Latest / Elon Musk Podcast / California Police Tickets Tesla and Self Driving Cars

Transcript

California Police Tickets Tesla and Self Driving Cars · Elon Musk Podcast

0:00
California law enforcement now issues formal traffic citations
0:03
directly to the artificial intelligence software operating
0:07
Waymo and Tesla vehicles. Right, which legally
0:10
reclassifies the corporate manufacturer as the actual
0:13
driver of the vehicle. I mean, that completely erases
0:16
the loophole that previously allowed empty vehicles to commit
0:19
traffic violations without any consequence.
0:22
Yeah, and you know this overlaps directly with Tesla initiating
0:25
mass production of vehicles completely lacking steering
0:27
wheels or pedals. Exactly.
0:29
So if the software is legally the driver, how exactly do we
0:33
judge a machines mistakes in a court of law when it inevitably
0:37
crashes? Well, we are basically watching
0:39
the entire legal framework of transportation being rewritten
0:43
in real time for generations. The Vehicle code was built
0:47
around a very specific assumption, right?
0:50
A natural person sits in the front seat, holds a physical
0:52
steering wheel and makes conscious decisions, right?
0:55
Because the law applies to human behavior.
0:57
Speeding, failing to yield, running a red light.
1:00
Those are citations written to a person who carries a state
1:02
issued driver's license. Yeah, a police officer pulls you
1:05
over, walks up to your window and asks for your physical
1:09
documentation. The entire justice system for
1:12
the road relies on the presence of a human being to accept
1:15
responsibility. Which is exactly why the
1:18
introduction of the Notice of Autonomous Vehicles
1:20
noncompliance is so structural. This is the new mechanism that
1:24
allows police officers to formally cite a corporate
1:27
manufacturer for moving violations.
1:29
Because before this existed, law enforcement officers were
1:33
effectively paralyzed when dealing with autonomous fleets.
1:36
I mean, imagine you are a patrol officer right?
1:39
You observe a vehicle blow through a stop sign.
1:41
Standard procedure dictates you hit your sirens, pull the
1:44
vehicle over, request a license and registration.
1:47
Yeah, when you walk up to the glass and look inside, there is
1:50
no operator. The driver's seat is completely
1:53
empty. The steering wheel might be
1:55
turning itself, but there's no human to hand a pen to.
1:58
The entire enforcement protocol just breaks down on the side of
2:01
the road. It really does.
2:03
The officer cannot arrest a computer.
2:05
They cannot suspend a software program's non existent driver's
2:09
license. There's a very specific, well
2:12
known incident that illustrates this perfectly.
2:15
Oh, the one in San Bruno. Yes exactly.
2:18
An officer in San Bruno was on duty and watched an empty Waymo
2:22
execute a completely illegal U-turn right in the middle of a
2:25
public Rd. Oh wow, so the officer witnessed
2:28
it happened right there. Yeah, witnessed it clearly.
2:30
The vehicle was undeniably in the wrong.
2:32
The officer activated their emergency lights and pulled the
2:34
robo taxi over, but the officer was entirely powerless to issue
2:38
a ticket. Because of the driver rule.
2:40
Right. The state law literally required
2:42
a natural person to be behind the wheel to receive the
2:44
citation. The officer just stood there
2:47
looking at an empty car and you'd only verbally notify the
2:50
company through a support speaker that their car did
2:53
something illegal. That creates a massive
2:55
accountability void. Think about the consequences if
2:57
you, a human driver, execute an illegal U-turn.
3:01
You get a ticket, you get points on your license, your insurance
3:04
premiums increase. And if you do it enough times,
3:07
the state takes away your privilege to drive.
3:10
It is a documented behavioral failure that follows you but for
3:14
an empty robo taxi. The incident essentially
3:17
vanished into a corporate server.
3:18
So the Police Department had no legal mechanism to find the
3:22
algorithm. None.
3:24
It created a situation where tech companies could test their
3:27
fleets on public roads without generating a public government
3:31
documented paper trail of their software failures.
