AI’s giant pool of hype

Summary

Podcast Summary: "AI’s Giant Pool of Hype"

Podcast: Click Here (Recorded Future News)
Episode Date: October 3, 2025
Hosts: Dina Temple-Raston
Guest: Gary Marcus (Emeritus Professor of Psychology & Neuroscience, NYU)

---

Overview

In this episode, host Dina Temple-Raston dives deep into the realities—versus the soaring expectations—of autonomous vehicles and artificial intelligence, with candid expert commentary from Gary Marcus. The conversation explores the technological, ethical, and regulatory challenges facing AI-powered driverless cars and the wider AI industry, highlighting a persistent gap between hype and actual progress. Marcus draws on years of experience to scrutinize bold claims and caution against overconfidence, warning of public backlash and calling for more prudent oversight.

---

Key Discussion Points & Insights

The Realities of Autonomous Vehicles

• Driverless cars are everywhere... and so is the hype
  • Notable observation: White, driverless SUVs silently navigating through traffic in major cities.
  • Gary Marcus: "Autonomous vehicles have arrived in Atlanta. And while you won't see the driver, Denver is the next frontier for Waymo." (00:38)
• Promise vs. Reality
  • Marcus likens the current state of AI to an era filled with "philosopher's stone" ambitions—big dreams with little solid gold delivered.
    • "I just imagine the age of alchemy being like, you know, I'm gonna turn lead into gold... It didn't work this year, but maybe it'll work next year." – Gary Marcus (01:17)
  • Despite public excitement, foundational problems persist, including startling edge-case failures and unresolved safety questions.
  • Quote: "The hype filled headlines are so dazzling, we're a little blinkered about how driverless cars are really doing." – Dina Temple-Raston (01:30)
• Current Shortcomings
  • Marcus highlights a recent recall of 1,200 Waymo vehicles due to software issues that led to avoidable crashes (01:39).
  • Real-world accidents, like a fatal rear-ending involving a Waymo vehicle in Tempe, highlight persisting risks and the system’s limitations (04:43–05:31).

The Edge Cases and Limitations of AI Driving

• Why “Edge Cases” Matter
  • AI systems struggle when faced with scenarios they weren’t trained for.
  • Example: A Tesla on "summon mode" crashed into a $3.5 million jet at an airfield simply because the system had never seen a jet on a parking lot before (06:16–06:31).
    • Gary Marcus: "It ran straight into a $3.5 million jet... These systems are basically a kind of glorified regurgitation machine..." (06:31, 06:49)
  • Current AI can handle routine, mapped environments, but would fail in chaotic or unpredictable conditions—e.g., the streets of Mumbai.
    • Gary Marcus: "If you said, okay, now we're going to take you over to Mumbai, we're going to stick you in a waymo, I'd be like, no way." (07:17)

Misleading Terminology and Customer Expectations

• "Autopilot" is Misleading
  • Terms like "Autopilot" create a dangerous illusion of safety and autonomy, encouraging driver disengagement.
    • Gary Marcus: "I hate the fact that they call it autopilot, right? Autopilot suggests that you don't need to pay attention." (07:39)
  • In reality, attentive human supervision is still essential (08:09).
  • Problem: Humans tend to tune out, placing too much trust in imperfect machines (08:09–08:35).

Ethical Dilemmas: Overhyped and Distraction?

• Trolley Problems vs. Real Challenges
  • Philosophical debates (e.g., who should the car save?) are in some ways a distraction from the urgent challenge of just making the cars safe.
    • Gary Marcus: "...in some ways we got distracted from the reality of how hard it is to just make these things go safely in general, even when there isn't a moral dilemma." (09:10)

The Regulatory Gap

• Blind Spots in Oversight

  • No strong regulatory framework exists; some government mandates (even such as windshield wipers on driverless cars) are being reconsidered as unnecessary (10:37).
  • Pressure from investors and innovators is prompting agencies to relax disclosure and safety requirements instead of tightening them.
  • Warning on Hype: "We're just early in the process, and we have this illusion that we're, like, right there and we're not." – Gary Marcus (10:21)

