That's not how people will see it.

Nuclear energy is the safest form of energy, it has much fewer casualties than gas, wind or solar. The difference is the risk of nuclear is indiscriminate, whereas people dying in a solar accident are the workers (rooftop fall).

If the driver is in control of their own car, they have themselves to blame (mostly). With self-driving, they are trusting their lives into a single entity they understand nothing about.

Nuclear energy is a good example of the situation that self-driving cars find themselves. A common saying in my industry is that with safety critical systems you do not get to decide whether the systems kill people or not, the best you can do is decide how many people get killed. Unfortunately for self-driving cars, like nuclear energy, even if the number is far less than existing technologies the public has little appetite for the indiscriminate death.

^This.

The census don’t, generally speaking, tend to believe statistics and facts. And the media will always spin a bad situation to boost their click through rates.

So, even though self driving cars may be 100% less prone to accidents (this is just an example), media headlines like “driver killed instantly when self driving car automatically steered into a wall without warning,” will cause the majority to push back and say they’re unsafe.

> “driver killed instantly when self driving car automatically steered into a wall without warning,”

Interesting example. A justifiable fear people have of self-driving cars is them crashing in scenarios where no (minimally competent, not ass-blasted drunk) human would reasonably crash, like randomly steering into a wall.

I think if self-driving cars were seen to crash in scenarios where a human could have made a similar mistake, their reputation would be better.

Not to say the media reporting of these incidents would be in any way accurate or fair.

This is why self-driving cars will feel more dangerous, even if statistically they are safer. I wouldn't much feel better if in the moments before my car drives into concrete barrier, after having determined impact is inevitable, it informs me that while it might have failed me and I will die, that statistically it has saved 10 other theoretical people.

If I had to guess, the decision will always be to protect the occupants of the car over everything else. No one is going to buy a car that will kill you to save others.

That's not what I was saying. That a car is statistically more safe is not comforting when it kills you in a way you would have easily avoided had you been driving.

there are plenty of sober people that randomly steer into walls/trees/you name it, every day

Nuclear power generation is also being used, you just don't hear about it much. You can see California's power supply by fuel here: http://www.caiso.com/TodaysOutlook/Pages/supply.aspx

Right now it's around 10%, but it's different for different regions.

I think the problem with nuclear energy is the overall number of people who die per incident. Because of this, the deaths are overreported.

We seem to be able to handle people dying one at a time, but once a whole town is in danger, even if this is something that happens once a generation (this is the hard part to understand), it crosses that threshold of "this isn't worth it".

That said, if self driving car deaths get reported as much as traditional car deaths (that is to say, not very often), perhaps people will think they are just as safe as driving themselves.

>I think the problem with nuclear energy is the overall number of people who die per incident.

IIRC those numbers are very, very, low. Your village should worry about the dam upstream, not the nuclear power station upwind.

> The moment it's an order of magnitude less, then self driving car will become default.

I have no surveys to back this, just intuition. But I don't think one order of magnitude is enough to counteract the fact that people don't feel as if they have control over their own mortality.

i.e. it'll have to be a lot better than just one order of magnitude before people can be swayed towards putting their lives in the hands of a robot.

That is not the only thing that matters. There is some interesting psychology at play here.

One thing I see rarely discussed is the fact you are getting killed by a programming error, not a human. Of course, the humans made the said programming error, but they have no face, there is nobody (individual) to blame, no-one to apologise, no-one to mourn for their actions.

At least if someone is hit by another driver, they will be prosecuted in the case of negligence, or if it's due to a mistake, be able to apologise. It's human. It's different.

I don't know. People still fly despite the occasional mechanical error, programming error, or other non-individual error leading to fatalities.

I think if it is an order of magnitude safer that would override many of those emotional concerns. Insurance companies are decent at thinking rationally about problems. People might have emotional concerns, but I think a halved auto insurance bill would be enough to sway a critical majority of people.

But when you fly there is still a human face ‘in control’ (even if not to a sufficient degree) you can blame: the pilot.

