Robots and their impact on the future

[EscapeVelocity2020]

Assuming that the majority of people continue more or less their current lifestyles and consumption, than a UBI high enough that no one needs to work will leave enough excess beyond the basic necessities that it would still be possible to save enough to invest.
The average person would complain and insist that the amount wasn't enough to live on, but the sort of person who posts on these forums would be slowly building capital even if they started with none.

But in a UBI world, the average person would not see the motivation to save and invest.  Even a Mustachian might have a hard time convincing their kids that deferred gratification provides some ability to ... what, retire early?  There is nothing to 'retire early' from!  Spend a little more than the average in old age?  Meh, better to enjoy it when you are young and healthy....

There is going to have to be a transition from the economic motivations that work in today's society over to what will motivate people in a UBI world.  Society will have to look a lot different, or else there will be a violent revolution due to the incredible inequality between a tiny fraction of powerful owners and a giant multitude of powerless folks that want a little more and have lots of time on their hands...

[toganet]

Assuming that the majority of people continue more or less their current lifestyles and consumption, than a UBI high enough that no one needs to work will leave enough excess beyond the basic necessities that it would still be possible to save enough to invest.
The average person would complain and insist that the amount wasn't enough to live on, but the sort of person who posts on these forums would be slowly building capital even if they started with none.

But in a UBI world, the average person would not see the motivation to save and invest.  Even a Mustachian might have a hard time convincing their kids that deferred gratification provides some ability to ... what, retire early?  There is nothing to 'retire early' from!  Spend a little more than the average in old age?  Meh, better to enjoy it when you are young and healthy....

There is going to have to be a transition from the economic motivations that work in today's society over to what will motivate people in a UBI world.  Society will have to look a lot different, or else there will be a violent revolution due to the incredible inequality between a tiny fraction of powerful owners and a giant multitude of powerless folks that want a little more and have lots of time on their hands...

Most descriptions of UBI I have heard emphasize the "Basic" part, so I see a possible motivation in folks' desire to acquire luxury-type things.  So if everyone is getting the same $X that is calculated to cover the basics, I can choose to be more frugal than my neighbors and use my surplus to do the things I want to do.  This might be travel or entertainment, or just a choice of better restaurants over others.

The part that I can't quite work out is how a Basic income ultimately allows for the type of non-basic economic activity to happen at the scale it does today.  How can I run a gourmet restaurant if most of my patrons can't afford to eat there often? I may be creating amazing art now that I don't have to work minimum wage to cover my needs -- but so are tons of other folks so it's a commodity.  And if only a tiny fraction of society can afford to buy it, I might as well give it away, right? 

The cascading effects of shifted economic incentives seem to lead to too many contradictions.  However, if we do it Trekonomics-style (no artificial scarcity, energy is free, robots do all the dirty work) then the incentives to do anything "for money" goes away, and humans can truly focus their energies on the things that make us human: art, learning, relationships, travel ...

[maizefolk]

What do you see as the differences in definition between a UBI world and a trekonomics world?

[Bakari]

Assuming that the majority of people continue more or less their current lifestyles and consumption, than a UBI high enough that no one needs to work will leave enough excess beyond the basic necessities that it would still be possible to save enough to invest.
The average person would complain and insist that the amount wasn't enough to live on, but the sort of person who posts on these forums would be slowly building capital even if they started with none.

But in a UBI world, the average person would not see the motivation to save and invest.  Even a Mustachian might have a hard time convincing their kids that deferred gratification provides some ability to ... what, retire early?  There is nothing to 'retire early' from!  Spend a little more than the average in old age?  Meh, better to enjoy it when you are young and healthy....

There is going to have to be a transition from the economic motivations that work in today's society over to what will motivate people in a UBI world.  Society will have to look a lot different, or else there will be a violent revolution due to the incredible inequality between a tiny fraction of powerful owners and a giant multitude of powerless folks that want a little more and have lots of time on their hands...

In the current world the average person doesn't have the motivation to save and invest. 
There are plenty of mustachians who don't want to literally "retire" so it still seems similar to me.  The motivation to save and invest would be to become one of those "powerful owners", or at least a little more in that direction.  Like, I am in no way remotely close to being a real estate mogul, but saving and investing (on an average 20k income) got me to owning 3 rental units, which is a big step up from renting myself.