• Silicon Valley’s “Don’t Regulate Us” Stance

  • Companies argue that regulation will hinder innovation, but Marcus points to history—such as train travel safety regulations—showing that oversight is often needed for public trust.
    • Gary Marcus: "...people didn't want to get on the trains until there was some regulation and people should be actually pretty leery of driverless cars." (11:36)

The Hype Cycle in Both Cars and Language Models

• Ride-hailing and robotaxis (Uber/Lucid partnership)
  • The market rushes ahead despite unresolved safety and technological issues (12:00).
• Large Language Models (LLMs) and AI Industry Parallels
  • Similar hype and shortfall issues plague AI language models, as seen with the disappointing reception of GPT5.
    • Gary Marcus: "I kept telling people for years, GPT5 is not going to be magic... And you know, now GPT5 came out and people are starting to realize that I had a point..." (12:41)
• Media Incentives
  • Stories about challenges and slow progress do not get attention; the public wants tales of instant transformation.
    • Gary Marcus: "...nobody wants to read that. They want to read Tomorrow the world will change and your children won't need to learn to draw. Like, that's a much more exciting story." (13:26)

Looming Risks and Public Backlash

• Prediction for the Near Future
  • Marcus foresees significant AI-related controversy (driverless accidents, LLM-related problems, cybercrime) playing a major role in the 2028 U.S. election.
    • Gary Marcus: "I'll go on public record as saying some aspect of AI... will actually be a fairly big part of the 2028 election... Nobody should be surprised if something pretty bad happens." (13:56)

---

Notable Quotes & Memorable Moments

• “There's what we call geofencing, and so they can't take certain routes and they can't take certain roads and so forth.” – Gary Marcus (04:36)
• “We're still struggling with the basics... just the basic ethical thing of don't run into things, don't run into people, turns out to be harder to get right in some of these systems than you would have thought.” – Gary Marcus (02:12)
• “If you said, okay, now we're going to take you over to Mumbai, we're going to stick you in a waymo, I'd be like, no way...” – Gary Marcus (07:17)
• “In reality, those moral dilemmas for now are the icing on the cake.” – Gary Marcus (02:12)
• “The message from Silicon Valley, which I think is very short sighted, is don't regulate us at all or we won't be able to make progress.” – Gary Marcus (11:36)
• “I kept telling people for years, GPT5 is not going to be magic... now GPT5 came out and people are starting to realize that I had a point...” – Gary Marcus (12:41)
• “I'll go on public record as saying some aspect of AI... will actually be a fairly big part of the 2028 election... Nobody should be surprised if something pretty bad happens.” – Gary Marcus (13:56)

---

Timestamps for Key Segments

| Time | Segment/Highlight | |----------|---------------------------------------------------------------------| | 00:38 | Autonomous vehicles on U.S. streets and public perception | | 01:17 | Alchemy analogy: expectation vs. delivery in AI | | 01:39 | Waymo recalls and persistent software flaws | | 02:12 | Real ethical challenges: the basics are still hard | | 04:05 | Gary’s personal experience riding in Waymo | | 06:16 | The infamous Tesla-airplane crash: “edge case” example | | 07:39 | The dangers of the term “autopilot” | | 09:10 | Why philosophical dilemmas don't match current technical struggles | | 10:21 | Are we anywhere close to true autonomy? | | 11:36 | Regulatory vacuums and historical precedents | | 12:41 | GPT-5 letdown and the hype cycle in AI | | 13:56 | AI risks, regulation, and the 2028 election |

---

Tone and Language

The episode maintains an accessible, conversational tone. Dina Temple-Raston grounds lofty technological subjects in everyday contexts, while Gary Marcus provides sharp, occasionally skeptical commentary rich with analogies and well-chosen examples. The overall message is cautionary yet engaging: amid the "giant pool" of AI hype, the gap between headlines and reality is not only wide, but potentially hazardous.

---

Conclusion

"AI’s Giant Pool of Hype" peels back the layers of optimism surrounding autonomous vehicles and AI, pushing listeners to consider the persistent real-world obstacles, the consequences of unchecked momentum, and the urgent need for thoughtful governance. Gary Marcus, leveraging years of research and critical analysis, provides timely warnings as the digital age accelerates, insisting on the value of skepticism amid technological gold rushes.