3:34
Well the notice of autonomous vehicle non compliance changes
3:37
that dynamic entirely. Creating a formal, government
3:41
documented paper trail of software failures limits a
3:44
companies ability to claim their testing phase is flawless.
3:48
Exactly. When you have a stack of
3:50
official citations issued by law enforcement sitting in a public
3:53
database, you can no longer go on television or tell your
3:56
investors that your safety record is unblemished.
3:59
The state is now creating an independent, legally binding
4:02
record of every single time the algorithm fails to follow the
4:05
rules of the road. It strips away the corporate
4:06
secrecy. Yet previously, if a car made a
4:09
mistake that did not result in a crash, the company could just
4:12
internally log it as a learning opportunity and delete the
4:15
footage. Now, the notice of non
4:18
compliance fundamentally shifts the power dynamic from the
4:22
manufacturer's public relations department directly to the
4:25
state's regulatory body. Wait back up.
4:28
I am trying to visualize this. A cop writes a ticket, but there
4:31
is no driver to roll down the window.
4:34
How do you physically hand a citation to a car with no
4:37
driver? That is a great question.
4:39
Are they just slipping a piece of paper under the wiper blade
4:41
of a robot and hoping a technician eventually finds it
4:44
when the car goes back to the charging depot?
4:46
No, no. The process is entirely
4:48
digitized and tightly regulated. The strict reporting mandates
4:51
require that these citations are not just left on the windshield
4:54
to blow away in the wind. So how does it reach the
4:56
company? When an officer issues one of
4:58
these notices, they input the vehicles unique fleet
5:01
identification number into their cruisers terminal.
5:04
The autonomous vehicle companies are then legally required to
5:07
report these specific violations to the state motor vehicle
5:10
department within a very tight window.
5:13
OK, so the physical paper doesn't matter at all.
5:15
The digital citation triggers an immediate compliance countdown
5:19
for the corporate headquarters. Exactly.
5:20
And if there is a collision involved, that reporting window
5:23
shrinks drastically, requiring them to notify the state even
5:27
faster. The system forces the companies
5:30
to maintain a constant open pipeline of data with the state
5:33
regulators. I see.
5:34
So it is not just about a multibillion dollar company
5:37
paying a $200 fine. The fine is completely
5:40
irrelevant to them financially. Totally irrelevant.
5:43
It is about logging the behavioral failure of the
5:46
software in a centralized state database.
5:49
Regulators can now track exact patterns.
5:52
If software version 4.2 is repeatedly running SKOP signs in
5:56
a specific residential neighborhood, the state has the
5:59
concrete data to identify that trend and mandate a recall.
6:03
Handing out tickets for rolling through a stop sign is one
6:05
thing, but dealing with active emergencies requires an entirely
6:08
different level of machine compliance.
6:10
Oh absolutely. If we look at the new emergency
6:13
operational requirements, there is a strict mandate for a two
6:16
way audio link. A remote human operator sitting
6:20
in a Control Center must answer within seconds of a police
6:24
officer or firefighter making contact with the empty vehicle.
6:27
Which makes sense, because that directly addresses the extreme
6:30
friction between emergency responders and autonomous
6:33
systems. Picture an active emergency
6:36
scene. A building is on fire, right
6:38
Fire trucks are trying to navigate a narrow city street,
6:41
hoses are being dragged across the pavement, and police are
6:44
securing a perimeter. They need civilian vehicles out
6:46
of the way immediately. Yeah, a human driver can be
6:49
shouted at, directed with frantic hand signals, or
6:52
aggressively told to reverse down an alleyway.
6:54
A human understands the urgency in context.
6:57
But a machine only understands its base programming, and the
7:00
programming usually tells the machine to simply stop moving
7:03
when it is confused, which is the exact opposite of what first
7:06
responders need. Yeah, if there is a massive
7:09
fire, an empty robo taxi stopping dead in the middle of
7:12
the only access Rd. blocks the ladder engines from reaching the
7:15
building. So this two way audio link
7:17
forces the tech companies to have a human being physically
7:20
ready to take responsibility, speak through the car's exterior
7:24
speakers, and communicate with the authorities on the ground
7:27
almost instantly to figure out how to move the vehicle.
