Questions I Have About Autonomous Vehicles
With yesterday’s death of a pedestrian by one of Uber’s self-driving cars, I’m writing this post. I’ve been thinking about these issues for a while but ended up putting the post up quickly. I’ll probably edit in the near future as I learn more.
When it comes to upcoming big changes to a big part of our lives, I naturally have big questions. This is a list of questions related to autonomous vehicles. If there are answers then I hope you can help me answer them. Note: I’m not accepting “positions” or “opinions” as answers for these questions. If answers exist today, then there should also be an understanding of why it’s so and who gains or loses because of it.
Also note: Many of these questions are about second-order effects that could result from such big changes. I haven’t seen much thoughtful writing about this yet. I’m going to take the claim that it shows a lack of thought to simply say that autonomous vehicles will automatically always be safer. If you think some of my questions are extreme edge cases, well, edge cases turn out to be more common than we think. Also, when self-driving vehicles operate at scale edge cases can be big issues. Without a measure of the risks that will be generated as things also get better most of the time, we expose ourselves to shocks when the new system breaks down. I’ll explain why below.
Here are my questions so far…
1. Yes, the pedestrian who was killed (mentioned above) was apparently “jaywalking,” which is prohibited in Tempe, Arizona and many other places. That jaywalking is prohibited still comes as a shock to me. Having spent so much time living and working in New York City, crossing at odd points within the block is a common occurrence. I believe that the city would stop working if all pedestrians had to wait at the crosswalk for the light to change. That’s no reason to push that model everywhere, but I generally believe that cities are best built as places for people, not cars. How much of the responsibility should be put on the pedestrian and how much on the autonomous vehicle company? I believe that the model for human driven cars is not necessarily a guide here.
2. How does the human reaction to liability change with self-driving cars? As in, if one casualty is caused by an autonomous vehicle (AV), what is multiple of liability from one casualty caused by a human-driven car? I believe that this multiple is large, at least today.
3. In the US, road fatalities killed over 37,000 people in 2016. That’s a little over 3,000 per month, or 100 a day. This great chart also describes the change in the US as more cars came on the road. If AVs work as they are described and there is widespread adoption, then those fatalities could indeed be driven down to something close to zero. But nothing ever works perfectly. What happens with cybersecurity breaches?
The above chart is where the US is today. Around 3,000 fatalities per month. For the last 40 years there have been more years of fatality reduction than increase (multiple reasons).
If all goes smoothly with widespread adoption (or even a mandate) for AVs, the red line in this graph could be shifted down to something close to zero, as below.
That’s great. Now what could happen if a single large cybersecurity breach takes place (intentionally not trying to estimate the upper bound).
That’s with a quick resolution. We’ll get to the financial and legal impact later. But for now, let’s just consider how a hacked national fleet, or national grid of AVs could possibly impact road fatalities. Here’s another example.
That lingering impact could be generated by anything from lost trust in AVs, leading to people bringing their human-driven vehicles back (even if illegal), inexperienced people driving cars (more accidents), sabotage of autonomous cars, or lingering effects of the hack that may or may not be not apparent, such as infrastructure damage created by the hack itself.
Here’s one more.
In this example, it takes a while for the hack to be resolved. Travel may decrease temporarily, but it is more dangerous than it was previously.
4. Life savings as a percentage of overall deaths is different depending on where you are in the world. The World Health Organization organizes the Top 10 causes of death globally and by country income level. While road deaths aren’t on the top 10 for high-income countries, these deaths are number 10 for both low-income and lower-middle income economies. That means that the last places likely to see autonomous cars (the poorer countries) are also the ones most likely to benefit from the decreased road deaths (assuming that road deaths indeed do decrease). The top of the list for Road fatalities per 100,000 inhabitants per year is Libya, Thailand, Malawi, Liberia, and Democratic Republic of the Congo. The top of the list for Road fatalities per 100,000 motor vehicles is Guinea, Benin, Democratic Republic of the Congo, Sao Tome and Principe, and Ethiopia. I’m not counting data for Road fatalities per 1 billion vehicle-km since there is data for only around 20 countries.
5. Interesting, the global average (2002 data) for road deaths per 100,000 people is 4x higher for men than it is for women. Does anything change about this with driverless cars? Remember that when Uber driver data was analyzed it revealed that women are paid 7% less than men (for factors mostly related to length of time as drivers, time and location of work, and speed).