Transcript

[Gary Marcus]

Chatgpt AI Machine Satellite engine ignition.

[Dena Templewastin]

Click here and lift up. I'm Dena Templewastin and this is Click Here's Mic Drop, a deep dive into one of our favorite interviews of the week. If you've driven around LA or San Francisco recently, you may have seen a white SUV gliding through intersections, making turns and stopping polit crosswalks. And then you notice something unnerving. No one's behind the wheel.

[Gary Marcus]

Autonomous vehicles have arrived in Atlanta. And while you won't see the driver, Denver is the next frontier for Waymo.

[Dena Templewastin]

Driverless cars aren't science fiction anymore. They're rolling out in cities across the country along with waves of hype.

[Gary Marcus]

Well, it's finally happening. The robots, they're coming. Look at all these sensors and cameras. Hi, Waymo. Wow. This thing could drive by itself.

[Dena Templewastin]

Gary Marcus has been charting the promise of AI for decades now. He's an emeritus professor of psychology and neuroscience at nyu, and he says there's a real philosopher's stone feel to all of it.

[Gary Marcus]

I just imagine the age of alchemy being like, you know, I'm gonna turn lead into gold. This is gonna be amazing. I mean, think how rich you would be if you could turn lead into gold. And it didn't work this year, but maybe it'll work next year.

[Dena Templewastin]

And that's the thing Gary says. The hype filled headlines are so dazzling, we're a little blinkered about how driverless cars are really doing.

[Gary Marcus]

Waymo, the self driving car service is recalling over 1200 vehicles due to a software flaw that caused some of them to crash into chains, gates and other roadway barriers.

[Dena Templewastin]

Earlier this week, we spoke with novelist Bruce Hulsinger about his new book, Culpability. It's a story about an autonomous minivan colliding with tragedy and the family inside left wrestling with responsibility. But in the real world, Gary Marcus makes a compelling case that we're not even at the moral dilemma stage yet. We're still struggling with the basics.

[Gary Marcus]

The reality is our ability to make systems perceive the world well and to reason over them is limited. And so in a way, those moral dilemmas for now are the icing on the cake. But just the basic ethical thing of don't run into things, don't run into people, turns out to be harder to get right in some of these systems than you would have thought.

[Dena Templewastin]

Stay with us.

[Podcast Narrator/Announcer]

This show is supported by Odoo. When you buy business software from lots of vendors, the costs add up and it gets complicated and confusing. Odoo solves this. It's A single company that sells a suite of enterprise apps that handles everything from accounting to inventory to sales. Odoo is all connected on a single platform in a simple and affordable way. You can save money without missing out on the features you need. Check out [e-mail] that's O D O O.com support comes from Dakota State University. What happens when you unite dynamic minds with one revolutionary vision? A rising force on the cutting edge of everything from a little university on the prairie to a cyber powerhouse of the plains? We're proving the nation's best cyber can come from where you least expect it. Now we're molding the future of cyber AI and quantum computing one graduate at a time. See what Dakota State University is building [e-mail].

[Dena Templewastin]

I'm Dena Templewost and this is click here's mic drop. Have you actually been in a Waymo or Tesla on Autopilot?

[Gary Marcus]

Definitely in UEMO's. I was in Uemo just a couple months ago to two rides in San Francisco and I sat in the front seat. Let's start with that. And I actually felt more comfortable there. I know that they've done a lot of work around San Francisco. I'm willing to do that in San Francisco. As a scientist, I think the odds are pretty good that it's going to be a safe trip.

[Dena Templewastin]

Waymo has built its reputation on caution, years of mapping carefully defined service zones, a company more interested in avoiding headlines than making them.

[Gary Marcus]

There's what we call geofencing, and so they can't take certain routes and they can't take certain roads and so forth.

[Dena Templewastin]

And when things get strange or confusing, the cars can call in remote fleet response teams to guide them through it. But even with that meticulous approach, last month Waymo had a brush with tragedy in Tempe, Arizona. One of its SUVs, empty at the time, was slowing to yield at a crosswalk, and a motorcycle rear ended it. Seconds later, another car struck the motorcyclist and then fled the scene. The rider was killed, and because Waymos are constantly filming their surroundings, the company was able to hand over the license plate number of the driver who left the scene. And at least for now, regulators aren't blaming Waymo. But Gary Marcus says that won't last.