7:30
Furthermore, we have the emergency geofencing directive.
7:34
Local officials can now digitally cordon off an active
7:37
emergency zone Wait. How does that actually work?
7:40
Basically, the police dispatcher draws an invisible digital
7:43
perimeter around A4 block radius where an incident is occurring.
7:48
The automated vehicles are legally required to clear that
7:51
specific area almost immediately upon receiving the wireless
7:54
signal. Oh wow.
7:55
Yeah, they must evaporate from the zone and prevent any other
7:58
cars in their fleet from entering.
8:00
Think about the staggering technical infrastructure
8:03
required to actually pull that off.
8:05
It forces tech companies to essentially run massive real
8:08
time emergency dispatch centers of their own.
8:11
Exactly. They have to maintain a staff of
8:13
remote operators large enough to handle simultaneous emergencies
8:17
across multiple major cities. They need the digital
8:19
architecture to push geofence updates to thousands of moving
8:23
vehicles, instantly overriding their destination programming.
8:27
It shifts these companies from being pure software developers
8:30
to acting as municipal utility operators.
8:33
They are managing public infrastructure flow in real
8:36
time, just like the power grid or the water supply.
8:39
It is an incredibly demanding requirement and honestly I think
8:42
it is an unfair standard. Really.
8:44
Why is that? Well, we are holding the
8:47
machines to a standard of perfection that human beings
8:50
never achieve. Human drivers frequently panic,
8:54
freeze, or cause absolute gridlock around fire trucks and
8:57
ambulances. You see it on your own commute.
9:00
That is true. People pulling over in the wrong
9:01
direction, blocking intersections because they get
9:03
flustered by the deafening sirens and flashing lights.
9:06
Humans are terrible in emergencies, yet we expect the
9:09
software to perfectly and silently evaporate from the
9:12
scene without causing any friction whatsoever.
9:14
I completely disagree with you on that.
9:16
If a machine cannot panic, it has no excuse for lingering in a
9:19
danger zone. But they are still navigating
9:22
unpredictable physical spaces. Sure, but humans freeze because
9:26
our biology gets overwhelmed. You hear a sudden air horn, your
9:30
adrenaline spikes, your heart rate jumps, and your brain
9:33
temporarily short circuits trying to process the threat.
9:37
A computer does not have a biological nervous system.
9:40
It does not get scared of flashing lights, it possesses
9:43
absolute perfect spatial awareness from its 360° radar
9:47
and LIDAR sensors, and it has immediate access to high
9:51
definition digital mapping fairpoint.
9:53
If it possesses those superhuman advantages, it should absolutely
9:57
be held to a higher standard of compliance.
9:59
It should not be allowed to act like a confused, terrified
10:02
human. OK, that is a fair counter
10:04
argument regarding its raw processing capabilities.
10:07
But whether it is an unfair double standard or not, it is
10:10
now the law, and this opens up the ability for regulators to
10:13
impose escalating penalties if a corporate fleet repeatedly
10:17
obstructs first responders. Right, it is no longer just a
10:19
warning. Exactly.
10:20
The state can actively restrict their fleet sizes.
10:23
If your car is block an ambulance, the state can legally
10:26
cap the number of vehicles you were allowed to deploy.
10:28
They can limit your operating hours to just the middle of the
10:31
night, or they can pull your operating permits entirely.
10:34
The state will literally force you to shut down your servers
10:37
and pull your physical cars off the public roads.
10:40
And these strict emergency mandates and compliance rules
10:43
are colliding directly with a massive hardware rollout
10:46
happening a few states over. Yeah, we have to look at the
10:48
hardware reality of the cyber cab.
10:50
We are watching Tesla's production ramp up at
10:53
Gigafactory Texas and the physical design of this vehicle
10:56
is radical. Radical is one word for it.
10:59
It is A2 seat vehicle built entirely with without a steering
11:02
wheel or pedals. There is no manual override
11:05
mechanism anywhere in the cabin. None at all.
