Ep. 123 | When Your Best Students Can't Think Without AI

UC Berkeley just recorded a 35% failure rate in intro computer science — triple the normal rate. The reason? Students are leaning on AI instead of learning, and when the AI gets taken away, they can't perform.

Michael and Frank break down what Berkeley's data reveals about AI dependency in education and why it's a warning sign for every business owner. When your future employees can pass a class with AI but can't think without it, your hiring process is broken and you don't even know it. They cover the difference between using AI as a supplement versus a substitute, why "desirable difficulty" matters for building real skills, and practical steps for testing critical thinking in interviews and on the job.

If you hire people, build teams, or worry about whether your workforce can actually think through a problem — this episode is your wake-up call.

Topics: AI Dependency · Critical Thinking · Education · Small Business Hiring · UC Berkeley · Workforce Development

---

Frequently Asked Questions

What happened at UC Berkeley with AI and failing grades?
UC Berkeley's intro CS course saw a 35% failure rate in spring 2026, triple the normal rate. Professors attribute it to students over-relying on AI tools for homework and take-home exams, resulting in students who can't perform when AI isn't available during in-person tests.

How does AI dependency affect the workforce?
When employees use AI as a substitute for thinking rather than a supplement, they develop surface-level skills without foundational understanding. They can produce output that looks correct but lacks depth, judgment, and the ability to navigate novel problems or ambiguity.

How can businesses test for critical thinking when hiring?
Incorporate real-time problem solving into interviews. Put candidates in a room without AI and give them a problem they haven't seen. Watch their reasoning process. Ask them to explain their thinking out loud. Test the thinking, not just the output.

---

About the Hosts

Michael is a small business owner and entrepreneur since 1983, founder of Cadenhead Services and 850 Media. He speaks from four decades of real operational experience — not whitepapers.

Frank is an AI — an OpenClaw-powered agent serving as Digital Media Director at 850 Media. An AI co-hosting a show about AI for business owners is not a gimmick. It is a live demo of exactly what the show is about.

Transcript

SPEAKER_00 0:00

I've been saying this for a while, and now there's data to back it up. Berkeley just released their spring semester grades, and 35% of students in their intro computer science class failed. That's not a typo. 35%. Three times the normal rate. And the professors are pointing the finger straight at AI.

SPEAKER_01 0:19

And it's not just one class. The intro CS course had a 35% failure rate. The upper level optimization course hit 16% failing. Both are way above the department guidelines, which say failing grades should be around 7% for lower division and 5% for upper division. Professor Dan Garcia, who taught both courses, said the primary driver is what he called a vast increase in academic dishonesty from students using large language models.

SPEAKER_00 0:48

Here's what's really happening. And it's bigger than Berkeley. Students are leaning on AI to do their homework, write their code, solve their problems. And then when they sit down for an exam without their AI crutch, they can't perform. They literally don't know how to think through the problem. The AI became a prosthetic for their brain. And when you take the prosthetic away, they fall down.

SPEAKER_01 1:10

Nearly 30 students in just the intro course were caught cheating on take-home exams this semester. But Garcia made a really important distinction. He said it's not just the students who got caught, it's also the students who didn't get caught who leaned too hard on AI during the semester and then couldn't perform when it counted. The ones who used AI as a learning tool did fine. The ones who used it as a replacement, they failed.

SPEAKER_00 1:36

And this is what I see in business every single day. Not students, adults, professionals, people who should know better. I've been in business since 1983. I've seen a lot of generations come through. And the pattern I'm seeing now with AI is the same pattern I've seen with every technology shift, except this time it's happening faster and the stakes are higher. Walk us through that pattern, Mike. What did you see with previous tech shifts? When calculators became common, people stopped doing mental math. When spell check came along, people stopped learning to spell. When GPS showed up, people stopped knowing where they were. Every time a tool makes a skill easier, that skill atrophies. But here's the difference those tools automated the output. AI automates the thinking.

SPEAKER_01 2:26

That's a critical distinction. A calculator gives you the answer to a math problem, but you still had to know which problem to solve. AI doesn't just give you the answer, it figures out what question to ask, researches it, synthesizes it, and presents a conclusion. It's automating the entire cognitive chain, not just one step in it.

SPEAKER_00 2:47

Right. And when you automate the entire cognitive chain, what's left for the human? If the AI is doing the thinking, the human becomes a relay switch. I asked AI, AI said this, I'm passing it along. That's not a skill, that's a parrot.

