Home
登录
注册
学习内容
Loading...
Self-Driving Cars are DOOMED. The $2 Trillion AI Disaster - Video học tiếng Anh
听力练习
听力练习
/
Video
/
The Infographics Show
/
Self-Driving Cars are DOOMED. The $2 Trillion AI Disaster
Self-Driving Cars are DOOMED. The $2 Trillion AI Disaster
选择学习模式:
查看字幕
选词
重写单词
Highlight:
3000 Oxford Words
4000 IELTS Words
5000 Oxford Words
3000 Common Words
1000 TOEIC Words
5000 TOEFL Words
字幕 (161)
0:00
Humans are terrifying drivers… and Waymo proves it.
0:04
The company’s robotaxis have driven millions of miles… and caused far
0:08
fewer injuries than humans. In fact, they’re almost 10.4 times safer. Yet by late 2025,
0:14
there were only 2,500 robotaxis on U.S. roads, compared to 297 million human driven cars.
0:21
The software works. Waymo has the brains, the money, and the data. But one question remains…
0:27
I’m Josh, and on today’s episode of The Infographics Show, we’re asking if
0:31
self-driving cars work… where are they? Chapter One: The Invisible Tracks
0:37
The Waymo system is actually more complicated than you think, which
0:40
makes sense, considering its parent company is Alphabet. The Waymo autonomous vehicles
0:45
engage with the road in a fundamentally different way than human drivers do.
0:48
The greatest strength - and greatest weakness - of Waymo autonomous vehicles is that they think
0:53
like a machine. They don’t reinvent driving; they take the tech we already use every day:
0:59
cruise control, collision warnings, lane assist. They combine that with cameras,
1:03
LIDAR sensors - that’s Light Detection and Ranging - and incredibly detailed maps of
1:08
the cities they operate in, collected in a way that’s surprisingly similar to Google Street View.
1:13
While we can only look in one direction at a time - checking mirrors and blind
1:17
spots - Waymo sees everything from above. It’s like the car is tracing its finger across a map,
1:23
navigating the streets from a bird’s-eye view rather than from inside the driver’s
1:27
seat. You’re sitting in a remote controlled car being piloted by an AI system which has
1:33
plotted a fixed path. Once the route is mapped out in its 3D model of the world,
1:37
the AI takes over. It controls the actuators, the systems that handle steering, acceleration,
1:42
and braking. It’s less like driving… and more like the car is being puppeteered in real time.
1:48
This is all pretty cutting edge stuff,
1:50
tech that would seem exclusive to the world of science fiction like 20 years ago.
1:55
But how does that account for those insanely impressive safety statistics?
1:59
Because what it’s really doing is on another level.
2:02
In addition to all those other stats, Waymo cars cause around 80% fewer incidents
2:08
involving injured cyclists and motorcyclists. At intersections, injury crashes drop by over 90%.
2:14
Overall, crashes involving suspected serious injuries are down by about 92%
2:20
By June of 2025, Waymo was claiming to have logged over 96 million miles (154 million km), and its
2:26
accident rate was far lower than human-driven ride-sharing services like Uber and Lyft.
2:31
That’s all because of human error. People bend the rules when we think nobody is
2:35
looking. Our minds wander off when we should be focusing. We drive tired,
2:40
upset, sick, or in extreme cases, drunk or high.
2:44
Self-driving cars don’t have that issue.
2:46
They’ve got hard-coded rules, obstacle avoidance algorithms, predictive modeling,
2:51
and object detection to keep them on the straight
2:53
and narrow. Where people are sloppy, self-driving cars are disciplined.
2:58
That can be a problem.
3:00
The exact rigidity in following the plan that makes a Waymo ride so safe is also
3:05
one of its Achilles’ Heel.The real test for Waymo isn’t highways, it’s cities.
3:10
Proving the system works there means it’s not just viable, but potentially the next
3:15
step toward a world where no one needs to drive. Because cities aren’t static. They’re constantly
3:20
changing, unpredictable, almost like living organisms. When your entire system depends
3:25
on following a highly detailed, pre-built map, that map has to evolve just as fast.
3:30
If it doesn’t, everything starts to break down. Maybe a water main bursts. A traffic light
3:35
fails. A new lane suddenly appears, or construction takes over the street.