11:08
Right, it runs purely on a vision based autonomous driving
11:11
system. It relies entirely on exterior
11:14
cameras to perceive the environment, process the visual
11:17
data through its neural net, and execute every single driving
11:20
task. The volume they are targeting is
11:22
critical to understanding the impact of this.
11:25
They have a production target of 2 million units annually. 2
11:28
million. This is not a small boutique
11:31
project. It is not a limited run test
11:34
fleet operating in a single sunny neighborhood.
11:37
Right. That production figure is meant
11:38
to completely replace their current consumer models as the
11:42
absolute highest volume vehicle they manufacture.
11:45
They are betting the entire industrial capacity of their
11:49
Texas factory on a vehicle that physically cannot be driven by a
11:52
human being. Removing the steering wheel
11:55
entirely limits human intervention TO0.
11:58
Imagine you are a passenger sitting inside inside this
12:00
vehicle. You are cruising down the
12:02
highway and you suddenly feel unsafe.
12:04
Right. Maybe the car is navigating a
12:05
complex construction zone very poorly, or perhaps it is driving
12:09
far too fast for a sudden torrential downpour.
12:12
There are no physical controls for you to grab.
12:14
You cannot slam your foot on a brake pedal.
12:16
You cannot grab a wheel and steer away from a swerving semi
12:19
truck. You are completely, utterly
12:22
captive to the decision making of the software.
12:24
So the hardware is totally committed.
12:26
There is no going back to human driving once that car rolls off
12:29
the assembly line. Creates incredible regulatory
12:31
friction. The hardware is designed from
12:34
the ground up for Absolute 0 human supervision.
12:38
The factory in Texas is churning out vehicles that physically
12:42
cannot be supervised or overridden by a human occupant.
12:46
But current state laws, particularly those strict
12:49
enforcement mechanisms on the West Coast we just discussed,
12:52
are actively tightening the leash on unsupervised
12:55
operations. Exactly.
12:56
You have hardware built for a purely autonomous, theoretical
13:00
critical future, entering a legal environment that is
13:03
actively hostile and increasingly skeptical of
13:05
unsupervised algorithms. And Tesla's push for this
13:08
unsupervised hardware is heavily complicated by a very recent,
13:11
highly consequential legal battle over what they've been
13:14
calling their software for the past decade.
13:16
Yeah, a state regulatory ruling recently found that Tesla
13:18
violated consumer protection laws by misleadingly marketing
13:22
its advanced driver assistance features.
13:24
For years they were using the terms Autopilot and Full Self
13:28
Driving to sell software to consumers.
13:30
But that software, despite the names, still legally required a
13:34
human being to pay constant attention to the road and keep
13:38
their physical hands on the wheel, ready to take over.
13:41
The immediate fallout from that regulatory ruling was severe.
13:44
The state threatened to suspend their manufacturing and dealer
13:47
licenses. Wow.
13:48
That is the nuclear option. It really is to avoid having
13:52
their ability to build and sell cars in the entire state
13:55
completely shut down. Tesla complied with the order.
13:58
They changed the marketing language on their web website
14:00
and in their vehicles. They renamed the software to
14:03
explicitly include the word supervised, changing it to full
14:07
self driving supervised. This made it legally explicitly
14:11
clear to the consumer that the human driver is still the one
14:14
responsible for the vehicle. But the twist here is
14:17
fascinating. Immediately after complying with
14:19
the order, immediately after changing the name and adding the
14:21
supervised label to satisfy the regulators and keep their
14:24
licensing in, Tesla turned around and sued the motor
14:27
vehicle department to reverse the original false advertiser
14:30
ruling. Wait, they sued after they
14:32
already changed it? Yes, they accepted the penalty,
14:35
implemented the requested change their entire fleet, and then
14:39
launched a massive lawsuit to wipe the original regulatory
14:42
finding off the books. Why spend the money to sue after
14:46
you have already complied? It all comes down to the
14:48
liability shield in civil lawsuits involving severe
14:51
crashes, Tesla's primary legal defense in court has always been
14:56
that the human driver was supervising the system.