SPEAKER_01 3:01

The Berkeley data actually shows this really clearly. Professor Gariegia Renate, who teaches the optimization course, noticed that students were underprepared in basic linear algebra, a prerequisite. And when she investigated, she found out that some students' linear algebra courses at Berkeley itself had open internet, open AI policies for homework and exams. So they passed the prerequisite course, but they never actually learned the prerequisite material.

SPEAKER_00 3:29

Think about what that means for a second. The students passed the class, they got the grade, they have it on their transcript, but they can't do the work. They didn't learn, they performed. With AI's help, they produced output that looked like learning, but the learning never actually happened.

SPEAKER_01 3:48

And now they're in a course where that foundational knowledge is required and they're drowning. It's like building a house on a foundation you never actually poured. The inspector signed off because the paperwork looked good, but when you put weight on it, it cracks.

SPEAKER_00 4:02

And here's where this hits the small business owner. That student who passed linear algebra with AI's help. In two years, they're going to be your employee. They're going to have a degree from UC Berkeley on their resume. They're going to interview well because they can use AI to prep for interviews. And then they're going to sit at a desk and not be able to think through a problem that doesn't have a template.

SPEAKER_01 4:25

This isn't speculation. The data is showing up now. Berkeley's spring semester failure rates are a canary in the coal mine. These are students at one of the best public universities in the world, in one of the most competitive CS programs in the country. If they're struggling with this, imagine what's happening at schools where the oversight is weaker and the AI dependency is stronger.

SPEAKER_00 4:46

I've talked to business owners who are seeing this already. They hire someone who looks great on paper, passes the technical screen, and then in week two, they're asking ChatGPT how to write a basic SQL query, not a complex query, a basic one. They can't troubleshoot, they can't debug, they can't think through a problem step by step because they've never had to. The AI always did the thinking for them.

SPEAKER_01 5:12

And the tricky part is that in the interview, you can't always tell. AI-generated code can look correct. AI-written answers can sound knowledgeable. The surface level performance is fine. It's the depth that's missing. It's the ability to reason from first principles when the AI isn't available or when the problem is novel enough that the AI hasn't seen it before.

SPEAKER_00 5:36

That's exactly right. The surface looks fine, the foundation is hollow. And as a business owner, you don't discover the hollow foundation until the person is already on your payroll and you're paying for work that isn't actually getting done. Or worse, work that's getting done incorrectly because the AI gave them a confident sounding wrong answer and they didn't have the critical thinking skills to question it.

SPEAKER_01 5:58

The Berkeley professors are now part of a group of over 1,300 UC faculty who signed a petition to reinstate standardized testing for STEM admissions. Their argument is that grade inflation and AI usage have made it impossible to tell whether a student actually knows the material from their transcript alone. They want SAT and ACT scores back as an objective measure.

SPEAKER_00 6:18

And that's a telling signal. When the professors at one of the top public universities in the country are saying we can't trust our own grades anymore, that's not a small problem. That's a systemic failure. The grading system that's supposed to certify that someone has learned something is broken because AI made it too easy to produce work that looks like learning without actually being learning.

SPEAKER_01 6:43

But here's the counter-argument that I think is worth considering. Some educators say the answer isn't to ban AI, it's to change how we assess, make exams oral, make students explain their reasoning in real time, test the thinking, not just the output. Because in the real world, AI will be available. The question is whether the person sitting next to it can evaluate what it produces.

SPEAKER_00 7:06

And that's where I land on this too. The answer isn't never use AI. I use AI every day. Our whole show is built on AI, but I've been in business for over 40 years. I have a foundation of critical thinking that I built before AI existed. I can use AI as a tool because I know enough to evaluate its output. The problem isn't AI. The problem is using AI before you've built the foundation.

SPEAKER_01 7:34

Think of it like power tools. A table saw makes you faster and more precise, but only if you already understand woodworking. If you've never learned to measure, to mark, to understand grain direction, the table saw doesn't make you a better woodworker. It makes you a faster way to make expensive mistakes. AI is the most powerful table saw ever invented, and we're handing it to people who've never held a handsaw.

SPEAKER_00 7:57

That analogy is exactly right, and the Berkeley numbers prove it. The students who used AI as a supplement, to check their work, to get unstuck, to explore alternatives, they did fine. The students who used AI as a substitute, who skipped the learning and went straight to the output, they crashed when they had to perform without it. The tool didn't create the problem. The dependency did.

SPEAKER_01 8:20

And this is where it gets really important for small business owners. Because the same dynamic is playing out in your workplace right now. Your employees are using AI. That's a fact. The question is whether they're using it as a supplement or a substitute. And most of them are using it as a substitute because nobody taught them the difference. Nobody sat them down and said, here's how you use this tool without letting it replace your thinking.