[Gary Marcus]

I always think we're like one really bad accident away from the public getting really upset about this, though.

[Dena Templewastin]

Originally he thought maybe these cars would be made more safe than the ones driven by humans. I saw something you wrote back in 2012 for the New Yorker, where you talked about someday humans might not be allowed to drive because driverless cars would be so safe. Do you still stand by that prediction, or has reality kind of shifted your view?

[Gary Marcus]

It will certainly happen. The main problem, which I've been writing about since 2016, is that these systems are not very good in unusual circumstances. What we call outliers or edge cases.

[Dena Templewastin]

Edge cases like what happened at this airfield.

[Gary Marcus]

My favorite example of this is a Tesla was at an airplane trade show. Someone summoned it to just drive across the parking lot, and it ran straight into a $3.5 million jet. You can watch it on YouTube and you can.

[Dena Templewastin]

As slapstick music blares, a Tesla drives nose first into a jet. T bones it. These systems train on neat, predictable worlds. Driveways, suburban streets. So when they encounter something unexpected like a jet plane in their path, it.

[Gary Marcus]

Just didn't know what to do. And that reminds us that people, these systems are basically a kind of glorified regurgitation machine, kind of copying circumstances they've seen and not necessarily having a deep analysis.

[Dena Templewastin]

In other words, Gary says these cars don't really understand the road. They're just replaying patterns they've seen before. And driving in San Francisco, as hectic as it may get, it's nothing like driving in Mumbai.

[Gary Marcus]

If you said, okay, now we're going to take you over to Mumbai, we're going to stick you in a waymo, I'd be like, no way. Because they have not done the same work there. The data or the traffic patterns are different. They're probably not even going to understand what a rickshaw is like. It's going to be a big mess.

[Dena Templewastin]

The problem isn't just coding limitations. It's also in misleading vocabulary.

[Gary Marcus]

I hate the fact that they call it autopilot, right? Autopilot suggests that you don't need to pay attention. The lay sense of autopilot sure sounds like, hey, it's going to take me from point A to point B and I can check my email or play video games or just fall asleep.

[Dena Templewastin]

But that's just not the case. Not yet, anyway. Right now, driverless cars still need passengers to be the equivalent of a driver's ed teacher, alert and ready to take over the controls in case of emergency.

[Gary Marcus]

The fact is, you have a car that is making decisions based on software at the current level of technology. You have a human driver who's obligated to pay attention to what the car is doing. You have the reality that. That human beings tend to Tune out after a while. And so they will tend, unfortunately, to give too much trust in the machines.

[Dena Templewastin]

When Gary Marcus first started writing about driverless cars, he didn't dwell on the coding glitches or fender benders. He was thinking about morality, the kind of thought experiment philosophers love. The trolley problem.

[Gary Marcus]

It laid out a dilemma like this, where there's a school bus going out of control and you're in the driverless car. Should the driverless car save you? Or the school bus full of children?

[Dena Templewastin]

It's the stuff of ethics seminars and late night dorm debates. But Gary Marcus says it turned out to be a kind of distraction.

[Gary Marcus]

And I think in some ways we got distracted from the reality of how hard it is to just make these things go safely in general, even when there isn't a moral dilemma.

[Dena Templewastin]

When we come back, lawmakers have been largely silent and what that vacuum means for the road ahead. Stay with us.

[Gary Marcus]

At Radiolab, we love nothing more than nerding out about science, neuroscience, chemistry.

[Podcast Narrator/Announcer]

But we do also like to get into other kinds of stories. Stories about policing or politics, country music, hockey, sex of bugs.

[Gary Marcus]

Regardless of whether we're looking at science or not science, we bring a rigorous curiosity to get you the answers and.

[Podcast Narrator/Announcer]

Hopefully make you see the world anew.

[Gary Marcus]

Radiolab Adventures on the Edge of what we think we know Wherever you get.

[Podcast Narrator/Announcer]

Your podcasts.