14:59
Oh, I see. They argued to the judge that
15:01
the software is just a fancy assistance tool like advanced
15:04
cruise control, and the human operator is the one legally and
15:08
morally responsible for the vehicle's safe operation.
15:11
Precisely having a formal documented state ruling on the
15:14
books officially calling the company a false advertiser
15:18
regarding their self driving capabilities completely
15:20
undermines that exact defense in front of a jury.
15:23
Because imagine you are severely injured in a crash and you sue
15:26
the manufacturer. The manufacturer's lawyers stand
15:28
up in court, point at you, and say you were supposed to be
15:31
washing the road. The manual says you must pay
15:34
attention. But if your lawyer can present a
15:37
formal government ruling to the jury that says this company
15:40
actively and falsely advertise as the car as capable of driving
15:44
itself, the jury is going to look at the corporate
15:47
manufacturer, not you. It changes the legal burden
15:50
entirely. Consumers can now point to the
15:53
state's official finding to argue they were genuinely misled
15:56
into thinking the car could drive itself safely while they
15:59
checked their phone, which is exactly why they took their
16:01
hands off the wheel. The corporate marketing itself
16:04
becomes the legal reason for the driver's negligence.
16:07
Now, while the regulatory environment on the West Coast is
16:10
highly adversarial, entirely focused on enforcement citations
16:14
and false advertising battles, other states are actively
16:17
rolling out the red carpet to attract these exact same fleets.
16:21
Yeah, we can look closely at the Texas preemption model.
16:24
Texas utilizes an authorization model managed directly by the
16:28
state's motor vehicle department.
16:30
It is a legislative framework specifically designed to
16:32
facilitate rapid commercial deployment rather than slow
16:35
incremental permission. The concept of preemption is the
16:39
entire key to their strategy. The state law explicitly forbids
16:44
local cities, counties or any political subdivisions from
16:47
imposing their own regulations, franchises or rules on automated
16:51
vehicles. So a local City Council in Texas
16:53
cannot pass an ordinance demanding a special congestion
16:56
tax on robotax. Right, They cannot limit their
16:59
operating hours around schools or require them to paint their
17:02
cars a specific neon color for visibility.
17:05
The state government in Austin has declared absolute top down
17:09
authority over the entire issue. Contrast this with the heavy
17:12
municipal friction you see elsewhere.
17:15
Without preemption, a tech company has to go city by city,
17:18
sitting in endless local council meetings, negotiating with every
17:21
single mayor, Alderman and transportation committee just to
17:24
get permission to map a few streets.
17:26
It is a logistical and political nightmare for a tech company
17:28
whose entire business model relies on scaling quickly
17:32
Because of preemption. In Texas, autonomous fleets can
17:34
launch commercial operations in Austin, Dallas and Houston
17:37
simultaneously without negotiating a single contract
17:40
with those individual city councils.
17:42
This preemption opens up immediate commercial deployment.
17:46
Companies essentially just need to certify to the state that
17:48
their vehicles can achieve A minimal risk condition during a
17:52
critical failure and they must maintain a massive insurance
17:55
policy. A minimal risk condition
17:57
basically means that if the main driving computer crashes or the
18:00
sensors are blinded by mud, the car has a backup system that
18:04
knows how to smoothly pull over to the side of the road, turn on
18:07
its hazard lights, and stop safely.
18:09
Rather than just freezing violently in the middle of a 70
18:12
mile per hour fast lane, right? If they can submit paperwork
18:16
proving that capability and they show proof of the insurance
18:19
policy, the state hands them the authorization to start charging
18:22
the public for rides. Hold on, what if the car is
18:25
actually dangerous in practice? If the state just hands out
18:29
authorizations based on paperwork without conducting
18:31
their own deep, independent testing of the software?
18:35
What happens when a fleet inevitably starts causing
18:37
accidents in downtown Houston? Is there a kill switch?
18:41
The enforcement mechanism in this model is entirely reactive.
