ethics issues, it comes back to a point where it can um have a situation where the company is so affected by it that it's it the company is no longer in existence because it's you know it's too many lawsuits and things like that, and and no one trusts the product. Right. So people stop purchasing it. It's it can be the death of an organization if you don't do this, if you don't integrate that within your technology. So it's those things we have to make these companies aware of.

SPEAKER_00 10:15

Right. You know, in engineering, we always say when you're building a product, security is not a feature. It's it's you design with security in mind, right? From the get-go. You're you're all everything's already designed to be encrypted, and you know, you have passwords and right. Same thing with ethics. Ethics can't be a feature anymore. It has to be designed in there. You know, they lowered the guardrails for a lot of these LLMs. A couple of weeks ago, uh, you had uh Pete Hexeth was in um uh where where was this? They had the big uh AI.

SPEAKER_01 10:53

Is it is it Devils? Oh no?

SPEAKER_00 10:56

Uh I'm trying to remember. But he was out there with Elon Musk and he was announcing all the new the new uh uh um you know uh military uh and do and all the all the stuff that they're doing with AI and stuff. And one of the things that they were you know ta talking about was we're not going to be using any sort of AI or that has certain guardrails um you know in place because you know in the military, you know if you there's a whole ethical, you know, uh conversation around should you be using AI in in the military anyway, but you know, uh if you if you had certain guardrails in there that would prevent the AI from taking certain actions that could be you know problematic. So naturally they're either retraining models to for specific military use, or these contractors like uh Google or you know, OpenAI, whoever they're using, uh wouldn't have to have certain models that where they lower the guardrails. And we've seen kind of what happens when this you know, there was a few incidences where kids have committed suicide um because the guardrails were reduced um allegedly, and uh and so there's um yeah, there's a lot of c you know considerations where now if if those guardrails are reduced and I build my technology on top of an LLM that doesn't have that, I uh it's not that I can uh I can't be relying 100% on OpenAI to uh have all the you know all the check boxes in there. I have to build my applications with my own guardrails in place.

SPEAKER_01 12:36

But you you also have to look at the application, right? If it's a military application, there's certain things that you know you you don't want as far as guardrails, but certain things you do, especially if you are going down the path of autonomous weapons. And and and the the uh you run the risk of the AI taking over, launching things that you should not launch at a time you should not launch it if it's not a human in the loop. Right. And and so it's certain other guardrails that you got to make sure you have in place. But that same technology, if it's if it's used uh everyday life or somebody is working with it with kids, there's a different set of guardrails that you need to make sure are in there. So it's it's really depends on the application of the AI that that you're using, what guardrails you need to have in place. Yeah, and that's common sense to me, that's common sense. Right.

SPEAKER_00 13:27

Granted, I don't I don't think that they're using it to launch weapons, but I think what they're probably using it for would be uh assessments of like, you know, taking all the data and distilling it down and coming up with assessments on like, you know, what is the chances of X, Y, or Z happening and you know, yeah.

SPEAKER_01 13:44

Well, you know, they that back in the day when that that movie War Games, when you you had uh you know simulation. Right. You know, uh and there's no really real winner. Um so it's it's a lot of different aspects of AI and that that could be really useful and and pertinent within the military and what they specifically want to do. And then there's AI and for the general public, AI for organizations. And so if you like you said, you have to develop it with the ethics in mind, security, like you said, the security with in mind, those are the things that you need to develop it with in mind as part of your requirements, testing, training, the whole nine. Yeah.

SPEAKER_00 14:22

When you do that, anything that touches a human, I would can always consider. Do you, you know, do you have guardrails about, you know, can somebody ask a question that is it's not designed for? Right. You know, if you're building a gaming chat bot or something, right? Somebody asks about drugs or suicide or something like that, those should be topics that should be off guard off uh limits, and then you redirect or or so on, right? Um so people have found ways of getting around it, and uh um you know even uh games right now where you know, Roblox and stuff. There's a lot of parents who are removing Roblox now from their kids' iPads because of all the scary stuff that goes on with that. And um, you know, so you you you gotta be vigilant as a parent, you gotta be vigilant as an entrepreneur, you know. Exactly. You can't don't take anything for granted that it's it's designed with all the safety precautions, you've got to build in yourself.

SPEAKER_01 15:25

Yeah, I I want to get your opinion on this. Um the you know, large language models, you uh AI has is been very good at developing code. Whether it's Python, C sharp, whatever. What are you what is your your your belief on just letting the AI develop the code and just installing it? You know? Is there uh I think there should be steps like any other software development, you know, F.

SPEAKER_00 15:58

I come from uh uh my background is in compilers and uh compilers the idea is you know taking high-level languages and you know uh compiling down to executable code. And I look at LLMs as being the new generation of compilers, basically take an idea and distilling it down to code which then gets executed. And you know, uh now the LLMs also have the ability to uh not just create the code but execute it, run it, you know, you give it access to your computer, to your email, and they can go ra you know, rain havoc.

SPEAKER_01 16:37

Right.

SPEAKER_00 16:37

I think that's why, you know, one of the big you know uh hurdles with this clawbot that everyone's talking about now, this application, is that you're giving it access to do anything. And and uh it could send out emails on your behalf, it could you know, it could be quite catastrophic if it publishes something or erases all your files, or what if somebody attacks it and from the outside sends an email and the the LLM responds back and sends all of your information to it, sends your passwords. So you you know, uh there could be a lot of even beyond just you know installing code or running it, uh it could it could very easily uh do its own thing. And and you know, even even using your AI chatbot as a a host, as a co-host, listening in with you telling not to uh talk, you know, and it still will operate and do its own thing. Um is you know, very simplistic example of how uh it doesn't follow all the directions that that you're given.