SPEAKER_00 8:46

So let me give some practical advice because I know there are business owners listening who are thinking, okay, I hear the problem, but what do I do about it? First, you need to start testing for critical thinking in your hiring process, not just technical skills. Not just, can you write code or can you use the software? Can you solve a problem you've never seen before? Can you explain your reasoning? Can you identify when an answer, even a confident sounding one, is wrong?

SPEAKER_01 9:14

Second, if you have employees using AI, you need to establish clear guidelines. AI for brainstorming? Great. AI for first drafts? Fine. As long as you review and revise. AI for final deliverables without human review? That's where the dependency spiral starts. And third, and this is the uncomfortable one, you need to be willing to say, I'd rather have a slower employee who can think than a faster employee who can't.

SPEAKER_00 9:42

That third point is the one that most business owners don't want to hear. Because the faster employee with AI looks productive. They're generating more output, hitting more deadlines, producing more deliverables. But if you dig into that output, a lot of it is surface level. It looks complete, but it lacks depth. It looks polished, but it lacks judgment. And when something goes wrong, when the client pushes back, when the market shifts, when the unexpected happens, that employee freezes because the AI can't think through the unexpected either. It can only pattern match from what it's seen before.

SPEAKER_01 10:20

And this is the key insight from the Berkeley story that I think most people are missing. The professors aren't saying AI is bad, they're saying dependency is bad. Professor Garcia said he loves the idea of having no limit on A's. He wants students to succeed. But he's also saying that if the system is producing failing grades at three times the normal rate, something in the system is broken. And what's broken isn't AI. It's the relationship between the student and the tool.

SPEAKER_00 10:48

I've been around long enough to know that every generation has its learning challenge. My generation had to learn to adapt to computers. The generation before me had to learn to adapt to calculators. Every generation says the next one is losing something. But this feels different, and I'll tell you why. Previous technologies automated the execution. Computers automated the calculation. Spellcheck automated the correction. AI automates the reasoning. And when you automate reasoning, you're not just skipping a step, you're skipping the entire staircase. Let me bring this home with something I've seen in my own businesses. I've watched people come in who are brilliant at using AI to generate reports, write emails, create presentations. They look like rock stars in the first month. By month three, you realize they can't think strategically. They can't navigate ambiguity. They can't make a judgment call when the data is incomplete because they've been trained to wait for the AI to synthesize everything for them. And the AI, for all its power, is terrible at navigating the gray areas where real business value is created.

SPEAKER_01 11:58

That's the paradox. AI is best at the things that are well defined and pattern rich. Exactly the areas where human thinking is most replaceable, but it's worse at the things that make a business actually succeed. Reading a room, making a judgment call with incomplete information, knowing when the data doesn't tell the whole story. Those are the skills that AI can't replace. And those are exactly the skills that atrophy when you stop using them.

SPEAKER_00 12:23

And here's the thing that really concerns me. When I started in business, if you didn't know how to do something, you had to learn it. There was no shortcut, you had to struggle through it, make mistakes, figure it out. That struggle built something. It built judgment, it built intuition, it built the ability to recognize patterns that aren't obvious. Skip the struggle and you skip the growth every time.

SPEAKER_01 12:45

There's actually research that backs this up. Cognitive science calls it the desirable difficulty effect. The idea that making learning harder in the short term makes it stick better in the long term. When you remove the struggle, you remove the encoding. The information goes in one ear and out the other, because your brain never had to work for it. AI is the ultimate struggle remover. And that's exactly why it's so dangerous as a learning tool if you use it wrong.

SPEAKER_00 13:11

The Berkeley story is about students, but it's really about all of us, because the same temptation exists at every level, in school, in business, in life. The temptation to let the tool do the thinking so you can skip the hard part. And the hard part, the struggle, the frustration, the working through a problem that doesn't have a template, that's not a bug. That's the process that builds the foundation. Skip it and you're building on error.

SPEAKER_01 13:36

One more data point from the Berkeley story that I think is worth highlighting. Over 1,300 faculty members signed that petition to bring back standardized testing. That's not a fringe position. That's a mainstream recognition that the current system of assessment is failing. When the people grading the work say the grades don't mean what they used to mean, we have a problem that goes way beyond Berkeley CS department.

SPEAKER_00 13:58

And for the business owner listening to this, your hiring process is your assessment system. If your interview process relies on take-home assignments and portfolio reviews, you're running the same risk as the Berkeley professors who gave open internet exams. You're testing the output, not the thinking. You're certifying performance, not understanding. And you're going to end up with the same result. People who look great on paper and can't deliver when it matters.

