Hacker News Reader

Comment by crazygringo | Hacker News Reader

> A model of power consumption by autonomous vehicles predicts that once widely adopted, they will consume as much power worldwide as data centers do.

It's referring to the energy to power the brains of self-driving, not to move the vehicle.

But that seems... fine? I can't tell if the article wants us to think that's a lot.

Data centers and networks account for something like 1% of global electricity use.

If autonomous driving saves a ton of lives then an extra 1% is well worth it. Just from productivity increases alone it may also be well worth it. But also, the "brains" will presumably become more energy-efficient over time (as electronics have generally), and the share of renewables is increasing.

I don't really see a problem here.

View on HN

> If autonomous driving saves a ton of lives then an extra 1% is well worth it

Your total scope is about 40,000 road deaths. Off the top 1/6 of those are pedestrians most often hit at night. Another 1/6 is motorcycles. A little more than 1/2 are those who are drunk or otherwise impaired. Around a 1/3 involve someone under the age of 24 and male.

People presume that driving is exceptionally dangerous. If you actually examine the data you realize it's one of the safest things we do, given the _trillions_ of miles that are driven every year in the US alone.

From this perspective, it's not at all clear that greater levels of automation are going to improve anything. From the history of introduction into airlines, it will initially make travel _less safe_ until drivers are retrained to use the new systems in a way that doesn't enhance risk. These will often be non-intuitive lessons.

If you put a human in a machine also designed by a human and use it around other humans, you will never eliminate human factors.

I don't know where you're getting 40,000 road deaths from, because:

> According to the World Health Organization (WHO), road traffic injuries caused an estimated 1.35 million deaths worldwide in 2016. That is, one person is killed every 26 seconds on average. [1]

I also don't know what the breakdown you give has to do with. If we can save the lives of pedestrians, drunk people, and males under 24, then amazing.

Also, a quick Google search reveals that in the US:

> Deaths and injuries resulting from motor vehicle crashes are the leading cause of death for persons of every age from 3 through 33 (based on 2003 data).

That sounds pretty dangerous to me if it's the leading cause of death. Maybe not that dangerous per-mile, but clearly those miles add up.

[1] https://en.wikipedia.org/wiki/List_of_countries_by_traffic-r...

> I don't know where you're getting 40,000 road deaths from, because:

US road fatalities. This is because NHTSA catalogs every one of them and the circumstances leading up to them. This product is called the FARS.

> If we can save the lives of pedestrians, drunk people, and males under 24, then amazing.

You're assuming they'll use the automation. Or you're presuming it will be forced on people. Which ignored motorcycles. Or bicycles. People use those.

> Deaths and injuries resulting from motor vehicle crashes are the leading cause of death for persons of every age from 3 through 33 (based on 2003 data).

I'd like to know where you got that statistic from. In the US, it is not likely to be true. Accidental self inflicted injury, a separate category, is often the most common until Heart Disease and Cancer become factors.

The reason to use the US is because it is a developed country, a wide variety of different states and road conditions, and it has a single reporting agency covering it. Presuming that automated vehicles are going to lead to major improvements outside of developed nations is a bit unfounded.

But even saving up to 40,000 lives for 1% increase in electricity demand sounds like a good deal?

Depends on how that electricity is being generated. If it's clean then it might be a good trade-off. If it's dirty then we're probably worse-off.

True. And now we contrast this with the additional pollution deaths the increased energy and resource consumption will lead to. 1.3 million people isn't very many.

Currently we're in a paradoxical situation where there are two very contrasting mantras in society: 1. We need to reduce our carbon emissions, cut down on resource consumption and move to only producing energy from renewable sources. And 2. We should have more automation, more AI, more data, more computation in general.

Both directly contradict each other, anyone who's looked into the state of batteries and renewable energy knows we simply don't have a way for both to be true. In fact, we drastically have to reduce energy consumption if we want to make that green dream work, there's no practical way as of today to store all the solar and wind energy that would be needed for it.

I was ready to go do the math and prove you wrong, but you're right!

Some random infographic I found suggests that the mortality associated with driving of all road vehicles is about 1.14 Fatalities per 100 Million Vehicle Miles Traveled. If we say that the average driver is mostly urban and mostly arterial driving, with a health amount on highway on a per hour basis that puts at us at about 1.14 fatalities per (100,000,000 * 1 hour / 45 miles ) 2 million hours.

