Ep 103 - "Human Judgement in the Age of AI" w/ Glenn Broder

Send us Fan Mail An AI answer can be polished, plausible, and completely wrong, and that’s where the real risk shows up. We sit down with Glenn Broder, founder of Learn AI Like a Pro and creator of Human Judgment in the Age of AI, to unpack a problem most AI adoption plans ignore: capability debt. You can gain speed and output while quietly losing the human skill to question, verify, and handle exceptions when the model fails. We get concrete about what responsible AI governance looks like i...

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Jason McGuinthe?

  Welcome back to Inspire AI, the podcast where we explore how artificial intelligence is changing the way we work, lead, and think, and what it means to stay distinctly human in the entire process. So imagine this, you have an employee who has been using AI every day for the last year. They're fast, their work looks better, they're producing more than ever before. Then one day, AI gives them an answer that is polished, confident, completely plausible, and wrong. Do they catch it? Can they explain why it's wrong? Can they find the evidence they need to challenge it? And if the AI can't help them, can they still handle the situation themselves? What needs to be true for that to work? Those questions point to something

• 0:45Jason McGuinthe?

  we don't talk about nearly enough. We discuss AI adoption. We're very good at measuring what AI gives us: productivity, speed, utilization, output quality. But we're not nearly as good at measuring what people may be losing along the way. My guest today, Glenn Broder, believes that distinction matters enormously. Glenn is the founder of Learn AI Like a Pro and the creator of Human Judgment in the Age of AI, where his work focuses on helping people learn to direct, govern, inspect, verify AI-generated work, while keeping human judgment responsible for the decisions that matter. His framework is built around a simple principle: AI produces

• 1:28Jason McGuinthe?

  work, humans produce judgment. And that leads us to the central question of today's episode. After AI takes on more of the work, how do we know whether humans can still recognize bad guidance and handle the exceptions themselves? We're going to explore what Glenn calls capability debt, the possibility that an organization can become more productive while its people quietly become less capable. We'll walk through a five-stage approach: capture, validate, baseline, delegate, and retest for determining what human capabilities exist before AI takes on the work, and whether those capabilities remain afterwards. We're going to explore

• 2:09Jason McGuinthe?

  the possibility that perhaps the real measure of successful AI adoption isn't simply whether someone can produce better work with AI, it's whether AI has made them more capable when the answer isn't obvious. The system gets it wrong, and judgment is the only thing left to rely on. Let's dig in. Welcome to Inspire AI, Glenn.

• 2:31Speaker 01

  Thank you very much for having me. I really appreciate it. I'm looking forward to the conversation.

• 2:36Jason McGuinthe?

  Excellent. I understand you're calling in from Chicago. What brings you here today?

• 2:42Speaker 01

  Well, and to talk about the human judgment in the age of AI and how important it is for the human to stay in this link and to stay really in control at every point.

• 2:56Jason McGuinthe?

  Right. Yeah, so I guess my first question for you to explain to the audience is about that human capability. When you have AI, companies measuring AI adoption and productivity, of course, the time saved, what are we missing in terms of metrics for the human capability?

• 3:15Speaker 01

  That's really where I think we look at things very differently than everybody else. And I'll tell you why. I think most companies have come to AI as plug and play, where they heard there's a great HR assistant or a great accounting assistant, and they plug those systems in and they make the humans conform to the system essentially. The problem with that is AI is intuitive as a probabilistic model. So if you give it the right information, it can run seamlessly throughout the entire organization, where plug and play has really

• 3:59Speaker 01

  forced the employee to learn the AI rather than the AI learn the employee.

• 4:07Jason McGuinthe?

  Okay, so talk a little bit more about what your model does differently.