SPEAKER_01 17:43

All the time. Right. So yeah, yeah. And and so that that that's that's my that's my angst. I mean, I kind of like, you know, I I I would use the a a model to generate code, I wouldn't install it until I reviewed it. Yeah. Until I tested it. You know, you you don't just write, I never wrote any code and just installed it. Right. You would we always tested it, you know, from from basic unit test to string testing to to integration testing, all those levels of testing. Yeah. And and and then, you know, this it's you know, you do red teaming and really hack at it, throw data at it, does that it shouldn't, you know, um, you know, see how it responds. All those things are important before you so you know what you what you have before you release it. Right. And and people uh are not doing that because I see folks that are really not versed in programming and developing systems, just letting, you know, like you said, giving the code access to different systems without first testing all of this first.

SPEAKER_00 18:51

Right.

SPEAKER_01 18:52

And I think that's that's that's a myth.

SPEAKER_00 18:54

Yeah, you know, um most of these uh tools like um Visual Studio Code and um uh what's the other one now that I use uh Windsurf, they they don't really handle large context windows all that well. So what ends up happening is you say write me some code or fix this code, and what it ends up doing is changing a lot of stuff because it doesn't have the you know, it doesn't contain the full context. And so what ends up happening is you think you have like a perfectly written code, it rewrite starts rewriting everything. And uh so you gotta keep your files, you know, it was one of the tricks of the trade, you gotta keep your files pretty low and pretty small and and being more modular. But um but yeah, yeah, it could have inadvertent effects if you just have the LM just write or you know, start editing everything, it could insert some garbage in there that breaks, and yeah, you gotta have the unit tests, you have to have a process for uh developing code for testing for you know pushing it out. And um, but uh who who's gonna do that? Somebody with hopefully the critical thinking skills.

SPEAKER_01 20:04

Here we go, here we go back to that. Yeah, and and and you know, we're we're old school, so we this is stuff we do, and that's I call it software development one-on-one. This is stuff we do. And it's like, have we forgotten this? Or people just say, you know, the this this current generation, they say, well, the L AI is gonna do it for me. You gotta make sure that AI understands what it what you want it to do. So it's those concerns um you know that that bother me, you know. So I I I know we covered a lot. Um, and um, you know, so what else? Tell me tell let's talk about the things you're doing. Let's just get into let's get into running. So testing and and and understand uh you know the things that you're doing. Because I've always found you to be very fascinating, interesting person, uh very um uh understanding of of of the trends and and someone that has a a great perspective. Yeah. Because every time we um I'm speaking in your courses, we have conversations and the students then benefit and the students jump in. So so talk about more of what you're doing.

SPEAKER_00 21:19

Well, I I'm I'm always uh innovating. Um in the back of my mind, I'm thinking 10 more things. So, you know, I uh I often struggle with balancing life and uh and and innovating. Um because I there's there's always something that I get really, you know, interested in and I I wanna, you know, attack it. So I I have uh you know, right now I'm working with a bunch of different students doing research and uh you know, testing different ideas and and so on. But uh, you know, my classes are um I'm always updating them. Uh so I I have that. I'm uh trying to uh uh improve our our master's program. We're trying to create new opportunities um for for students, trying to help them get jobs. And a lot of that comes down to uh rethinking the role of a master's program for students, you know, making sure that there's an you know uh opportunities for them. Right. So my courses are often evolved around like how do I market myself, get the best skill sets and things like that. So I most of my day is spent you know working with students. I I I I teach probably twice twice a week but then I spend the next four days meeting with students, guiding them, whether it's on research or projects or or things like that. I have students who are becoming entrepreneurs. So I'm working with them. And then I have a coup you know my companies on the side that I'm working on which are all have elements of of AI and and so on. I I do work in consulting a lot of it also is in AI. So I keep myself pretty busy uh trying to innovate and um you know the the things that are sort of on the on the uh um in the entrepreneurial space that I'm working on are still sort of under the radar right now while I'm still working on it. But uh there there's inevitably uh what I'm finding is you know I I I I I start going down a path and then there's like 20 new things that come out and you're like I gotta I gotta look at these things. All the AI everything changes so rapidly so you know staying staying afloat that's that's the big challenge but it's also one of my big passions. So my next uh big project right now is is launching my own podcast you know uh and and being able to just have these conversations it forces you to be be up to date on what's going on and having these conversations I think are really important.

SPEAKER_01 23:56

Um and so uh those are some of the big things on my my agenda right now but uh you'll be hearing about some of my launches uh hey hey I'm I'm looking forward to that I'm looking forward to that um and also looking forward to the next time I'm uh come to Northwestern and to oh yeah converse with uh you and your students so uh that's always that's always uh open open for you to do that for just you know give me a call or shoot me an email whatever and we can we can make it make it happen yeah you're always uh on that on the top list of of things uh that that uh part the AI ethics part is and the legal have really become more and more so uh um more interesting uh and and pivotal parts of those courses because that's you know I think that's where students lack you know engineers they can always pick up new skill sets but nobody teaches them the ethical or the legal components ramifications of their work and um so I've been stressing that more and more uh in these recently and so I I really enjoy that part of the conversation. Fantastic. Well I'd like to thank you uh David I really appreciate your time and your willingness to come um speak to me today on the podcast I can want to return the favor to you oh yeah when you launch yours I'm definitely looking forward to that and um you know we go from there. Yeah thank you sir appreciate your time my man great to be here all right if you enjoyed this conversation don't forget to subscribe comment like and share because we will be diving into AI past present and the possible future soon