I read the post, the comments and the gist of why you made this post. It got me curious to learn more about this situation you presented. Let me "copy and paste" my AI research, if you dont mind.
i just want to put this out here for your consideration. I cannot promise to reply back or have further discussion on the topic. I dont have time for that.
I do not have programming of robots background, but I am not behind in AI developments. I understand (conceptually) how machine learning works, how models are trained, computer vision,etc. I don't claim to be super knowledgeable in this field, but I know enough how to ask questions to satisfy my curiosity to find answers.
____
While the author's physics-based approach offers a pragmatic escape from endless philosophical debates, automotive engineers, legal experts, and safety analysts point out several massive real-world flaws and counterarguments to this framework.
The primary challenges fall into four main categories:
1. The "Empty Property" Blind Spot
The author suggests the AV should aggressively route itself into "unoccupied property," such as parked cars or static infrastructure. However, a machine cannot always verify with 100% certainty that property is empty.
The Risk: A parked car might have a sleeping infant in the back seat, a parent adjusting a car seat, or someone looking under the hood.
The Consequence: By choosing to smash into a parked vehicle to avoid a pedestrian, the AV might accidentally kill a hidden occupant, creating a new ethical nightmare rather than solving one.
2. Legal Liability and Owner Pushback
Consumer psychology and the legal system pose massive hurdles to an "aggressively self-sacrificing" vehicle.
The "Suicide Vehicle" Problem: Would you buy a car knowing that its code is explicitly programmed to violently scrape a concrete wall, trigger a rollover in a ditch, or total itself because a pedestrian stepped out without looking?
Liability Shift: If a human driver swerves into a wall to miss a dog, it is viewed as a split-second reaction. If an AV manufacturer programs the car to intentionally destroy itself and injure its own passenger to save someone else, the manufacturer faces immense liability lawsuits from the passenger.
3. Cascade Failures and Secondary Collisions
Treating the environment as an "active tool for survival" assumes that crashing into infrastructure happens in a vacuum. In reality, physics models are notoriously unpredictable during chaotic, multi-car incidents.
Debris Fields: Scraping a guardrail at 45 mph or hitting a concrete pillar can tear parts off the AV. This flying debris can blind or injure other nearby drivers, cyclists, or pedestrians who weren't originally part of the hazard.
Rebound Hazards: Intentionally inducing a controlled spin or hitting a parked car can cause the AV to bounce back into moving traffic, triggering a secondary, much larger pile-up.
4. Overestimating Sensor and Physics Capabilities
The technical framework heavily relies on the AV calculating hyper-precise variables in milliseconds, such as roadside mud depth, surface slickness, or the exact structural integrity of a barrier.
Sensor Limitations: Rain, snow, or nighttime conditions heavily degrade laser (LiDAR) and camera sensors. An AV cannot easily distinguish between a "soft mud patch" that will bog the car down and a hidden concrete block covered in weeds, or deep mud that will trip the tires and cause a violent rollover.
Real-Time Calculation Latency: Calculating the fluid dynamics of a controlled slide on a wet road in the span of 200 milliseconds is computationally staggering. The processing power required to solve these chaotic physics equations in real-time might actually introduce the exact "computational latency" the author is trying to avoid.
The Ultimate Irony: By trying to escape the Trolley Problem, a physics-based system inherently creates a new one. The moment the AI decides to sacrifice the vehicle's frame (and potentially harm the passenger inside) to reduce total impact forces for people outside, it has still made a calculated, programmatic choice about who takes the brunt of the kinetic energy.
i just want to put this out here for your consideration. I cannot promise to reply back or have further discussion on the topic. I dont have time for that.
I do not have programming of robots background, but I am not behind in AI developments. I understand (conceptually) how machine learning works, how models are trained, computer vision,etc. I don't claim to be super knowledgeable in this field, but I know enough how to ask questions to satisfy my curiosity to find answers.
____
While the author's physics-based approach offers a pragmatic escape from endless philosophical debates, automotive engineers, legal experts, and safety analysts point out several massive real-world flaws and counterarguments to this framework.
The primary challenges fall into four main categories:
1. The "Empty Property" Blind Spot
The author suggests the AV should aggressively route itself into "unoccupied property," such as parked cars or static infrastructure. However, a machine cannot always verify with 100% certainty that property is empty.
The Risk: A parked car might have a sleeping infant in the back seat, a parent adjusting a car seat, or someone looking under the hood.
The Consequence: By choosing to smash into a parked vehicle to avoid a pedestrian, the AV might accidentally kill a hidden occupant, creating a new ethical nightmare rather than solving one.
2. Legal Liability and Owner Pushback
Consumer psychology and the legal system pose massive hurdles to an "aggressively self-sacrificing" vehicle.
The "Suicide Vehicle" Problem: Would you buy a car knowing that its code is explicitly programmed to violently scrape a concrete wall, trigger a rollover in a ditch, or total itself because a pedestrian stepped out without looking?