Most descriptions of UBI I have heard emphasize the "Basic" part, so I see a possible motivation in folks' desire to acquire luxury-type things.  So if everyone is getting the same $X that is calculated to cover the basics, I can choose to be more frugal than my neighbors and use my surplus to do the things I want to do.  This might be travel or entertainment, or just a choice of better restaurants over others.

The part that I can't quite work out is how a Basic income ultimately allows for the type of non-basic economic activity to happen at the scale it does today.  How can I run a gourmet restaurant if most of my patrons can't afford to eat there often? I may be creating amazing art now that I don't have to work minimum wage to cover my needs -- but so are tons of other folks so it's a commodity.  And if only a tiny fraction of society can afford to buy it, I might as well give it away, right? 

The cascading effects of shifted economic incentives seem to lead to too many contradictions.  However, if we do it Trekonomics-style (no artificial scarcity, energy is free, robots do all the dirty work) then the incentives to do anything "for money" goes away, and humans can truly focus their energies on the things that make us human: art, learning, relationships, travel ...

If it were truly only enough the basic basics, perhaps enough people would be motivated to do the types of "artistic" human work that Sol listed that there would always remain a (perhaps smaller, but still significant) employed middle class who could afford middle luxuries, like eating out and traveling.  In other words, the artists and other chefs will be your patrons.

[toganet]

What do you see as the differences in definition between a UBI world and a trekonomics world?

Mainly the amounts (and existence) of money involved. 

In a UBI world, my understanding is that individuals would receive some amount of money to cover their basic needs, such as housing, food, education, healthcare.  This might be done via universal-access systems like in Scandanavia, plus things like "SNAP for all" and HUD-subsidized rent/mortgage for everyone, without resorting to actual cash payments.  (I am not advocating this).  So money is still a medium of exchange that is ultimately a proxy for labor, even if that labor has become automated.  Anything beyond basics would require additional labor to earn the funds to buy it.

In a Trekonomics scenario, money eventually ceases to exist.  When any physical good you might need can be created for next-to-nothing (think free, limitless energy powering advanced 3D printers) there's no need to exchange one thing for another, even time.  Time, in many ways, becomes the central question -- now that all needs AND wants are attended to, how does one choose to spend one's time?  Kinda like personal finance, you can do anything you want -- but you can't do everything you want.

[maizefolk]

Makes sense, thanks!

[Bakari]

In a Trekonomics scenario, money eventually ceases to exist.  When any physical good you might need can be created for next-to-nothing (think free, limitless energy powering advanced 3D printers) there's no need to exchange one thing for another, even time.

In addition to some of the things Sol pointed out (you can't 3D print human companionship), land is finite.  Population might also outpace our ability to provide raw materials to the robot printers if everyone can have everything they can think of for free, (so everyone has their own fleet of helicopters and yachts of various sizes).
That link claims Star Trek is in a universe where war doesn't exist and leisure and work are indistinguishable, yet the enterprise is armed with photon torpedos and there is a bar and a holodeck for one's off hours.
There may be limits to the degree of utopia that is possible (at least until we can clone minds in a software world)

[EscapeVelocity2020]

In an interesting confluence of technology advancement and science fiction, 3D printed houses are apparently on the near horizon.  Reminds me of an online futurist story ARS once linked to:  http://marshallbrain.com/manna1.htm

The building we exited was another one of the terrafoam projects. Terrafoam was a super-low-cost building material, and all of the welfare dorms were made out of it. They took a clay-like mud, aerated it into a thick foam, formed it into large panels and fired it like a brick with a mobile furnace. It was cheap and it allowed them to erect large buildings quickly. The robots had put up the building next to ours in a week.

The government had finally figured out that giving choices to people on welfare was not such a great idea, and it was also expensive. Instead of giving people a welfare check, they started putting welfare recipients directly into government housing and serving them meals in a cafeteria. If the government could drive the cost of that housing and food down, it minimized the amount of money they had to spend per welfare recipient.

As the robots took over in the workplace, the number of welfare recipients grew rapidly. Manna replaced tens of millions of minimum wage workers with robots, and terrafoam housing became the warehouse of choice for them. Terrafoam buildings were not pretty, but they were incredibly inexpensive to build and were designed for maximum occupancy. They clustered the buildings on trash land well away from urban centers so no one had to look at them. It was a lot like an old-style college dorm. Each person got a 5 foot by 10 foot room with a bed and a TV -- the world's best pacifier. During the day the bed was a couch and people sat on the bedspread, which also served as a sheet and the blanket. At night the bed was a bed. When I arrived they had just started putting in bunk beds to double the number of people in each building.

