2022 number: 42,795
https://en.wikipedia.org/wiki/Motor_vehicle_fatality_rate_in_U.S._by_year
1,000
1.00What scenarios do bettors imagine would decrease it? Safety through new car models and FSD technology?
Average age of vehicle ownership is fairly high (12 years in 2022).
@parhizj once FSD clearly reduces deaths by say 99% the remaining deaths will start to feel real again. If a single state got serious about at least requiring emergency safety systems where the self driving takes over to protect you and others, there could be large state-state gaps. If one state has 2k annually while it's neighbor had like ten, it does look bad. Especially once kids start being raised in neighborhoods where the constant threat of cars hurting them is gone and we get used to that freedom again
That said on general you don't hear much about practical issue anymore, like alcoholism, gambling, accidents harming families and kids so mayve even with a solution to this available it might still just go on
@Ernie So I think a general model is then: manufacturer fleet demonstration (most narrowly in a model of a manufacturer's cars), followed by small scale market deployment (state scale deployment -- new car sales in California), followed by widespread compulsory deployment (1/3 of cars driven in states).
Consider just California as being the first to ban new car sales not equipped with such AI safety systems in their states (similar to emissions regulations).
It will take a few years for the majority of manufacturers just to include such systems in the design of vehicles (2-3 years on average for a car design). It should take at least this long after legislation before manufacturers can produce enough cars to meet demand in the state. Imagine that such safety technology is demonstrated by a manufacturer in it's fleet by the end of 2025 (deployed by a manufacturer fleet wide at the start of 2025), so California studies the data and makes it mandatory starting 2027 requiring new car sales employ this technology. Car makers even anticipate this in 2026, and all such new vehicles begins design/incorporation of such safety tech immediately and finishes design so that sales of all cars featuring this technology is commences at the start of 2029.
So the start of widespread adoption is likely to only commence then: 2 years (fleet demo + studying + legislation) + 2 years (manufacturer adoption) = 4 years from start of fleet scale (demonstration) deployment of such a technology to the earliest start of wide spread scale adoption (in this case all cars are made with it, but it could also be the case that this is not so).
Then it will take time for the current fleet of vehicles in a state to be replaced. In California ~1.8 million new car sales last year, https://www.factorywarrantylist.com/car-sales-by-state.html#google_vignette, but with 31.4 million registered vehicles https://www.statista.com/statistics/196010/total-number-of-registered-automobiles-in-the-us-by-state/. If you assume the number of registered vehicles and the number of car sales will stay constant (31.4 / 1.8), it will take ~17 years to replace all those vehicles.
Consider another case, an optimistic all-at-once scenario: To meet the requirements of this question we only need to reduce the number of fatalities by roughly 1/3: let's naively assume that: (a) 100% of new car sales in the US have such systems (with a federal mandate or equivalent adoption by car manufacturers), (b) manufacturing meeting the same sales rates as before, (c) using California vehicle sales / registered vehicles as a proxy for the country, (d) a design delay to majority market production of 3 years subsequent to a mandate, (e) and the number of vehicles registered does not decrease, for which we will use as a proxy for number of vehicles driven on the road in the same proportion (this is also quite a large assumption -- that given the option, people will not avoid driving less safe vehicles in their commutes). Working backwards, this scenario would be reached at earliest then 17/3 ~= 6 years after such adoption is compulsory, and for the number of cars to reach this number by the start of 2039 suggests that car sales starting in 2033 are all with such safety features. Furthermore, that the majority of car companies (levels of car manufacturing) have already reached this threshold (assuming 3 years again, this means they would have to start designing such vehicles ready by 2030.
There are manifold ways we can relax these constraints: for instance by staggering the adoption (California goes first for instance, and perhaps other states follow) so that universal adoption by every state is not even required or so that universal adoption can happen later than 2033 so in the end the required number of vehicles (1/3 of all driven vehicles) that are safer can be purchased later than 2033. We could also remove the design delay assumption by making another assumption that car manufacturers switch to the new safety tech and stop making the less safe/more affordable automobiles for various reasons such as.
These scenarios still presupposes the demonstration of the existence of such tech prior to widespread or compulsory adoption, meaning the underlying safety technology must be deployed in a demonstration in a large enough market 6 years + 2 years (design adoption across all manufacturers) + 1 year (legislation) + 1 year (fleet demo) = ~10 years total prior to the start of 2039, so (a statistically large enough sample) of these "safer" cars must be deployed by 2029 somewhere (although not necessarily in the US).
In your scenario where even an individual state goes first like California, we can further roughly guess that would delay the adoption by (at least) another year, such that the fleet deployment demo need start by 2028.
So do we think there will be a manufacturer's fleet wide demo deployed by 2028?