Liability Shift: If a human driver swerves into a wall to miss a dog, it is viewed as a split-second reaction. If an AV manufacturer programs the car to intentionally destroy itself and injure its own passenger to save someone else, the manufacturer faces immense liability lawsuits from the passenger.
3. Cascade Failures and Secondary Collisions
Treating the environment as an "active tool for survival" assumes that crashing into infrastructure happens in a vacuum. In reality, physics models are notoriously unpredictable during chaotic, multi-car incidents.
Debris Fields: Scraping a guardrail at 45 mph or hitting a concrete pillar can tear parts off the AV. This flying debris can blind or injure other nearby drivers, cyclists, or pedestrians who weren't originally part of the hazard.
Rebound Hazards: Intentionally inducing a controlled spin or hitting a parked car can cause the AV to bounce back into moving traffic, triggering a secondary, much larger pile-up.
4. Overestimating Sensor and Physics Capabilities
The technical framework heavily relies on the AV calculating hyper-precise variables in milliseconds, such as roadside mud depth, surface slickness, or the exact structural integrity of a barrier.
Sensor Limitations: Rain, snow, or nighttime conditions heavily degrade laser (LiDAR) and camera sensors. An AV cannot easily distinguish between a "soft mud patch" that will bog the car down and a hidden concrete block covered in weeds, or deep mud that will trip the tires and cause a violent rollover.
Real-Time Calculation Latency: Calculating the fluid dynamics of a controlled slide on a wet road in the span of 200 milliseconds is computationally staggering. The processing power required to solve these chaotic physics equations in real-time might actually introduce the exact "computational latency" the author is trying to avoid.
The Ultimate Irony: By trying to escape the Trolley Problem, a physics-based system inherently creates a new one. The moment the AI decides to sacrifice the vehicle's frame (and potentially harm the passenger inside) to reduce total impact forces for people outside, it has still made a calculated, programmatic choice about who takes the brunt of the kinetic energy.
Vampyre · 51-55
@CookieCrumbs I really like this response.
The article on a suicide vehicle is the one that got me going... I like that you brought this back. I also appreciate the other arguments as well. If my thesis passes, it will take time to get implemented, and by that time I expect all AVs to be equipped with thermal vision as well. As far as an infant sleeping in the back (super illegal), the police vehicles will be scanning for this, as well, however with different interest... The AV will be able to determine all life forces around it, and being a microprocessor driven "vision" it will input everything as it happens and constantly run safety scenarios.
The idea is not to save non-driving people over passengers; rather to minimalise injury to all humans. Inclusive and primarily passengers, then pedestrians on the outside. It is intended as an emergency velocity reduction.
If the vehicle is in auto drive mode, it will be travelling at the speed limit. Speed limits are designed to be a safe rate of speed for the environmental variables. If the impact with human life is truly inevitable, our desire is to preserve ALL human life. Regardless of wealth, age, race, or any other demographic "qualifiers". The passengers of any AV will be priority for all safety protocols. It is not intended to be used to come to a complete stop and look both ways.... When all sensors give more than 99% probability of impact with human life (not animals).
If the impact with a deer will only cause vehicle damage, then the AV will behave accordingly and come to a safe stop with all authorities having been notified.
Preservation of human life is focused on the passengers first and peripheral outside humans secondary.
Also if this emergency action must be taken and utilized in reality, it should only minimalize the injury to prevent death. Injury is resultant from miss-use and other factors.
Injury is expected. Death must be prevented. If the AV calculates no resultant human death from a certain course of action, then it should follow that course. Safely bring the vehicle to a halt and report all injuries.
We may need an extra line of code that covers how it is to never cause harm to any human not imminently involved (pedestrians, kids chasing a ball, a cyclist falling over, inhabitants of other vehicles or establishments...). Upon finalization there will be thousands of named variables.
I will add your data into my thesis and expound upon it.
Thank you for the excellent feedback!
The article on a suicide vehicle is the one that got me going... I like that you brought this back. I also appreciate the other arguments as well. If my thesis passes, it will take time to get implemented, and by that time I expect all AVs to be equipped with thermal vision as well. As far as an infant sleeping in the back (super illegal), the police vehicles will be scanning for this, as well, however with different interest... The AV will be able to determine all life forces around it, and being a microprocessor driven "vision" it will input everything as it happens and constantly run safety scenarios.
The idea is not to save non-driving people over passengers; rather to minimalise injury to all humans. Inclusive and primarily passengers, then pedestrians on the outside. It is intended as an emergency velocity reduction.
If the vehicle is in auto drive mode, it will be travelling at the speed limit. Speed limits are designed to be a safe rate of speed for the environmental variables. If the impact with human life is truly inevitable, our desire is to preserve ALL human life. Regardless of wealth, age, race, or any other demographic "qualifiers". The passengers of any AV will be priority for all safety protocols. It is not intended to be used to come to a complete stop and look both ways.... When all sensors give more than 99% probability of impact with human life (not animals).