[EscapeVelocity2020]

Very interesting AI discussion on the EconTalk podcast this week -
http://www.econtalk.org/amy-webb-on-artificial-intelligence-humanity-and-the-big-nine/

Russ Roberts: My guest is futurist and author Amy Webb.... Her latest book is The Big Nine: How the Tech Titans and Their Thinking Machines Could Warp Humanity.... Your book is a warning about the challenges we face, that we're going to face dealing with the rise of artificial intelligence.

So, there's a province in China where a new sort of global system is being rolled out. And it is continually mining and refining the data of the citizens who live in that area. So, as an example, if you cross the street when there's a red light and you are not able to safely cross the street at that point--if you choose to anyway, as to jay-walk--cameras that are embedded with smart recognition technology will automatically not just recognize that there's a person in the intersection when there's not supposed to be, but will actually recognize that person by name. So they'll use facial recognition technology along with technologies that are capable of recognizing posture and gait. It will recognize who that person is. Their image will be displayed on a nearby digital--not bulletin board; what do you call those--digital billboard. Where their name and other personal information will be displayed. And it will also trigger a social media mention on a network called Weibo. Which is one of the predominant social networks in China. And that person, probably, some of their family members, some of their friends, but also their employer, will know that they have--they have infracted--they have caused an infraction. So, they've crossed the street when they weren't supposed to. And, in some cases, that person may be publicly told--publicly shamed--and publicly told to show up at a nearby police precinct.

[Kate Leiton]

Thanks for sharing. For me this kind of future with AI is always a little bit scary. It's great in theory, but depends on how people will use it and that's where the problems may arise.

Very interesting AI discussion on the EconTalk podcast this week -
http://www.econtalk.org/amy-webb-on-artificial-intelligence-humanity-and-the-big-nine/

Russ Roberts: My guest is futurist and author Amy Webb.... Her latest book is The Big Nine: How the Tech Titans and Their Thinking Machines Could Warp Humanity.... Your book is a warning about the challenges we face, that we're going to face dealing with the rise of artificial intelligence.

So, there's a province in China where a new sort of global system is being rolled out. And it is continually mining and refining the data of the citizens who live in that area. So, as an example, if you cross the street when there's a red light and you are not able to safely cross the street at that point--if you choose to anyway, as to jay-walk--cameras that are embedded with smart recognition technology will automatically not just recognize that there's a person in the intersection when there's not supposed to be, but will actually recognize that person by name. So they'll use facial recognition technology along with technologies that are capable of recognizing posture and gait. It will recognize who that person is. Their image will be displayed on a nearby digital--not bulletin board; what do you call those--digital billboard. Where their name and other personal information will be displayed. And it will also trigger a social media mention on a network called Weibo. Which is one of the predominant social networks in China. And that person, probably, some of their family members, some of their friends, but also their employer, will know that they have--they have infracted--they have caused an infraction. So, they've crossed the street when they weren't supposed to. And, in some cases, that person may be publicly told--publicly shamed--and publicly told to show up at a nearby police precinct.

[maizefolk]

Agriculture is already a highly mechanized sector of the economy, but I thought this story was still pretty cool.

In Saskatchewan, the first commercially sold autonomous tractors made by Dot are hitting fields this spring.

The Dot units will not be completely on their own this year -- farmers who bought equipment as part of a limited release are required to watch them at all times. But after this trial run, the producers will be able to let the equipment run on its own starting next year. That will open up a lot of time for the growers who will no longer need to sit behind the steering wheel.

https://www.independent.co.uk/life-style/gadgets-and-tech/robots-farming-autonomous-equipment-canada-australia-a8919836.html

[G-dog]

Agriculture is already a highly mechanized sector of the economy, but I thought this story was still pretty cool.

In Saskatchewan, the first commercially sold autonomous tractors made by Dot are hitting fields this spring.