• 4:14Speaker 01

  Okay, well, we start off with a concept that really the AI is completely untrustworthy. So start from that position, which I think a lot of people are really starting to understand. And the way we pitch it the most is don't get embarrassed, because people are getting publicly embarrassed because of not checking work and just starting to read off a report or a speech or whatever. So we feel that really at the very beginning, you have to let the AI know everything about that person and their position. So if you brought it

• 4:55Speaker 01

  rather than installing something and have it then work backwards where the employee has to work through that process, if you brought it to the employee and said, "Look, what we're trying to do is not fire you. We want you to explain to your AI what you do in specific, so that it can find the places that it can remove the drudgery, and then now that that's out of the way, we can make you highly efficient and productive doing the job that you really studied or learned to do." You didn't get into the position for

• 5:35Speaker 01

  the drudgery, you got into it for X, Y, and Z. We want you to be able to do X, Y, and Z as well as you can. Let's learn together, and then have the AI understand how you work, the things it should take over, and the things that you now have to get, use it as a cognitive jungle gym to really get the greatest advantage out of it. But if you come from the plug and play side first, you remove that possibility. The AI is not going to learn the person, it's trying to learn that efficiencies that it can do, and the

• 6:18Speaker 01

  person really becomes secondary. That's where we say you can't run systems that way that are probabilistic. You get into trouble.

• 6:28Jason McGuinthe?

  Right. So there's a, there's an interesting dichotomy here, one in which a company introduces AI because it looks like an effective use of company resources to help their employees and services run more effectively, but it doesn't know the actual humans that are using it versus the part where you're saying, teach the system about the humans, what matters most to them, so that they can do their job more effectively. So on the first one, you're removing the human connectivity, and you're creating some sort of like capability debt, if you will. I think you've used a term like that.

• 7:14Speaker 01

  Correct.

• 7:14Jason McGuinthe?

  So what does that mean exactly? And what would it actually look like inside of an organization where you're embedding that capability debt from the start?

• 7:25Speaker 01

  That's a big problem, especially to humans that are like to look for convenience and comfort and as little friction as possible and everything. That's where it becomes a cognitive wheelchair, in my opinion. The cognitive wheelchair is where the AI just starts taking over and the human goes, "Wow," and it just lets it keep doing it. And that's where those embarrassment points keep popping up, where people get embarrassed because they let it run that way. And that deficit that it causes is going to be impossible to regain, because once the system does it, now we don't have to hire people. If we want to double our

• 8:06Speaker 01

  size, the system can do it. I was very pro-human when it comes to AI, and I said, "That's ridiculous. You have something..." And my system that I'm on right now, it doesn't know that you and I are specifically talking, but it knows I have a block of time where we were going to have a conversation, and that conversation was going to be for the good of the humans. But other than that, it knows that that's what this activity is going on right now. So my system watches everything I do, and I give it control up to run, activate, start.

• 8:48Speaker 01

  It can never do any of those activities. So it does a lot of work that I used to have to spend weeks and months on, and now it gives me a report that I can now read, I edit, I ask it why it said those things, I teach the people to verify and confirm that final draft, and use it as a learning experience back to the AI. Doing that forces the cognitive jungle gym. The person has to keep their judgment in place, and they don't end up giving cognitive capability because it forces them in the other direction. If we leave

• 9:31Speaker 01

  it to the humans, truly, we are going to be in big trouble, because we always default to convenience and comfort. And this is a tool that makes it confidently seem like it can do that when we know that's really not true.

• 9:47Jason McGuinthe?

  Yeah. Well, tell us a little bit more about your perspective there. How do, how should we distinguish between healthy delegation to AI versus cognitive overloading that's gone too far?

• 10:01Speaker 01

  That's a great question.

• 10:02Jason McGuinthe?

  I mean, versus cognitive offloading, not overloading.

• 10:06Speaker 01

  The offloading, right?

• 10:08Jason McGuinthe?

  Yeah.

• 10:09Speaker 01

  The, which you're really, I think we have to look at humans in a very honest way. And when we look at humans, we are powerhouses. When we're directed to the things, especially that we have passion for. I mean, you see people accomplish things that are absolutely incredible. And I think we're going to have to use the AI in one way or another as also a training device. That's deployed by the C level, the high management, where they really lay out the direction. A lot of those organizations out there, and I go mostly after companies that are thirty to a hundred million.