6. What about hacks that cause no loss of life but have a different type of cost? For example, a hack that caused less fuel efficiency (less efficient routing, driving speed, or other actions) for a national fleet or network may not be noticeable to passengers. However, this could have a big impact nationally. Like bank robbery, it is not worth it for the autonomous car company to disclose it.
7. What about hacking that does not cause loss of life, but diminishes the value or trust of AVs? E.g. driving someone to the wrong location? Eavesdropping or other activities?
8. Hacking, whether by outside criminal forces or internal company forces where people are driven somewhere they don’t want to go, or where people are intentionally targeted for surveillance for voyeurism? If that targeting already happens within video surveillance systems, how do we prevent it in AVs?
9. Even if self-driving cars are less susceptible to hacking than vehicles driven by humans, remember that the value of hacking a fleet, company, brand, etc is much higher than hacking individuals. Large entities will pay to get control back or keep things quiet. Individuals much less so, or not at all. How does that change the intent of groups wanting to break into AV networks or specific vehicles?
10. Are there some issues where the autonomous vehicle will refuse to accept passengers, some where it will accept but refuse to drive, some where it will drive but impose a fine? For example, while a human driven car (rideshare or taxi) may refuse to serve a drunk, dangerous, or offensive person, what about the AV (autonomous vehicle)? Why not serve them but impose a fine, whether monetary or to reputation/ranking, which could impact future pricing and rides.
11. Business model. Today, there are automotive transportation companies that require human drivers to own or rent the vehicles (most taxi companies and rideshare companies today). One of the future models is for fleet operators (could be Uber, Google, or “even” Ford) to offer services using AVs. How does that change the business models for these driverless car companies? Today, it is typically the drivers who cover vehicle costs, including purchase or lease, regulatory licensing, fuel, maintenance and cleaning. That’s a large upfront cost of buying the vehicle plus the ongoing operational costs. In a transition to AVs providing transportation services, the service provider must cover those costs. How much does that change the presumed savings and the pricing extended to customers?
12. Savings: human labor, extra fuel cost of carrying human driver, extra cost of non-essential car parts (driver’s seat, trunk), ability to run more efficient vehicles by changing the fleet more frequently (more usage in shorter time periods), extra waiting time required since human drivers must see and accept potential customer rides, inefficiencies related to human drivers (need to take food, bathroom, and rest breaks).
13. What happens during shocks? For example, oil (if AVs are still gas powered), or shocks impacting electric vehicles, for example cost increases caused by access to components of battery technology or new costs put on machinery using batteries to account for future disposal and pollution caused?
14. Cleanliness: When many people ride in a car during the day, by the time you get into the vehicle, how clean is it? Problems include both refuse left behind by previous passengers, but also smells… Will there be businesses established just to clean the inside of autonomous vehicles? Or, we could imagine a coating on the inside of the vehicle which can be peeled off and disposed of when needed…
15. What about surveilling riders themselves for no reason? Is there anything prohibiting open and intentional surveillance of passengers, done with the claim of public safety? What about security back-doors for government access to monitor movements of individuals?
16. What about passive surveillance for the purpose of data collection. E.g. compiling a list of conversation topics to improve customer satisfaction or develop new products?
17. Can children ride in an AV by themselves? What about babies? Before you say “no way!,” what’s your reasoning? What if the vehicle included child care options?
18. Could people just put their kids in an AV when they need to get some work done? Baby-sitter on the go when none is available.
19. Will an AV prevent an unsafe number of people from squeezing inside? Who decides what is unsafe? Or will this be handled by pushing a fine to the passengers, rather than preventing them from entering?
20. Will an AV prevent someone from unbuckling their seat belt? Will the car refuse to drive or pull over if the seat belt is unbuckled? If so, what about unbuckling the seat belt when needing to take off a jacket? Does a small adjustment like that trigger a larger response? What kind of AI is needed to understand the associated human intent?
21. Will an AV prevent someone from smoking (if that’s prohibited)? Or just generate a fine for the passenger?
22. Do driverless cars lead to the development of something called the “60-mile per hour club”? I’ll leave you to draw your own conclusions of what that is (hint: think of the “mile-high club”). If so, what do AV companies do about that, if anything?