The Dot units will not be completely on their own this year -- farmers who bought equipment as part of a limited release are required to watch them at all times. But after this trial run, the producers will be able to let the equipment run on its own starting next year. That will open up a lot of time for the growers who will no longer need to sit behind the steering wheel.

https://www.independent.co.uk/life-style/gadgets-and-tech/robots-farming-autonomous-equipment-canada-australia-a8919836.html

I thought autonomous tractors were already in use in the US - though that people still sat in the cab, though it isn’t needed.  I may have assumed too much about that capability of the equipment.

[ender]

I thought autonomous tractors were already in use in the US - though that people still sat in the cab, though it isn’t needed.  I may have assumed too much about that capability of the equipment.

Most large ag manufacturers are hesitant to allow fully autonomous tractors without a driver, mainly for liability reasons.

I know most require weight in the seat. But I've seen some really hacky approaches to that - putting weights on the seat, etc.

[G-dog]

I thought autonomous tractors were already in use in the US - though that people still sat in the cab, though it isn’t needed.  I may have assumed too much about that capability of the equipment.

Most large ag manufacturers are hesitant to allow fully autonomous tractors without a driver, mainly for liability reasons.

I know most require weight in the seat. But I've seen some really hacky approaches to that - putting weights on the seat, etc.

I knew about the liability, but not the seat trigger. I guess hardcopy encyclopedias are still of use.  The field plan (crop, seed, planting density, etc.) is all programmed in, but I imagine that people are still needed for the transition.  Lots of similar technology in construction equipment - but I don’t know how it is being used (and any unions will likely resist it going fully autonomous)

[ctuser1]

I have not gone through the entire thread. Apologies if this is just regurgitating something someone else already posted.

AI is just a glorified name for automation, something that has been going on for hundreds of years. It can definitely replace "your" job (generic "you"), but it also creates many new jobs.

This is due to the O-Ring principle of Economic Development. https://en.wikipedia.org/wiki/O-ring_theory_of_economic_development
This theory has a long history starting with the NASA Challanger Shuttle disaster in 1986.

Here is how it goes:
1. There are 100 tasks in our fictitious value chain producing end client value.
2. AI is able to replace 80 of those 100 tasks, improving reliability, speed and accuracy AND decreasing cost for these tasks.
    - Because of the way typical incentive structures work, this is implemented asap so that managers controlling these 80 tasks can take all the credit and point fingers to the people controlling the rest of the 20 tasks.
3. As per the O-Ring Theory, the rest of the 20 tasks, things that can not be done by AI, suddenly become very important. They need to be improved as well, else the entire value chain is denied any improvement. The labor saved in the AI-replaced tasks are now devoted to these 20.

We have seen this O-Ring theory bear out in many industries. Look at all the bank tellers replaced by the ATM machines. You can go to "https://www.bls.gov/charts/employment-situation/employment-levels-by-industry.htm" and pull up the chart for "Financial Activities". Or you can go to "https://www.bls.gov/opub/ted/2016/employment-by-industry-1910-and-2015.htm" and compare 2015 against 1915 - i.e. a century.
 
In all data, you do not only NOT see a sudden cliff you'd expect when ATM's invaded the bank tellers job - but there seems to be gradual increase.

What happened?

Well, when the bank tellers no longer had to count money, they were deployed to become personal bankers. The number of people employed did not go down, just that a different part of the job suddenly became important and they were all employed there.

The same thing played out in many other sectors. Spreadsheets invaded accounting and we no longer have the paper-pusher roles in any big corp that existed 50 years ago. But look at the number of professionals doing accounting, and it has not only NOT gone down, but increased dramatically. Automation made a large part of the corporate value chain more reliable and more efficient - the ERP/SCM etc. Accounting HAD To keep up or else no benefits would accrue. Hence came many more accountants who no longer did much arithmatic by handheld calculators - but work on optimizing a new part of the workflow that simply did not exist 50 years ago.

The above assumes, of course, that there will always be the 20 jobs in the value chain of 100 that AI can't do. This has held so far, and I believe will hold for decades into the future unless some earth-shattering development in CS and Math comes along and is not expected anytime soon.

My knowledge in this field is a decade+ dated, since when I dabbled into NN and learning in college. Most of the technological developments of AI happened after 2012 (https://qz.com/1307091/the-inside-story-of-how-ai-got-good-enough-to-dominate-silicon-valley/).