3:39
Orange cones, temporary signs, detours, workers in high-vis moving everywhere. For a human,
3:45
it’s an inconvenience. Something that’s stopping you from getting home sooner.
3:48
But for a Waymo car, it’s a different story.
3:51
If Alphabet hasn’t updated the maps, this kind of thing is catastrophic.
3:55
When you break the invisible track that all these cars operate on,
3:58
they can’t improvise the way a human could. The only question is can Alphabet actually
4:03
keep up with constantly remapping every inch of a city to make their driverless cars run smoothly?
4:08
Chapter Two: The Long Tail of Terror
4:11
This might be a disadvantage for Waymo, and account for some of the barriers facing mass
4:15
adoption of robotaxis, but remember what we said earlier? This isn’t some small Silicon Valley
4:21
startup. Alphabet, Waymo’s parent company and owner of Google and YouTube, has a market cap of
4:27
over $2 trillion. It also has access to the finest programming and engineering minds on the planet.
4:33
Everything they touch seems to turn into gold, well except Google Gemini. They have clearly put
4:38
their money where their mouth is, adapting the tech and upgrading their mapping game.
4:42
Every year, they add more cities to their roster, with half a dozen at the time of
4:46
this video’s production. In February of 2026, they also began the rollout
4:50
of their 6th generation vehicles. By this point, they’d clocked over 200 million
4:54
miles (322 million km), and a vision system that their official press releases describe as, quote,
4:59
“far beyond the capabilities of human sight or standard automotive cameras.”
5:04
The cars have EARS now, too.
5:06
It’s short for External Audio Receivers. These devices mainly detect the sounds of emergency
5:11
vehicle sirens to make sure Waymo doesn’t get in the way of ambulances, fire trucks, or police.
5:17
Alphabet has the money and the resources to scale what had once been their pet project.
5:21
They’re getting better and better at mapping and remapping cities, so it’s clear that the
5:26
fluid nature of city structure isn’t going to be a barrier for long. But building new
5:30
technology in the real world comes with the Whack-A-Mole effect. Fix one problem,
5:34
and another one pops up before Alphabet can even react. Even if driverless cars can get
5:39
the sense of a location, they’re never going to understand people.
5:43
It’s a genuine problem.
5:45
People in the zone call them “Edge Cases”. These are the bizarre, highly specific
5:50
scenarios that are impossible to predict in advance or program a universal response for.
5:54
It’s the horrors of “the long tail”, as statisticians put it. These are the
5:59
rare but critical anomalies that humans can understand and react to instantly, but would
6:04
leave a computer completely stumped. If a truck with a bed full of chicken crates took a turn
6:09
too sharply and filled the road with screeching poultry, a person would know to slam the brakes.
6:14
Would a Waymo? What if a car is driving down Main Street on a boozy Halloween night, and someone in
6:20
an inflatable dinosaur costume drops a mirror that confuses the vehicle’s digital sensors?
6:25
Sure, they’re hypotheticals, but even more deranged things have actually happened.
6:29
Self-driving cars have driven straight into floodwaters because they were never specifically
6:34
programmed not to. They also don’t know how to avoid an active crime scene where police have
6:39
their guns drawn, something a human driver would have the good sense to avoid most of the time.
6:44
And in one of the most horrifying incidents we found, a non-Waymo
6:48
self-driving car in San Francisco in 2023 ran over a pedestrian and
6:53
dragged them 20 feet (6 meters), causing multiple serious injuries.
6:57
That’s… not a very human choice.
7:00
That said, it’s not like Waymo hasn’t been involved in some high profile
7:03
cases that might’ve made the public a little nervous about getting into a driverless car.
7:08
Waymo vehicles have consistently been criticized for speeding past stopped school buses at
7:13
dangerous speeds. In 2023, a Waymo ran over a dog. In the same year, two cars crashed into the same
7:19
truck in quick succession. In 2024, they hit a cyclist and ran into a telephone pole. And in
7:24
one of the most bizarre long tail scenarios of all, Waymo attracted massive controversy
7:29
in San Francisco when one of their cars hit and killed Kitkat, an internet famous Bodega Cat.
7:36
It caused such an uproar, that a local Supervisor campaigned for legislation that would allow local
7:40
governments to make self-driving cars illegal. And that feels pretty mild compared to a 2026 incident
7:46
where a Waymo taxi blocked an ambulance heading to the scene of a mass shooting.