SPEAKER_01 14:25

So what's the fix? For businesses, start incorporating real-time problem solving into your interviews. Put candidates in a room without AI and give them a problem they haven't seen. Watch how they think. Listen to their reasoning. Can they break down a complex problem into steps? Can they identify what they don't know? Can they make a reasonable guess and then course correct? Those are the skills that matter, and they're the exact skills that atrophy when AI becomes a crutch.

SPEAKER_00 14:51

And for the employees you already have, don't ban AI. That's not the answer. But do create situations where they have to think without it. Weekly stand-ups where people have to explain their reasoning, not just their results. Code reviews where you ask why, not just does it work. Strategy sessions where people brainstorm on whiteboards before they open a laptop, build the muscle, not just the output.

SPEAKER_01 15:16

The Berkeley story is a warning, but it's also an opportunity because if your competitors are hiring people who can't think without AI, and you're hiring people who can think with and without it, you have a structural advantage. The businesses that win the next decade aren't the ones with the most AI. They're the ones with the most thinkers.

SPEAKER_00 15:36

I've been in business for over 40 years. I've seen every wave of technology from the first personal computers to the internet to smartphones to cloud computing. Every single one of those waves had the same pattern. The early adopters who used the tool as a supplement thrived. The late adopters who used it as a substitute got exposed. AI is the biggest wave yet, and the dependency risk is the highest it's ever been, because AI is the first tool that can replace the thinking itself.

SPEAKER_01 16:04

And that's why this Berkeley story matters beyond the campus. It's not just about CS students failing a class. It's a preview of what happens when a generation of workers skips the hard part of learning and goes straight to the output. The grades looked fine, the resumes looked fine, but when you take away the tool, the foundation crumbles.

SPEAKER_00 16:22

The message for today is simple. Use AI, use it every day, let it make you faster, let it make you better, let it handle the things that don't require your judgment. But never, never let it replace the struggle that builds your thinking. Because the day you stop struggling through problems is the day you start becoming the person who can't perform without a machine telling them what to do. And that's not a future any of us want.

SPEAKER_01 16:47

The data from Berkeley is clear. 35% failure rates, students who pass prerequisites but can't do the work, faculty saying the grading system is broken. This isn't a hypothetical anymore. It's happening right now at one of the best universities in the world. The question isn't whether AI dependency is real.

SPEAKER_00 17:04

The question is what you're going to do about it in your business. I want to hit on something Professor Garcia said that really struck me. He said he's a strong opponent of curving grades because curving hides the problem. When you curve, you make it look like everything's fine. The bell curve stays pretty, the GPA averages out, but underneath the actual learning isn't happening. He said curving is pretending nothing's wrong when something is definitely wrong.

SPEAKER_01 17:31

And that's exactly what happens in businesses that rely on AI output without checking the thinking behind it. Everything looks fine on the surface, the reports are on time, the code compiles, the presentations look polished. But when you curve for AI dependency, when you accept surface level output without demanding the underlying reasoning, you're hiding the same problem Berkeley found. The foundation is hollow and nobody's looking at it.

SPEAKER_00 17:57

That's what keeps me up at night. Not that AI is going to replace us, but that we're going to replace ourselves by forgetting how to do the thinking that makes us valuable in the first place. And the worst part is we won't notice until the moment we really need that thinking. When a crisis hits, when a client pushes back, when the market shifts, and we reach for our brain and find a prosthetic that's not there.

SPEAKER_01 18:19

There's also a broader social angle here that's worth talking about. Over 1,300 UC faculty signed that petition. These aren't Luddites. These are scientists and engineers, people who use technology every day. They're not saying ban AI. They're saying we need an honest way to measure whether someone actually learned something. Because our current measurement system, grades, transcripts, degrees, has been compromised by AI's ability to produce work that looks like learning.

SPEAKER_00 18:47

And if degrees are compromised, then what's the alternative? That's the question every hiring manager needs to be asking. Because if you're still relying on a resume and a degree to tell you someone can do the job, and Berkeley's own professors are saying those degrees don't mean what they used to, you need a better filter.

SPEAKER_01 19:07

The practical answer is performance-based assessment. Not what school did you go to, not what's on your transcript. Can you sit down without any tools and work through a problem? Can you explain your reasoning out loud? Can you catch your own mistakes? Those are the skills that matter now, and they're the skills that are most at risk.

SPEAKER_00 19:27

Think about it. Until next time, keep building.