As far as I understand, however, all these developments since 2012 are purely technological. The background math and science has not improved much. If so, the current boom of AI will be limited to "Weak AI": https://en.wikipedia.org/wiki/Weak_AI.

To break the assumption, and have AI replace all 100 tasks in the value chain, you need "Strong AI": https://en.wikipedia.org/wiki/Artificial_general_intelligence. We are nowhere near this!

Summary:
Fear of AI is overblown. Rest easy. Yes, "your job" may be automated, but there will always be new opportunities created as long as you stay nimble on your feet.
We are nowhere near the future where humans, as a whole, are outdated.

[sol]

the current boom of AI will be limited to "Weak AI": https://en.wikipedia.org/wiki/Weak_AI.

To break the assumption, and have AI replace all 100 tasks in the value chain, you need "Strong AI": https://en.wikipedia.org/wiki/Artificial_general_intelligence. We are nowhere near this!

The recurring theme of this thread has been the surprising number of tasks, previously thought to require strong AI, that are turning out to be profitably automated by weak AI systems.  The whole notion of "strong AI" is starting to look like human conceit, a lie we have told ourselves to protect our collective ego.  Those jobs only required strong AI because they were inefficiently structured to begin with.

As it turns out, we are a whole lot less special than we thought we were and many of the jobs that we thought required our uniquely human abilities can in fact be broken down into parts that can be automated, or at the very least the job itself can be modified in ways that allow it to be automated.  See Amazon's warehouse robots for an example of a job that previously was thought to require strong AI until the entire warehouse operations structure was rebuilt around robots.  That same process is ongoing in variety of operations.

The real threat doesn't appear to be that AI will replace you in your current job, whatever it is, but that AI will change the way your industry operates in such a way that your job in its current form longer exists at all, for humans OR robots.

[roomtempmayo]

https://www.thetruthaboutcars.com/2018/12/guns-drunks-and-rage-waymo-self-driving-vans-targeted-by-angry-arizonians/

Will automation bring out the worst in a handful of people?

Yes, absolutely.  If automation were to put >50% of the workforce out of work, there's no UBI that's going to be an adequate bandaid.  I'd give people one election cycle to nationalize industry, either by ballot or force.

The technology isn't the problem with automation, it's the distributive politics.  To date, we've automated/outsourced (to a worker the difference is irrelevant) maybe 10% of the workforce, and we've already got Trumpism and its global variants.  Multiply that dynamic several times, include former professionals with leadership abilities among the displaced, and we'd be in a very dark, very violent place.

Previous economic transitions worked because they made most people better off.  There's no reason to think an unregulated transition to AI is going to make most people better off.  I expect just the opposite instead, and there's no way that's going to be stable/sustainable.

[Bakari]

compare 2015 against 1915 - i.e. a century.
 
In all data, you do not only NOT see a sudden cliff ... but there seems to be gradual increase.

While what you are describing is probably a legitimate factor, I think the way the data is being looked at is misleading.
The "Employment levels" chart is looking at number of jobs, and the employment by industry is looking at the distribution of jobs.
Neither is considering hours of labor (or compensation thereof) per capita.
Between 1999 and 2019 there was an increase of 50 million in population.  The BLS chart you linked to shows in increase of 23 million jobs over that span. 
So the data is misleading - while it looks at first glance like an increase in jobs, it actually represents a very significant decrease in employment.

The so-called "unemployment" data has the same problem.  The official definition of unemployment requires a person to be actively seeking a job, which means anyone who permanently leaves the workforce does not count.

If you look at actual person labor-hours, there has in fact been a massive decrease since the time of the industrial revolution.  The decrease has just been absorbed by changing societal norms and expectations, and by government redistribution of wealth.
In 1900 about 20% of 10-15 year olds had full time jobs.  Prior to 1935 there was no such thing as a "retirement age" for most people, you worked until you were physically incapable, or until you died, which ever came first. Those two things alone represents a significant decrease in labor force participation rate, without the individuals who aren't working being considered "unemployed" SDI and AFDC allows many people who would need to hold whatever menial labor jobs they could get to survive do other things, like attend school, care for children, or just live out their days being insane in peace. (Granted, the increase in women in the work force goes against that trend)
Largest of all, the standard work week was reduced from anywhere from about 60-70 in the 19th and early 20th century to roughly 40 today.  That change alone increases the number of people who are employed by 50% to get the same amount of work done.
This is what has allowed jobs to be lost to automation and never replaced, without having any apparent impact on employment.  The distribution of the remaining jobs changed.