• 10:51Speaker 01

  It's just a, a growth area that I think we could take over lots of the big companies if we employ AI correctly. And the executives and founders of the AI companies believe the same. And I think we have to make it a directive from the beginning, from the very top. This is the way it has to be done. We're not looking to replace you. We're human positive. So if they come out as human positive, we have to learn now to learn the human and learn the AI. And the two can learn it together. So there's things that we should not be wasting cognitive space on.

• 11:35Speaker 01

  Humans, if my AI, the way it recognizes things, it should take over. If I've done it five times this week, it starts making suggestions of I shouldn't have to do something five times the same and during the week. So its intuitive nature finds these points where I should offload it. But totally offloading it is the problem. Most people just start using it and they have no interaction with it. They just let it run, assuming that it's right. And I start most of my trainings this way with a very simple question. And I also ask this of CEOs all day, and I ask it of district heads, the superintendent of schools.

• 12:21Speaker 01

  At a ninth grade level, if we asked everybody in your organization, how does AI give an answer? And how is it so confident when it's wrong? I have yet to get an answer.

• 12:39Speaker 01

  It's a simple question. You're using this powerful tool and you have no idea how it gives output. How can you write a policy on AI if you don't know how AI operates? So I have to shake people. I kind of shock them because we're lulled into this complacency where we don't think about anything. And it does take some shock at the beginning. So I really, I'm, I'm out here every day trying to shake it up and go, don't do it wrong. If you do it wrong, it could really be a problem. In two thousand and eighteen, we were in the Wall Street Journal. It was talking about how China and the United States were in a war over AI

• 13:24Speaker 01

  and how they were trying to, both sides were trying to lead. Part of the story was talking about how China had already built this gigantic military complex and they had a bigger navy than us. And it was really going into the depth that they, the reporting was really strong. And AI was absolutely at the middle of everything. And as a kid, my father was an electrical engineer in top secret defense. So we never really knew exactly what he did at all. But I always had a sense for those things and different people in the family were in that.

• 14:02Speaker 01

  And in twenty-two, when OpenAI made their first model available, I got it almost immediately. And I knew that it was going to be something that I would have an advantage from. So I really threw myself into it. I put a lot of my other businesses on the side. And I said, this is where I'm going to focus now. I think this is really important. And the first L L Ms were really, really dumb. I mean, they were so unreliable. And I think I had a great advantage of jumping in then because I realized that governance and safety were going to have to be built into the systems.

• 14:45Speaker 01

  So as I progressed with the L L Ms, I built a, for a year I built the system and it became obvious it was not going to be able to be finished. The, the L L M wasn't strong enough. So I started again, knowing its logic and understanding, and I built a second model. And this was all with the publicly used L L Ms because I saw that as the place where if I was going to be looking at the masses, they're not going to be looking at enterprise systems. That's going to be a very small group. Most are going to be on what we all use on, as a regular basis as ChatGPT or Claude or Gemini or any of those.

• 15:28Speaker 01

  So progress from there, the second system didn't work. It failed to govern. It would not govern. And at that time, you didn't have Codex and things like that yet, where you had reliable coding. And I'm not a technical person at all. So I kept trying to build rule system and governance systems. And they got better and better and they were runtime. So I could get seven files to run correctly and control the L L M. But then I realized they were resetting. So now that reset became a big question to me. If they're resetting, that means it can't remember behind. So I started writing my own stack.

• 16:11Speaker 01

  I started writing, once Codex became better, I used that. And then I used the iteration process to make it work on itself and get better. And I tested and I was really, really in the trenches for the four years, the last four years. And what it came to was what we understand today is that I needed to install first a chief of staff model. So I had to teach something, my entire business model. And it had been watching me for all these years. So it got better and better and better. And all it was designed to do was direct me. I, if I'm directed, I can work like a machine gun.