If the impact with a deer will only cause vehicle damage, then the AV will behave accordingly and come to a safe stop with all authorities having been notified.
Preservation of human life is focused on the passengers first and peripheral outside humans secondary.
Also if this emergency action must be taken and utilized in reality, it should only minimalize the injury to prevent death. Injury is resultant from miss-use and other factors.
Injury is expected. Death must be prevented. If the AV calculates no resultant human death from a certain course of action, then it should follow that course. Safely bring the vehicle to a halt and report all injuries.
We may need an extra line of code that covers how it is to never cause harm to any human not imminently involved (pedestrians, kids chasing a ball, a cyclist falling over, inhabitants of other vehicles or establishments...). Upon finalization there will be thousands of named variables.
I will add your data into my thesis and expound upon it.
Thank you for the excellent feedback!
DeWayfarer · 61-69, M
Wrong approach. And if you knew anything about programming, you would say so there as well.
There is several little used thought processes use in programming AI that could be better utilized.
Heck even AI has recommended them. The problem there is they are afraid to do so. Which is a truly debatable topic.
Do we, listen to them, and let them out grow us or not? 🤷🏻♂
BTW your thinking is cyclical. That is where you are going wrong. Of course you can't make endless tests in a cyclical process. Dump the cycle. Make it based on events. Events don't happen every cycle.
There is several little used thought processes use in programming AI that could be better utilized.
Heck even AI has recommended them. The problem there is they are afraid to do so. Which is a truly debatable topic.
Do we, listen to them, and let them out grow us or not? 🤷🏻♂
BTW your thinking is cyclical. That is where you are going wrong. Of course you can't make endless tests in a cyclical process. Dump the cycle. Make it based on events. Events don't happen every cycle.
View 25 more replies »
DeWayfarer · 61-69, M
@Northwest So you are referring to the post and not this guy's statement! 🤣
Not even your own statement.
Not even your own statement.
eMortal · M
All this will need 10x more compute than what is currently used in AVs, better proximity sensors and state of the art computer vision models.
All this will need 10x more compute than what is currently used in AVs, better proximity sensors and state of the art computer vision models.
DeWayfarer · 61-69, M
@Northwest All that matters to this threads conversation is the initial comment!
Not the post, not other threads, not your own comment!
That is all I am talking about. Anything else is irrelevant to this conversation. Because it just hasn't been mentioned until you just interjected that AI submitted question.
Not the post, not other threads, not your own comment!
That is all I am talking about. Anything else is irrelevant to this conversation. Because it just hasn't been mentioned until you just interjected that AI submitted question.
helenoftroy2000 · 22-25, F
Vampyre · 51-55
@helenoftroy2000 yes it is, however if the vehicle can orchestrate a path of heavy resistance, (even 35 mph will kill), and inadvertently cause it's own destruction, then the unavoidable impact can be lessened to a minor injury over hitting anyone straight on.
It should not be programmed to save only one person... But to protect every human.
It should not be programmed to save only one person... But to protect every human.
Northwest · M
@helenoftroy2000 Looks like we're dealing with a bot here.
helenoftroy2000 · 22-25, F
@Northwest this site can't afford bots
Vampyre · 51-55
@Northwest I was recently asked a very specific question that led me to this dissertation.
If an autonomous vehicle determines an imenent danger without any possibility to avoid crashing into a person, with 3 people of varios ages, races and wealth, jumping immediately into the path... Which should the vehicle sacrifice?
If an autonomous vehicle determines an imenent danger without any possibility to avoid crashing into a person, with 3 people of varios ages, races and wealth, jumping immediately into the path... Which should the vehicle sacrifice?
Northwest · M
@Vampyre
And amazingly, you decided that the 50,000 word salad presented the problem more succinctly?
If an autonomous vehicle determines an imenent danger without any possibility to avoid crashing into a person, with 3 people of varios ages, races and wealth, jumping immediately into the path... Which should the vehicle sacrifice?
And amazingly, you decided that the 50,000 word salad presented the problem more succinctly?
Vampyre · 51-55
Actually I am saying... In a nutshell...
The av should be equipped with knowledge of all reported, recorded and recovered accident crash sites. The basic autonomous maneuvers should include a vast array of police defensive driving from around the world,
and the question asked whom should the av kill of the 3. My response is that you don't allow exception in human life. Program it to react in a calculated slide or tailspin
..
They have had all this information for years. Put it to good use and create an av that will not kill another person.
The av should be equipped with knowledge of all reported, recorded and recovered accident crash sites. The basic autonomous maneuvers should include a vast array of police defensive driving from around the world,
and the question asked whom should the av kill of the 3. My response is that you don't allow exception in human life. Program it to react in a calculated slide or tailspin
..
They have had all this information for years. Put it to good use and create an av that will not kill another person.