18:45
The state retains the power to revoke the authorization if they
18:49
determine the vehicles operation is actively likely to result in
18:52
serious bodily injury. OK, so they can pull the plug.
18:55
Yes, and operating a fleet after that authorization is pulled
18:58
becomes a criminal misdemeanor for the corporate officers.
19:01
So the state's strategy is to step out of the way, allow the
19:04
innovation to happen as quickly as possible, but maintain a very
19:08
heavy criminal hammer to drop if the technology proves hazardous
19:12
to the public. So Texas relies heavily on
19:15
stopping operations only after things get dangerous on the
19:17
road. A neighboring state relies on
19:20
rigorously recording the danger exactly as it happens.
19:23
Right. This brings us to the Arizona
19:25
black box standard. Arizona law forces autonomous
19:28
vehicles to maintain a constantly rolling data
19:30
recorder. It acts exactly like a flight
19:32
data recorder, the black box on a commercial airplane.
19:35
The strict data requirements encoded in this law are
19:38
fascinating. The vehicle must record high
19:40
fidelity operational data for 30 seconds before and 30 seconds
19:44
after any collision. That part is somewhat expected.
19:48
I mean, if a crash happens, we obviously want the telemetry to
19:50
know why, but more importantly, the law mandates it must record
19:55
and preserve data for any near miss.
19:58
And the law does not leave a near miss up to interpretation.
20:01
It specifically defines it. It is not just a passenger's
20:03
subjective feeling of danger. Right.
20:05
A near miss is legally defined as avoiding a collision by a
20:09
margin of less than two seconds or any situation requiring
20:13
emergency hard braking. So if the vehicle's radar
20:16
detects A pedestrian stepping off the curb, and the algorithm
20:19
has to slam on the brakes or swerve violently to avoid an
20:21
impact within a literal 2 second threshold, all the data from
20:25
that event must be preserved by law.
20:28
This severely limits corporate secrecy.
20:30
In the event of a crash or even just a close call, the exact
20:34
operational parameters of the algorithm are permanently
20:37
captured and must be provided to state authorities.
20:40
The manufacturer cannot hide behind a vague PR statement
20:43
about unexpected road conditions or unavoidable circumstances.
20:47
No. The data recorder captures the
20:49
raw sensor inputs, the exact millisecond of the software's
20:52
reaction time, the precise steering angle, the hydraulic
20:55
braking force. Every single calculation the
20:58
machine made is laid bare. I still argue this is an
21:02
impossible standard to enforce evenly across the transportation
21:06
network. Human drivers experience near
21:08
misses every single day without having to report them to the
21:11
government. Wait, really?
21:12
You think it's impossible? Yeah, think about it.
21:14
Every time you drive to work and slam on your brakes because
21:16
someone aggressively cut you off in traffic, that qualifies as a
21:19
near miss. We do not require human beings
21:22
to upload a data log to the state every time they brake.
21:25
Hard to avoid a Fender Bender. Why should the machine be
21:29
burdened with this massive, expensive data collection
21:32
mandate for events that do not even result in a scratched
21:35
bumper? Well, because humans don't
21:36
operate on centrally updated collective algorithms.
21:39
What do you mean? If I learn a lesson from a near
21:42
miss today, say I learned to watch out for blind spots on a
21:45
specific intersection, it only makes me individually a better
21:49
driver. It doesn't instantly magically
21:51
improve the driving skills of every other human on the road.
21:55
I see where you're going with this, right?
21:56
But with an automated fleet, fixing one machines near miss
22:00
fixes the entire global fleet instantly.
22:03
If a robotaxi in Phoenix almost hits a Edestrian because its
22:06
vision system got confused by a shadow, the data from that near
22:10
miss is analyzed. The core code is rewritten by
22:12
engineers and the software update is pushed to every other
22:15
vehicle on earth. OK, that is a massive
22:18
difference. Making the data vital for
22:20
continuous fleet wide safety improvements absolutely
22:23
justifies the strict reporting. All this mandated data recording
22:27
and the new traffic ticketing mechanisms feed directly into
22:30
the much larger problem of how the justice system actually
22:33
handles a civil lawsuit against a robot.