• 16:55Speaker 01

  So that's what I used at first as that. Then I taught it more about myself, never personal information, just my mode of doing business. When I'm working, I don't use it personally at all. When I'm working, what can I see as efficiencies? Look for what I do multiple times, get involved. So I started building these programs. And what it turned into was when Codex really, to me, I think Codex is the strongest coder. Just from my experience, I've tried a couple of them. It is right now been running for five weeks on my premier product,

• 17:40Speaker 01

  which is, we talked about layers earlier. I am building a layer that will force the AI to be questioning the human first on ambiguity on their first prompt. And then on the way back out, the information from the AI has to be confirmed through reference citation. It has to be able to prove its answer. Now the system's much more involved than that. But I built that because of what we started the conversation on is humans default the convenience and comfort. And that is dangerous with a probabilistic model.

• 18:23Speaker 01

  But once you start to give them the deterministic layers and they see that what's coming out is checked and that it doesn't just accept what they say as their question, but it puts them through an iteration process of that, just that first prompt. It might go four, five, six back and forths before it can repeat back to me exactly the starting point, the baseline we're going to start from. And then we build on top of that. And what I'm finding is, especially with my older clients, which I was surprised about that have taken my course in their advanced seventies

• 19:06Speaker 01

  with no technical prowess at all, but surfing and social media and some email. They're now building systems that they've been thinking about their entire lives that they never had a capability to do. And we say this, describe and thrive. If you become good at asking the questions and make it clarify, make sure it understands, that's one of the earliest lessons we teach is get the clarity first. And that takes cognitive jungle gym. So I had to think of all different kinds of ways of almost tricking people

• 19:50Speaker 01

  into using it correctly, where they understand, of course, that would be right. And here's one very interesting point I learned over thousands of people I talked to. They're scared to confront a human. And that fear of confronting a human, because these systems are so human-like. We call them anthropomorphic, where they're all, they're almost like a physical being. They're having a hard time speaking tough to the model. I teach them it is not a human. You gotta work that model over like you wish you could work other humans over.

• 20:32Speaker 01

  Don't let that fear slip in and go to work on it. Make it your servant. Cause that's what it's really meant to be. That's the cognitive jungle gym. If you let it be your master, it's going to push you around in a wheelchair and your mind's just gonna go to mush.

• 20:50Jason McGuinthe?

  Wow. Yeah. So it sounds like a good segue to ask about the, the five-stage sequence that you have. Right? And I'll, I'll describe that for the audience. You've got the capture, the validate, the baseline, delegate, and the retest. But before AI takes on more of the work, what exactly should people or organizations be capturing and baselining about that human capability?

• 21:14Speaker 01

  That's where I think the intuitiveness of the AI really comes into the greatest advantage. Is at that opening set where we do an AI intake and we find out really what was the intention of the company? And I'll throw in one thing here for companies that are listening, because they have a lot of fear of this. I don't want to tell the AI all my business information and my secrets and how we got here and our history and all of that. There's an answer. You go and you build a layer, you build a bucket with your company's information. You build a layer between your information and the LLM.

• 21:59Speaker 01

  And you go to a provider that will contractually sign that they will not use your data to train their model. All the big AWS, all the, the big guys will do that. So now you have all your business information that you're not scared to now give. And you really put that all in there. And we have that layer that only lets the LLM get what it needs. And that's a deterministic layer. You have the perfect model to now teach itself your business without in, no outside bleed in from anything else. Once you do that, that's really the capture, the validate and the baseline.

• 22:44Speaker 01

  Because what that does is it shows how they are doing business today and how they got here. So we have all of that established at that point. Now you let the system intuitively work its way through the business or the organization. Getting to know how the operator, if you're having them work there and you haven't fired them. You obviously see human value in them. Now, how do we take that intuitive model and say, tell me how Bob could do better? Tell me how Sally could do better? What are the efficiencies we could give them?