[Dena Templewastin]

Getting into the backseat of a Waymo might feel like something out of the Jetsons. But Gary wonders whether they should even be on the road.

[Gary Marcus]

We're just early in the process, and we have this illusion that we're, like, right there and we're not. And that's probably because there's some ideas about how to build this stuff that we are lacking. Just like the alchemists were lacking the understanding of the periodic table.

[Dena Templewastin]

And now the excitement from innovators and the pressure from investors is starting to cloud rational risk assessment. The government's own Automated vehicles framework is paring back what companies have to disclose. And at the Department of Transportation, officials are even proposing to loosen mandates like the rule requiring windshield wipers on driverless cars, because, they say, if no one's driving, why should anyone need to look out? But that logic isn't just confined to cars. It's the story of AI writ large. A recent MIT study found that 95% of AI pilots have failed, and when that number surfaced, investors got skittish. All of which makes the stakes that much higher. Once the literal rubber meets the road, where experts warn we're scaling up far faster than we're making it safe.

[Gary Marcus]

The message from Silicon Valley, which I think is very short sighted, is don't regulate us at all or we won't be able to make progress. And the reality is you have historical examples like trains, where people didn't want to get on the trains until there was some regulation and people should be actually pretty leery of driverless cars.

[Dena Templewastin]

None of this has stopped companies from trying to cash in on the hype. Uber just announced a new Robotaxi partnership with Lucid, set to launch next year.

[Gary Marcus]

I think that the last decade has been so. It's almost been drowning in hype and we've mostly been talking about driverless cars. These days I mostly think about large language models and it's kind of the same story there where there's just constant hype about these things. And the hype isn't really true, though.

[Dena Templewastin]

Every now and then the hype and the reality are so mismatched it's impossible to ignore. Take the recent release of GPT5.

[Cybersecurity Announcer]

ChatGPT5 is the worst model ever.

[Gary Marcus]

GPT5 is horrible. Users were saying the new model felt dumber, more robotic and straight up less helpful than before. I kept telling people for years, GPT5 is not going to be magic. And the whole industry like dumped on me. They all went after me for even daring to raise the question of like, was this idea of scaling, which also is partly behind the driverless cars, was it going to kind of produce these magical results? They hated me for it. And you know, now GPT5 came out and people are starting to realize that I had a point, but it took, you know, three years of just like constant hype and constant not quite meeting expectations for people to realize that maybe this stuff is oversold and every incentive is to overhype these things. So, for example, if you're a journalist, you write a story about, yeah, it's sort of going slowly, but it's not, you know, it's not really there yet. Nobody wants to read that. They want to read Tomorrow the world will change and your, your children won't need to learn to draw. Like, that's a much more exciting story.

[Dena Templewastin]

Gary says this mismatch between hype and reality, while it's been mostly bypassed so far, he thinks there'll come a point when people stop turning a blind eye.

[Gary Marcus]

I'll go on public record as saying some aspect of AI, whether it's driverless cars or LLMs leading people to delusions or something, that some aspect will actually be a fairly big part of the 2028 election. Because we have all these people running without regulation and we've seen what happens with that before, like with, you know, banks ran without regulation for a while and then we had a banking crisis. Like nobody should be surprised if something pretty bad happens. Cybercrime, driverless car, something like that in the next couple of years. Such that by time 2028 comes a bunch of the public is like, hey, this has got to stop. You need to do a better job here of Regulating.

[Dena Templewastin]

From Recorded Future News this has been Click Here's Mic Drop. It was written and produced by Megan Dietre, Sean Powers, Erica Gaeta, Zach Hirsch, Lucas Riley and me, Dina Temple. Rest. It was edited by Karen Duffin. We'll be back on Tuesday with an all new episode of Click Here. Have a great weekend.

[Cybersecurity Announcer]

If you're looking for a daily guide to cybersecurity news and policy, sign up for the Cyber Daily from Recorded Future News. It serves up today's most interesting and important cyber stories from our sister publication the Record, and then aggregates all of the big cyber stories you might have missed from news outlets around the world. Just go to the Record Media and click on Cyber Daily to get all you need to know about the world of cybersecurity right in your inbox.