23. What human vs machine competitions will self-driving cars spawn? For example, competitions where humans try to maneuver AVs (think Kasparov vs Deep Blue, or even John Henry vs the machine). I assume the AVs are less interested in the competitions than the humans.
24. Who owns your travel data? Who can obtain the data and under what conditions? What compensation is made in the event of a data breach?
25. What other population data of interest could be collected? For example, animal populations seen while passing, insect populations (hitting the windscreen etc), roadkill species identification. A lot could be done to automate these counts.
26. There are road conditions which are dangerous at outset — severe storms for example. But could passengers pay a premium to travel during incredibly dangerous conditions? Why or why not?
27. How does parking change when most or all cars are AVs? On the one hand, passengers can have their cars (in this case assuming ownership and not rideshare) drive to a parking location and return later. On the other hand, what impact does this bring to traffic and parking availability? Do human-driven cars take priority when it comes to parking?
28. Will many vehicles have no one in them? The reasons could include that they are traveling elsewhere where there is demand, that the vehicle is not primarily used for human transport but instead is a traveling store, entertainment center, or service provider (for example, cleaning services as described above).
29. If it is cheaper to hire an AV truck to move your items when moving houses, how does that impact moving services and truck rentals?
30. Could you sleep in the car overnight for purposes of long-distance travel? Could the AVs include a shower, bathroom, and proper bed?
31. What about including a gym, cafe, restaurant, office, or other specialized spaces? How does AV development impact these businesses?
32. How do AV companies deal with antisocial behavior from humans? For example, humans who harass AVs either as pedestrians or while driving their non-AV vehicles.
33. Should weight be a factor in pricing? E.g. heavier people, or people bringing luggage charged a premium? Could an extra charge to people above a certain body mass index be used in pricing — basically taxing individuals who don’t fit current standards of healthiness?
34. Could you ride while high? How does the ability to monitor likelihood of substance abuse factor in to legal action?
35. Could you ride naked? Is the AV a public space or a private space?
36. Could you pay a premium to have your identity and the identity of your co-passengers kept secret? This question somehow reminded me of this Monty Python skit.
37. What type of advertisements are passengers exposed to? In many taxis, there is some type of ads, from visual print ads to videos that play with sound. Can you opt in or out? Is there a cost or benefit associated with that choice?
38. How do passenger travel patterns change with AVs? Does anyone ride mass transit anymore?
39. What types of entertainment grow or decline? Does video grow in relation to audio (which is the only entertainment you can get if you’re driving). So, does anyone do podcasts anymore if most travel is via AV (that is, there are only passengers and no drivers)?
40. What other types of businesses change or stop working? Billboards, rest stops, fueling stations, parking lots, retail are all up for big changes.
41. Do AV ride completion success rates (no accidents) vary by location because of differences in locale road and weather conditions, including differing human attitudes toward to vehicles, their passengers, variations in comfort with ambiguity globally…?
42, Tesla now has a second fatality (occurred after I originally wrote this list). How far in the other direction will acceptance of AVs swing now?
43. New addition after this post by Jason Calacanis on April 6, 2018. Writing that “[t]he most disturbing and frustrating trend, in all four deaths, is that the human drivers played a significant role in them,” misses the point. It will be easier to expect that people will continue to act as people when building autonomous vehicles. He further wrote “[w]e won’t know for a while, but there is a chance that if the driver — who was being paid to drive the car — had not been blatantly and knowingly breaking the law, they might have been able to apply the brakes in time.” It’s as though we should also blame the killed pedestrian for being so inconvenient as to allow herself to be hit. Edge cases are plentiful in driving. I believe that thinking on this issue will evolve over time.
44. Should a small fee per driverless mile be set aside in a fund to pay for future costs related to the potential system risks? Otherwise, who pays if the system breaks down?
45. Will AVs not even bother to park? Or will they only park when the economics work out? See this tweet about how AV demand for parking may be zero, depending on cost to park and AV operating cost.
46. Are AVs exposed to the potential to issues with GPS network?
47. Along with operational effectiveness, AVs bring systemic risk.
I will add to this list, especially as I receive feedback and learn more. Thanks!