That's a good point. Pilots are still an important psychological factor that makes it easier for people to fly.

> Of course, the humans made the said programming error

Are you sure?

You do understand that self-driving vehicles inherently rely upon machine learning systems (and I am not talking simple regression here), which are trained and ultimately make decisions by means that we still don't fully understand?

What I'm trying to get it here is that, barring faulty sensors or mechanical systems, any decision that the control system of a self-driving vehicle makes is most likely arrived at via the inner workings of one or more machine learning systems (likely deep learning neural networks of some nature, but there are other systems being experimented with as well).

They are not a huge series of if-then-else statements or anything of that nature; such algorithmic approaches have been tried in the past and dismissed as unworkable due to the sheer complexity that is inherent in driving.

Honestly, what drove the field forward was mainly work done by CMU, mainly ALVINN; but there was interplay between CMU, Hans Moravec, Stanford (the Cart, among others), Thrun, and others. It has a very long, convoluted, but detailed and rich history (I encourage anyone who's interested in the fields of robotics, ai/ml, and/or self-driving vehicles to read up on it - it's very fascinating). ALVINN ultimately gave the hint toward using neural networks and deep-learning, but the tech couldn't make the leap forward until the convergence of a number of technologies.

I know this. You know this. But we are in a technology bubble. Machine learning also has no face to the regular public. Telling to the public that AI killed their relative is even worse than saying the team at Google did it.

You've articulated the idea underlying a nascent field of research: algorithmic accountability.

> The moment it's an order of magnitude less, then self driving car will become default.

This is rational but I don't think it's true. ~37,000 people die from car crashes every year in the US, from about ~35,000 accidents. Each individual Tesla crash that results in any serious injury or death is still a national news story. Humans are bad at estimating probability; it's hard to imagine popular support for self-driving cars if they kill thousands of people each year. Even if the alternative is tens of thousands of deaths, we have already collectively priced that in as a risk of living in a developed, automobile-driven society, such that fatal crashes never rise beyond local news.

A third of fatal accidents involve drunk drivers.

I find it interesting that the apparent objective is to make driving safer, but apparently the powers that be have decided that it is easier to make cars that drive themselves than to convince people not to drive while drunk.

Quibbling... 28% in 2016 (https://crashstats.nhtsa.dot.gov/Api/Public/ViewPublication/...), but this is after decades of awareness campaigns and enforcement. Then there's the ~70% of fatal accidents that happen with sober drivers. There might not be more to be had by addressing human factors (though I personally believe graduated licenses with more rigorous testing and much higher investments in public transit would help) so the focus on self-driving vehicles may not be misguided.

You're off by some number of orders of magnitude on the accidents, unless you mean, "There are ~35,000 accidents that involve one or more fatalities (and many more that do not)."

> ~37,000 people die from car crashes every year in the US, from about ~35,000 accident

OP reads correctly to me

Even if overall self-driving cars are safer, they won't feel safer to the individual if the situations in which it will kill them are situations they could avoid if they were driving themselves.

The self-driving car might be able to avoid that t-bone situation more often than the driver can, but if the death scenario for the self-driving car is it gets confused and drives into a river, that will feel unacceptable to a human because it seems so simple to avoid to them.

Most people aren't concerned how many other people survive, but are generally primarily interested in their own survival over population-level statistics.

> The only thing that matters is how many people are dying because of self-driving cars vs dying without.

This is not how human emotion or layperson risk assessment works.

I bet it'll be more about money than lives. As soon as car insurance is an order of magnitude more expensive when "driving yourself", then we'll have self-driving cars as default!

It's not like that's going to happen next year or ten years from now.

Insurers like boring, happens at a consistent rate, human error accidents so long as their slice of the population is not making them at an appreciably different rate than the competitors' slices.

Insurers really don't like hard to predict accidents that cause people to max out their collision and medical (which is exactly what would have happened in the Tesla cases had the customer not died) which is exactly the kinds of accidents self driving cars seem to be having.