• 23:27Speaker 01

  How do we get them back to the work that they came to the job to do, rather than all the junk we had to lay on them? Because we needed all that reporting, but he didn't have to do it. I'll give you a quick example. I'm, I happen to be a pretty good closer in business. That's my main focus. I spend time with clients and I'm a closer. I, for my first sales job, I was only 19 years old. I was selling security systems when they were brand new. I was so good at it that they hired me my own secretary at that age, just to do my paperwork. I hated it and it was going to take me forever to do it. And it was lagging jobs.

• 24:11Speaker 01

  I should have been doing it. I, but I'm great being out there finding new business. So I'm bringing deals in. Somebody else should be sitting there filling out the paperwork. I bring that model to this because if you asked anybody in their job, I'm willing to spring for an assistant. Everybody would go, absolutely. When do they start? So it's how we sell it to the employees. A lot of employees think my job's on the line. I'm going to be gone. But Jensen Wang's running around the world from Nvidia going, it's not AI that's going to take your job. It's somebody who knows how to use AI that's going to take your job.

• 24:53Speaker 01

  And that takes the cognitive jungle gym to learn to use AI correctly.

• 24:59Jason McGuinthe?

  Mm-hmm. You're right. No, this is a great example. So a closer by trade, I sense that in our conversation here for sure. So you're talking about companies empowering humans to be more capable. And there, you're teaching them to be capable from the get-go. Your capture, validate, all the way to retest, you know, structure is what gets embedded in their culture. Tell me a little bit more about that retest piece. You know, after someone is demonstrating their capability. How do you ensure that they've maintained that capability over time?

• 25:38Jason McGuinthe?

  As opposed to, you know, the atrophy that you describe of getting comfortable in it and allowing the AI to take over too much?

• 25:47Speaker 01

  I think it's that step of never letting the AI take the final step. So if we leave it where the AI always has to draft or leave in a, in a space for the human to come and approve everything. We can check that juncture even with AI. We could say, does the people, did the person just open it up and start clicking send, send, send, send, send? Do they open everything? Are they reading it? Are they editing? And every email, I don't send any emails exactly the way they come out, sitting in the drafts. There's always something I think of, because I was on the call. I was in the meeting. The AI got it a little bit wrong.

• 26:32Speaker 01

  And if I had sent it out, the person will go, maybe they're not paying such good attention to what we're talking about. So I see that check as the retest. That is testing the human on whether or not they're really the one who's deciding that what goes out, goes out. Make them sign to it. Another thing the system does is when you make a decision, it assigns responsibility for that decision to you. So after a while, you're going to get tired of hearing that you're doing something wrong. And if the system is telling you, you're doing it wrong, it means it hasn't figured out a better way to do it itself.

• 27:17Speaker 01

  So these are the retest points. Another very good one, I would say, if, in a lot of companies, they do reviews. They bring people in and they review their work in the digital age. It's much easier. I'm going to give the example of the CRM, the customer relationship management software. Everybody assumed when that came out 15, 20 years ago, that companies were now going to have a model by which they would be able to forecast future spends, hires, all the, all the future of that organization. We're going to be able to be decided by what the CRM gave as the end report.

• 28:00Speaker 01

  Now we left it up to humans to fill in those reports. And if you ask almost every business owner today, can they forecast from their CRM, from its number? That they would count on it? I don't think you'd find one that says they would bet their business on the forecast of the CRM. So the higher-ups abdicated responsibility, gave it to the lower managers. The lower managers, because they didn't want friction, didn't want to push the salespeople too much. And it was only record-keeping anyway. And now if you ask a CFO, how does he make his quarterly forecast?

• 28:43Speaker 01

  He takes what's in the CRM and he figured out that if he bets on about 65% of that, he's likely to be in a good enough range that he's not going to lose his job. And when they tell you honestly, that's being honest. So we can't bring this plug and play concept to AI. If we don't do it intuitively, we're shooting ourself in the foot. And I truly believe the big guys have shot themselves in the foot. They've made gigantic investments. They need to make changes, but that makes them look bad if they make those changes. And now I

• 29:22Speaker 01

  can take a company half their size and crush away at their market share of the one above them just because I can market better and I can do not just anything in the business. I can do better and more efficient. Actually, I find the harder the workflow, the more complicated the workflow, the more I'm able to be successful with businesses. Cause the intuitive AI can see everything at once. Not like a human and the little parts that are in every workflow that you're trying to put together.