22:36
This is where we have to look closely at the legal theory
22:38
analysis regarding how to judge automated systems in a court of
22:41
law. Currently, the justice system
22:43
relies on traditional product liability.
22:46
Right. If your kitchen toaster suddenly
22:48
explodes and burns your house down, you sue the manufacturer
22:51
for selling a physically defective product.
22:54
But applying traditional product liability law to an artificial
22:58
intelligence driving a car is proving to be a massive failure
23:01
in the courts. The failure stems entirely from
23:03
the concept of information asymmetry.
23:05
Can you explain that? Sure, think of information
23:08
asymmetry like going to a mechanic, but the hood of your
23:11
car is welded shut and only the manufacturer has the key.
23:15
If your brakes fail, you know they failed, but you cannot
23:18
prove which specific gear snapped because the company will
23:22
not let you or the jury look inside the engine.
23:25
Because it is nearly impossible for a regular citizen to prove a
23:28
specific software coding flaw caused their crash, the driving
23:32
algorithms are heavily protected proprietary black boxes.
23:36
Exactly. If you sue a car company, your
23:38
lawyer has to prove exactly what line of code failed.
23:41
The manufacturer possesses thousands of elite software
23:44
engineers and millions of lines of proprietary code.
23:47
A plaintiff simply cannot penetrate that wall of corporate
23:50
secrecy to prove a design defect.
23:52
So legal scholars are actively exploring alternative
23:56
frameworks. We have a few competing legal
23:59
standards trying to replace product liability specifically
24:03
for autonomous vehicles. The first major alternative
24:07
being debated is strict liability.
24:09
Under this legal theory, the manufacturer simply pays for any
24:13
damage the vehicle causes, regardless of fault.
24:16
If the robo taxi hits a mailbox, the company pays to replace the
24:19
mailbox. Yep, there is no long court
24:21
battle over software bugs or sensor failures.
24:24
The machine caused the physical damage.
24:27
The corporate owner pays the bill.
24:28
But strict liability severely discourages innovation if
24:31
companies are forced to pay for accidents they genuinely
24:34
couldn't avoid. How so?
24:36
Well, if a pedestrian literally jumps out from behind a parked
24:38
van directly in front of the robo taxi, and the physical laws
24:41
of momentum make stopping in time impossible, why should the
24:45
company pay millions in damages? Under strict liability they
24:48
would have. To right and that creates an
24:50
immense unpredictable financial burden on the companies trying
24:53
to develop the technology. The second alternative is the
24:56
reasonable computer standard. This involves judging the car
24:59
against other autonomous vehicles currently on the
25:02
market. Oh, so if your robot taxi
25:04
crashes into a concrete barrier, the court looks at the data to
25:07
see whether a competitor's robotaxi would have avoided that
25:10
exact same barrier? You are judging the software's
25:13
performance against the current industry standard for artificial
25:17
intelligence. Exactly.
25:19
And the third alternative is the reasonable human standard.
25:23
Under this framework, the manufacturer is liable for
25:26
damages only if a competent, attentive human driver would
25:30
have avoided the crash in the same circumstance.
25:32
Wait, let me get this straight. We are holding the multi billion
25:34
dollar supercomputer to the standard of a guy driving home
25:38
from work. That's it exactly.
25:39
You asked the jury a very simple question.
25:42
Would a normal, sober, awake human being have stopped in
25:45
time? If a human would have inevitably
25:47
crashed due to the physics of the situation, the computer is
25:50
not held liable, right? But if a human would have easily
25:54
seen the obstacle and avoided the crash, the manufacturer
25:57
pays. It uses a standard of care that
26:00
human juries already deeply understand and apply every day.
26:04
The new traffic citations we discussed earlier perfectly
26:06
bridge this gap. The official notice of non
26:09
compliance serves as immediate government proof of negligence.