I tried to find other routine activities with any element of risk that were safer than that and had trouble doing so.

Airline travel is safer - but the amount of "safety stuff" done there is much more extreme, considering how there's little vehicle training, not much maintenance, and driving continues at all times and in all weather.

This isn't to say we shouldn't try to make it safer, but that we should correctly understand where the risks are.

Fatalities/VMT isn't the right stat to use to evaluate safety because injuries and deaths cluster around conflict points(intersections, crossovers to oncoming lane, offramp exits, etc.). It can be simultaneously true that most of the miles driven are safe, but also that there are danger zones which some road users are repeatedly subjected to. And when we talk about AVs being safe, we aren't saying "they'll be safer at avoiding a crash when racing", a situation where one might credibly consider using VMT. It's relatively mundane situations of making a clean left turn, stopping at stop signs, decelerating when exiting the freeway, and so forth. Those are things you encounter with higher frequency when you are making relatively shorter trips.

Self driving cars are probably safer for pedestrians as well. They can see in the dark and should have better reflexes.

Besides being fine, I also doubt it will stay true.

Datacenters have been consuming more and more energy because the work they are expected to do is always increasing. Self-driving cars have some phenomenally large problem to solve... and that's it. Any efficiency gain on those computers will realize as reduced consumption.

You need to take Wirth's law into consideration.

Please define the uncommon words you use...

https://en.wikipedia.org/wiki/Wirth%27s_law

The only thing that's going to constrain how much compute that cars will consume is cost and impact on range.

Except that's not it.

The problem will get bigger and bigger, requiring more and more compute power. You see, whatever system you design to autonomously control a car will never be "enough", or perfectly safe. There are simply too many variables in the environment, too many possibilities. So a certain design, taking a certain amount of compute power, will give you a certain level of safety. But it won't be perfect. People will want more safety and better systems. That'll take more power (probably exponentially more, as we've already seen with other simulations). The cycle will never end.

that cycle will break at the target markets and their price sensitivity.

similar to how current safety mechanisms are implemented. * two airbags, low weight constructions in entry segments targetting higher mileage versus 5+ airbags, advanced manufacturing processes that score high in crash tests and more as the segment price increases.

But sometimes it's saved elsewhere. Home office often increases energy usage in data centers through chat software, video conferencing etc. But it saves much more in energy otherwise wasted for commute.

Likewise, other computations in data centers might also save energy used somewhere else. Given all that servers do, 1% of electricity really doesn't seem much.

They can also be programmed to drive in a more efficient way that humans so there is a net saving of power.

I think that pretty much sums it up on many levels. I think of not just better route planning, but that one delivery driver can service 100's of customers who would otherwise make a trip themselves.

For delivery you're right but we could have this today already. There's no reason not to put minimum wage drivers in a car and have them do this today. We have computers and mapping already, self-driving isn't needed for this at all and paying someone 100 dollars a day or whatever is guaranteed to be much cheaper than the self delivering truck - at least for now and in the foreseeable future.

So why isn't it a thing? Clearly not for the lack of self-driving tech.

Do you actually have any data to back up the 'net saving of power' part?

It's more palatable to be moving at only 65mph or even 55mph if you're a passenger and don't have to pay attention, as opposed to being an active driver going 75-80mph to finish the task as soon as possible. And I doubt self driving cars will be programmed to speed and then slow down when they spot a cop, where human drivers are generally overspec'd enough to do that.

If this is a serious question then you can look to eight years of autonomus vehicle records held by Rio Tinto in the mega scale mining sector (800 million tonne / annum moved) running dump trucks, trains, etc.

Continous 24/7/365 operations show operational energy amd maintenance efficiences from supervised autonomous trucks over human operated trucks.

No, but I think it’s quite obvious humans do not drive with efficiency in mind the vast majority of the time.

> If autonomous driving saves a ton of lives then an extra 1% is well worth it. Just from productivity increases alone it may also be well worth it

I think that's a pretty optimistic assessment of it. So far they've only been capable of driving in conditions where humans don't have accidents anyway, and driving in more difficult conditions is a bit of a "ladder to the Moon" scenario.

I don't buy into the whole "think of all the time you'll save" thing either.

Reply on news.ycombinator.com