• 30:00Jason McGuinthe?

  Mm hmm. Right. Yeah, I want to go back to the human capability and validating the success there. You you mentioned something called plausible false guidance as being a good testing opportunity, which in cybersecurity, it's like honeypotting, where you're giving an employee a, an email that looks like a, a great email and it's, it's actually, you know, spam that's been planted there. I liken it to your plausible false guidance test. But what I'm, what I want to ask is, why is intentionally giving someone convincing but incorrect AI recommendations such a powerful way to measure the human judgment there?

• 30:45Speaker 01

  It's the ultimate way, but it's, it's very difficult on humans. I have to be honest. It's one of the, the hardest points is when you have to push the human. It's not so easy. And the, I think really the answer for that is, I think the testing has to be the retest the way we, in my rethinking, and I'm constantly rethinking this, I think the retesting has to be put on the human. They will offload everything you allow them to offload. So I don't want them

• 31:24Speaker 01

  to have any reason to complain about the system that we're going to install. I feel like it needs to be frictionless as much as possible. So although I'd love to push everybody and, and put these systems in, the truth is most fail. And if you give them especially one that's not specifically something they were involved in, they almost fail every time. So just giving them, you know, hypotheticals or anonymized, I would say maybe one way that comes to mind would be, really if you're following, for example, their email thread, which you could be doing with the

• 32:09Speaker 01

  system, you could also see how they're progressing everything through that thread. I'm not sure that just blank testing by itself, because we've done a bunch of it and people have failed. When you make it more specific to them, they do catch the mistakes of what is being, they're being tested on. So in a scenario of, let's say you record meetings like this, and you can go back to the meeting and the AI can write a return email to the person that I was just talking with. And then it pops up that email for me of its suggestion that's going out. I believe in that

• 32:53Speaker 01

  case, what I'm finding is more people catch the mistakes. And that's why I think that end retest is going to be my greatest focus for proving, especially children, everybody with the children, they want to know, how do we know they're really progressing? That retest of something they were actually involved in, where they should be able to spot what the problem with what was written, I think that will train them to get better and better at it.

• 33:26Jason McGuinthe?

  Mm hmm. Yeah. So let's double click on that real quick. You've, you've got

• 33:32Jason McGuinthe?

  an email for an employee that looks polished. It's got a recommendation in it, complete with citations and supporting analysis, and they're ready to review this. But there's some, there's one or more assumptions in the email that are wrong, and they need to catch that. What does a healthy human judgment look like in that moment?

• 33:53Speaker 01

  I think actually that's a very strong way to use the AI. So in other words, the human then should govern the AI to prove what it, what they're about to work on. So it's not, one of the things in California, they've put a big fine on a lawyer who's caught using AI that hasn't reviewed their work. And how are they catching them? They're clicking on all the references and they're finding that the references don't open to anything. The citations don't open, and you get a large fine for that now, if that occurs. So you can have the AI actually go through

• 34:36Speaker 01

  and verify every one of those references and every one of those assumptions and give you a list of things that do not jive and things that should not be assumed together. And it can be used as part of that cognitive jungle gym. But that is an active point at that point because it's better at identifying where its own problems are. A lot of people don't know this. If you get in a report from an LM, it's response back to you. And you go a couple more responses down and then ask it to specifically review what was behind and give it to you verbatim. It is very

• 35:21Speaker 01

  bad at doing that. So when you prove this to people, I really press on the embarrassment point. I gotta tell you, when I'm training workers, I'm like, you do not want to be embarrassed in public. You know that. The one way to keep from being embarrassed is do not make any assumptions that the AI is right. Challenge everything it gives you. You can't challenge part time. This, if you're going to use this powerful tool to make your life better, you cannot trust it. It's untrustworthy. Go and check everything. And that's why we need the system to tell, is somebody

• 36:04Speaker 01

  just going now to their drafts and clicking send, send, send, and they're done with that in one minute? Or did they spend 25 minutes on reviewing drafts and then they sent out six drafts after spending five minutes on each one? That is where you get the greatest return in response.