26:12
Yes, it completely bypasses the need for a crash victim to audit
26:17
the proprietary code. If the state has already issued
26:20
a formal citation to the software for running a red
26:22
light, the victim in the resulting crash doesn't need to
26:25
hire a team of software engineers to prove the car was
26:28
defective. The government citation itself
26:30
proves the machine breached its basic duty of care to the
26:33
public. But judging A lightweight
26:35
passenger sedan running a stop sign is one thing.
26:38
The stakes are about to get exponentially heavier as the
26:41
vehicle weight classes change. Because the regulatory ceiling
26:45
on vehicle weight has recently been lifted, automated vehicles
26:48
weighing over £10,000, which includes heavy duty freight
26:51
trucks and massive commercial semis, are now legally permitted
26:54
to operate on public highways. We are moving from small
26:57
passenger cars navigating city streets to massive multi ton
27:02
logistical vehicles hurtling down the Interstate without a
27:06
human being inside the cab. The testing requirements for
27:09
these heavy duty systems are intensely rigorous compared to
27:12
passenger cars. A heavy duty system must log
27:15
half 1,000,000 miles of physical testing at each development
27:18
phase before it is approved for commercial deployment.
27:21
Compare that to just 50,000 miles for a standard passenger
27:24
car. The physics of stopping an
27:26
80,000 LB fully loaded freight truck are fundamentally
27:30
different from stopping a sedan. Yeah, and the potential for
27:33
catastrophic mass casualty damage requires an order of
27:36
magnitude more validation from the state.
27:38
This connects directly to the broader economic model driving
27:42
this entire industry forward. Target operating cost for
27:45
passenger robo taxes is around $0.20 per mile.
27:48
If travel becomes that incredibly cheap, the
27:51
fundamental design of our cities changes entirely.
27:54
We currently design massive portions of our urban spaces
27:57
around the assumption that humans need to store their heavy
27:59
cars in parking lots while they work or shop.
28:01
Which brings us to the concept of the zombie car.
28:04
The zombie car. Yeah, imagine you commute to a
28:07
busy city center every morning. You currently have to pay for a
28:10
parking garage, which might cost 30 or $40 a day, but if the
28:15
operating cost of the automated vehicle is only $0.20 a mile, it
28:19
is vastly cheaper for your emty vehicle to slowly circle the
28:23
block for three straight hours rather than pay for that
28:26
expensive parking garage. Think about the reality of that.
28:29
Imagine downtown Manhattan or Chicago, but instead of parked
28:33
cars quietly lining the streets, there are 10,000 empty vehicles
28:38
endlessly circling the blocks at 5 mph.
28:41
They're just burning cheap electricity, creating an endless
28:43
loop of traffic simply because it makes financial sense for the
28:46
user to keep the car moving. The.
28:48
Streets will be absolutely clogged with ghost cars.
28:50
We are talking about pure algorithmic gridlock.
28:54
This immediately opens up the necessity for cities to
28:56
implement aggressive congestion pricing, specifically targeting
29:00
empty autonomous vehicles. Municipalities will have to
29:03
charge a high fee to vehicles driving without passengers in
29:06
dense urban cores just to prevent total gridlock.
29:10
The economics of cheap driverless miles will force
29:13
cities to heavily tax the empty space inside the vehicle.
29:16
Yep, they have to. So the legal identity of the
29:19
driver has officially transferred from the human seat
29:22
to the corporate boardroom. This forces manufacturers to
29:25
design hardware that survives real time municipal enforcement
29:28
and absolute legal accountability.
29:32
The state is no longer looking for a plastic driver's license,
29:35
they are auditing the algorithm itself.
29:37
And if the cost of travel drops to almost nothing, can the
29:40
machines become statistically far safer than us?
29:43
Will human driving eventually be outlawed entirely just to lower
29:46
the insurance premiums for the automated fleets?
29:49
If you're not subscribed yet, take a second and hit follow on
29:52
whatever a you're using. It helps us keep making this.
29:54
We appreciate you being here. Also, check out our YouTube
29:57
channel for more business and tech updates.
29:59
There's a link in the description.