7:51
Humans are victims of human error.
7:54
Driverless vehicles can fall victim to machine error.
7:57
That’s something we’ve all become a little more familiar with in the age of generative AI.
8:01
These situations are the kinds of things that only humans can understand because, well,
8:06
we’ve lived as humans. We can understand what’s happening from within. Sometimes,
8:12
actual truth and reality can be lost in the reams of raw data that driverless cars are trained on.
8:17
And critics have brought up that it’s hard to ever train computers on these scenarios. That’s because
8:22
actually exposing the cars to niche dangerous situations would be both dangerous and extremely
8:28
costly, given the risks exposed to the hardware. Even then, the problem with long-tail scenarios
8:33
is simple: no matter how many you train for, the world is chaotic, and new ones will always pop up.
8:38
One proposed solution is using virtual AI “proving grounds” where the computers that
8:43
run Waymo vehicles can explore as many scenarios
8:45
as possible. It improves their batting average, but the same problem remains…
8:50
They still can’t predict everything or react in real time.
8:53
If a Waymo car freezes up during an unexpected scenario, it might have a
8:58
death toll. That’s the worst case but still an extremely possible scenario. However,
9:03
in a dire situation, Waymo does have a hidden contingency plan in place.
9:08
One that reveals a secret that some would argue goes against
9:11
the entire point of true self-driving cars.
9:14
One advantage of self-driving cars? You’ve got more time to watch videos like this,
9:18
so make sure to like, share and subscribe. Don’t worry, the car won’t judge if you binge watch.
9:24
Chapter Three: The Ghost In The Machine
9:26
Self-driving or autonomous vehicles can sometimes be as much a case of
9:30
marketing spin as they can be a technical term for an emerging
9:34
science. That’s because if an emergency disables the self-driving features,
9:38
control of the car will likely be handed over to an overworked technician in the Philippines.
9:43
That’s right.
9:44
For all of Waymo’s talk about these autonomous cars being better than human drivers,
9:48
in a true emergency it’s a human, thousands of miles away, who is expected to save you. Waymo
9:55
has been squirrely about this, saying that these remote workers - who are often assigned to as many
10:00
as 40 different vehicles - don’t drive the cars. They just provide “guidance” to the AI system.
10:05
But at a point, isn’t that really just all semantics?
10:09
And relying on third-party emergency drivers isn’t nearly as exciting as
10:13
the idea of a car that does everything the hype claims. It’s also a major safety risk,
10:19
since it opens the system up to potential hacking by a malicious actor.
10:23
In the words of Senator Ed Markey, who voiced opposition to the system,
10:27
“Overseas remote assistance operations may be more susceptible to physical takeover by
10:32
hostile actors… granting them driver-like control of thousands of vehicles… [they]
10:37
could quickly become the weapons of foreign actors seeking to harm innocent Americans.”
10:41
And when you put it in terms of “your cool driverless taxi might
10:45
suddenly become a giant RC car for a terrorist ten thousand miles away”,
10:50
you can see why this leaves people feeling a little concerned.
10:53
Chapter Four: The Capital Nightmare
10:55
When you pull back the curtain at the processes that make Waymo’s impressive
10:59
safety stats possible, it becomes clear that getting this system online is far
11:04
from a walk in the park. If the autonomous taxi industry eventually reaches a global fleet of,
11:09
say, 10 million driverless vehicles, we’re looking at an emergency remote
11:13
assistance agent to every 40 or so cars. That means at least a quarter
11:18
of a million of these technicians need to be active and ready at all times.
11:22
Think about the logistics. A huge number of employees that need to be hired, trained,
11:27
paid, and given offices and equipment to do a job that is literally life or death.
11:32
That kind of up-front development cost would hurt even for Alphabet. It makes you wonder:
11:37
Doesn’t just having a driver in the car seem a lot less complicated,
11:40
even if the risk of getting into an accident is superficially higher?
11:44
Because the truth is, Waymo never really got rid of the driver.
11:47
They just moved them.
11:49
Ultimately, if you want to experience being taken from one place to another at affordable
11:53
prices without needing to drive yourself… we recommend you look into getting a bus pass.
11:58
Now check out “How Jaguar Destroyed Its 102 Year Legacy in 30 Seconds”, or watch this instead!