@parhizj ah, I had also imagined that something like comma_ai would be more believable in this world. It's a cheapish device which you plug into your cars control ports and connect to cameras and it just drives for you. They advertise it as level 2 but who knows how good it can get
https://www.reddit.com/r/Comma_ai/comments/nztkm5/will_comma_ai_invest_some_time_in_providing_legal/
So a scenario might be: people realize/see Tesla FSD is superhumanly safe (not the case yet but possible in 1-2 years say) because Tesla drivers change their behavior (sleeping during their commutes to work, no nag screen, starting to do overnight self driving trips, cars on road with no driver in the seat at all)
Once people notice that, and a state that cares about traffic fatalities a lot (eg Utah
https://www.udot.utah.gov/strategic-direction/zero-fatalities.html )
takes big steps like paying for this further, partnering with insurance companies
Add ten years for more devices to be sold and made to apply even to cars which are hard to handle, like old mechanical ones, and that's the best story I can think of.)
@parhizj the other possibility is that say, FSD is working enough to take over by 2028. Tesla makes tons selling it to other companies or others copy the tech. Then insurance or state governmentd start mandating such systems for bad drivers, people with DUI on their records, etc.
Then they license it as an addon improved version of comma_ai for say 7k as a back port for older cars - a good chip to run it, six cameras plus a connection to your car's controller system. At what point does this become mandatory enough that state or national govt, charities are willing to pay? or individuals get it? When does MADD get hooked up to DUI recipients and mod their cars to have mandatory supervision?
Id buy it if it can be made super defensive where even if other people are psycho it'll keep you safe. Id pay a lot for that. Stick one on my family members non Tesla cars just as a backup!
I agree 2040 is early but it's also crazy that we allow this insanity. See image - this guy drove a Ford f250 south, didn't notice the road curved and went head first into a van killing one, seriously injuring his wife, at 3pm in clear day. Simple lane keeping stuff like comma could have stopped this, let alone FSD. I bet there are configurations of his F250 or at least modern versions that might notice they were barreling into incoming traffic... At what point does the ntsb make them take it seriously? At least horns should have been blasting in this druggies ears, and at least auto slowing him down...
The guy just got life, clear as day he shouldn't have a license or be driving without mandatory supervision given his history
Anyway I don't have much hope we'll actually do anything about it, but a slightly smarter version of humanity would figure something out here
@Ernie After reading your comment, I looked up was the number of fatalities reported involving impaired drivers. https://www.cdc.gov/transportationsafety/impaired_driving/impaired-drv_factsheet.html
It's hard to derive a breakdown, but 30% of drivers in fatal crashes were drunk drivers; the statistics for opioids, marijuana, cell phone use are harder to figure out what is added on top of that (as some might be drunk AND using one of the other drugs or a cell phone).
The alcohol detection devices placed in cars (DADSS) doesn't seem to have a majority consumer support (40%), but as that article notes:
"The National Transportation Safety Board (NTSB) recommends that all new vehicles should be equipped with technology that prevents or limits the vehicle from operating if the driver is impaired by alcohol. The Bipartisan Infrastructure Law calls on the National Highway Traffic Safety Administration to issue a Federal Motor Vehicle Safety Standard requiring all new passenger vehicles to come equipped with advanced technology for the prevention of impaired driving." If this does get enacted and the compulsory rule doesn't get challenged, that would bring the total down by quite a bit (maybe ~ 20%?, as per https://nap.nationalacademies.org/read/24951/chapter/6#218 IIHS estimates that about 7,000 lives could have been saved in 2015 if DADSS devices set at a BAC limit of 0.08% had been in every vehicle (Cicchino, 2017)." -- 7000 lives of 35000 deaths is ~20%).
Right now the number of deaths is over 40,000 in the last few years. I think in the automation example I worked previously I used a number of about ~30000 and had an error in the calculation -- I assumed we'd need 1/3 cars replaced (that was just wrong as I was thinking to reduce by 10k fatalities not reach below), when we need actually 75% of cars replaced (with perfect machines) to reach the absolute number of 10000: so it would be 75% of 17 years for California which would be about 12-13 years instead of 6 years. This is on top of the 4 years or so after deployment of the fleet demo in the example. So it would be ~16 years from when we first see only cars with self driving safety tech to reach those numbers.
If the rule gets adopted for these alcohol detection systems, then we cut 20% off of those 40,000, that leaves a remaining 55%, so it will take then ~9 years (17 *.55) then plus 4 years so 13 years after the first fleet demo deployed. That would mean in this combined DADSS + (self driving) safety tech scenario we would have to see the start of the deployment of the self driving tech by 2039-13= 2026 (2 years from now). It seems not very plausible from just these two technological fixes alone in these rough sketches.
@parhizj agree. I think things which may change the calculus are:
If these devices are viewed as self defense
If they can be added to old cars rather than needing to completely replace the car
For #1 in the previous crash example, for example, smart careful people could outfit their own cars with defensive devices to protect them from drunks.
It's very strange to me that in the US the dialogue around musk is about meaningless Twitter conflict rather than on accurately measuring if FSD or similar, which he is by FAR the most outspoken about, can actually save all those lives.
Similarly, nobody talks about how SpaceX has given the USA and our military and reconnissaince departments a huge advantage in space, at least superficially. How much is our ability to destro the CCP's ability to use gps playing in our strategic competition with them?