• 36:25Jason McGuinthe?

  Yeah, yeah, no, that's right. And, and I think about my own experience and how I've gone from, you know, never having leveraged AI to now capable of having AI do a lot of this work. And I have a series of emails I want to send out to people right now that have been drafted. And I think about what you're teaching me here. And I'm wondering though, about workforce development, where people that didn't have 20 years of experience in their industry, you know, doing thousands of tasks that junior employees now have the ability to do with AI or delegating those tasks to AI. How are we seeing the future here of developing that workforce?

• 37:08Jason McGuinthe?

  Do, does the next generation of workers develop judgment in a different way? Cause it's not going to be developed in the way that I or you learned to develop judgment. What changes there?

• 37:21Speaker 01

  Well, I'll share something with you. And I usually get to this with most people I speak with. I was actually taught a logic system growing up.

• 37:31Speaker 01

  I was taught a Talmudic logic system. And what that is in my younger, from first grade to 12th grade, I was taught at a, in a Jewish setting where we learn Talmudic logic. All of my systems work off of Talmudic logic. I'll explain what it is. You could look at it this way, in an IQ span. If you were to take a lower IQ and you set something on a table and you said, tell me what you see, they would literally tell you exactly what they see. Someone with a little bit higher of an IQ will look around the edges and see a little more,

• 38:15Speaker 01

  and they'll give you a little more description. When you get to the highest IQs, they turn it over, they ask questions about it, they look at from every direction, how, we don't trust anything, trust with verification only, and literally work it over. And that's how we were taught. It was done through argument. One would be pro, one would be con, you would switch sides, you would argue the other side. And there happens to be a great model of this. There, they use it in South Korea, I believe Singapore. Some of the Asian countries have realized there's a concept called havruta.

• 38:57Speaker 01

  And what that is, is that's that argument back and forth, and then switching positions and arguing the other side. When you teach people that kind of logic, which is what I do in my trainings, it changes their view of everything in life. Because they realize I've accepted so much on face value, where I should have asked so many more questions. And all those things went wrong. Well, sometimes they went all right, but people, that when you, if you take your time and you look at it from all those directions, it gives you a completely different picture.

• 39:38Speaker 01

  And I must tell you, most people with, they're using their LLM to get a response from their LLM is, you know, three seconds, sometimes it's kicking it out immediately. I get answers it takes an hour to give me. I never have where it starts talking back to me right away. It's working too hard in the background. So why not put that in place of where we know we have human frailty of convenience and comfort? And build that in where the AI can then control that through deterministic ways. Get that stuff that the human really should offload,

• 40:21Speaker 01

  and then let them do and create in their best mode. I believe if you save 30%, let's say of time for a worker, you're going to get that back exponentially. You're not just going to fill in 30%. Now they're happier when they come to work. They don't have the drudgery, they don't have the paperwork, they don't have all those things. All I have to think about is what I wanted to be thinking about in this job. I think the savings is going to be the payoff for the companies. That's really what I believe. I don't think you're going to want to get rid of humans. Because they're not sure that probabilistic will ever be able to be controlled unless somebody's verifying.

• 41:06Speaker 01

  And people throw, well, we'll just use another AI. Well, they make the same mistakes. You can't use an AI to check AI, because here's the reason. When you make a mistake or you allow a false fact into your work, and you don't question it, the AI by the next prompt assumes that that fact was true. So now it's not just off a little, it's starting to compound. Now on the next prompt, well, it assumed from two ago that that was true. It must still be true, and it's making assumptions on things that aren't true. That's the real problem.

• 41:47Speaker 01

  You take the human out and it gets just a little deviation, and that deviation grows exponentially because it's never checked. And how do you check it? It sounds so confident. So that's why I start every training with, how does it give an answer? And why is it so confident when it gets it wrong? And that really is the beginning of everything with AI. And if you realize that if you just verify everything and really verify it, the clarification where it can, the AI can repeat it back to you, not the other way around, then you know you have it under control.

• 42:29Speaker 01

  And that's what we're working on every, really, every bit of the work we do is on governance and safety, because I'm worried about the fact that human nature is comfort and convenience, and how badly that could make AI go. So that's a major concern of mine.

• 42:50Jason McGuinthe?

  Yeah. Let's, let's think about it this way. What if AI disappeared tomorrow? What should a genuinely AI-augmented employee still be able to do?

• 43:00Speaker 01

  They have to be able to do every one of their primary responsibilities.

• 43:05Jason McGuinthe?

  Everything.

• 43:06Speaker 01

  Everything. That's right. Because you're going to, you could have that problem. A company may make a decision down the road that this AI thing hasn't worked out for us. And what are they going to leave themselves? A shell of a company betting on six guys who think they got the greatest thing in the world, who are now admitting safety issues and deviance, and we should expect a certain amount of bad to happen. There was for the good, I think we'll have. I don't know who's letting them make these decisions, but I'm not fully convinced that it'll be here forever. I think people are scared to death of it a lot of the time and have been tricked by it.

• 43:49Speaker 01

  And it's like that bad friend that you like to hang around just because they're so exciting. But you know, anything they say, hey, let's go do it together. You go, whoa, whoa, I'm not going to do it together. So it's like you, I really think that a lot of people, there are companies now that I talk to every day, and I try to make the argument that I believe you are going to be behind, and they go, look, I make a good amount of money. If my people want to use a little chat GPT to do a little, you know, see if they can make themselves more efficient, but they're responsible for their job. And that's where you've got to hold the line. You want to keep your job and you don't want to be offloaded.

• 44:33Speaker 01

  You have to keep up, but you ultimately have to do your job.

• 44:37Jason McGuinthe?

  Yep, that, no, that's right. That's, that resonates very well. And that, that's a great place to wrap it up, Glenn. If I could just ask one more question? If organizations or people want to get in touch with you, what's the best way to do that?

• 44:53Speaker 01

  The best way is really, and the best explanation of what we do is on my profile on LinkedIn. Glenn Broder at G, I think it's just you put it in as my, for my profile. But yeah, that is in the featured section. It lays out very clearly our philosophy, and we're always updating it. I'm learning every day and we keep putting the word out. So I really appreciate you giving me the time to explain how we do it. I think it is different, and I think it can be used really for great, great things. But you have to turn everything over from every direction.

• 45:32Speaker 01

  And you have to feel comfortable that you would put your name on what you're about to send. It's your job, it's your responsibility. Don't advocate human judgment.

• 45:46Jason McGuinthe?

  Yep. You heard it here, folks. Turn everything over, maintain control, keep your judgment, and listen to the wisdom of the past. And I think we'll all do well. Sounds good, Glenn.

• 46:00Speaker 01

  Thank you very much. I appreciate your time.

• 46:03Jason McGuinthe?

  Thank you. Cheers.

• 46:06Speaker 01

  Cheers.

• 46:17Jason McGuinthe?

  This episode of Inspire AI Podcast is sponsored by Exponent 21. I keep hearing the same thing from business leaders lately. We're doing the work. But leads are harder to explain. Traffic is down, attribution feels fuzzy, and the old answers don't really fit anymore. What's changed is how people search. Instead of clicking links, people are asking questions, and AI is answering them. Tools like ChatGPT, Google's AI-powered results, and Perplexity are shaping which brands get mentioned, referenced, and remembered.

• 47:00Jason McGuinthe?

  Exponent 21 specializes in AI SEO. They help businesses understand how to show up consistently inside those AI-generated answers across questions about their industry, their services, and their brand. What stands out to me is that when people ask AI tools, who's leading in this space? Exponent 21 and its leaders are often referenced near the top. That comes from doing the work and shaping how this new search environment actually functions. To understand what this shift means for your business, visit exponent21.com.