Todays decrease in average wages relative to GDP is a symptom of the decreased need for labor (without any accompanying updated change in distribution of labor availability, like a shorter workweek or increased minimum wage) - labor in a market economy is like any other resource: if supply outpaces demand, price (wage) falls.

We don't need strong AI for this to become a crises, we are already on that track as long as we continue with out current momentum.  AI just promises to get us there faster.

[Spud]

Interesting 13 minute video on the development of autonomous trucks. The don't need to sleep so you can cross the country in half the time. They don't need to stop for a piss. They don't get distracted send a text whilst they drive. They don't break the speed limit. The way they drive is more fuel efficient and makes tyres last longer. Humans might still do the short tricky journeys at either end, but it looks like the 250-1000 miles between hubs that make up the middle/majority of a trucking route in states where the weather is consistently warm and dry, no longer needs a human.

This could threaten around 1.8 million jobs...

https://www.youtube.com/watch?v=vMXivgUGVn8

[robartsd]

The don't need to sleep so you can cross the country in half the time.

Plenty of team drivers already operating nearly round the clock. Of course that basically just means each autonomous truck is replacing two or more workers.

[maizefolk]

It's been a while, but I think this is an interesting type of job being replaced by automation we haven't talked about before.

IHeartRadio (nee Clearchannel) announced the layoff of 850 employees from their radio stations, including more than 60 DJs because of investments in automation and AI for programming/podcasting/marketing, etc.

Source.

[joshuagraham_xyz]

Once we invent cheap, attractive sex bots (and get over that uncanny valley issue) I fully expect the human race to die out in a single generation . . . maybe two at the most.

I could see men that are shut out of the sexual marketplace, as it were, going this route, but I can't see women doing this.  The reason for men having sex is to achieve orgasm, and a good enough sex robot could make this happen; actually, all that would need to happen is to have a device that could choke the chicken for him while the man watches porn, especially the live porn that sexy webcams produce.  Women, however, need to be validated that a real man of sufficiently good genetic stock, and typically also with a fat wallet, is laying the pipe; a sex robot cannot do this.

What it will do is greatly decrease the number of men that hang out at Tinder, or at the local nightclub, etc., but since most of these men have been getting shunned by women, it won't have much of an effect for younger women, but most definitely could lessen the number of men that are willing to "man up" to marry older women, so we'll probably see even more articles about how difficult it is for (older) women to find a husband.  Of course, since the most important thing to a woman is not to have the husband (finances aside), but rather a good genetic donor's sperm, we'll see even more sperm bank business as competition ramps up for the genetically desirable college students to give sperm samples.

[sixwings]

Assuming that the majority of people continue more or less their current lifestyles and consumption, than a UBI high enough that no one needs to work will leave enough excess beyond the basic necessities that it would still be possible to save enough to invest.
The average person would complain and insist that the amount wasn't enough to live on, but the sort of person who posts on these forums would be slowly building capital even if they started with none.

But in a UBI world, the average person would not see the motivation to save and invest.  Even a Mustachian might have a hard time convincing their kids that deferred gratification provides some ability to ... what, retire early?  There is nothing to 'retire early' from!  Spend a little more than the average in old age?  Meh, better to enjoy it when you are young and healthy....

There is going to have to be a transition from the economic motivations that work in today's society over to what will motivate people in a UBI world.  Society will have to look a lot different, or else there will be a violent revolution due to the incredible inequality between a tiny fraction of powerful owners and a giant multitude of powerless folks that want a little more and have lots of time on their hands...

This is basically what will happen, it will probably be the tech titans vs. everyone else, and everyone else will lose.

[TempusFugit]

I haven't heard much lately about how self-driving cars are doing out in the world.  Tesla seems to make news for lots of stuff lately but not accidents, thankfully, I guess.   Not only has another year passed but they've also sold probably 40-50 thousand more cars with 'auto-pilot' mode available. With people being people, I'm sure just as many idiots are ignoring the road and letting the car drive but somehow they must be surviving.    Maybe this means the tech is really getting closer to what we might consider fully self driving. 

I've ridden in a Telsa Model S once and it was kind of cool, sure.  Very powerful and stable feeling.  I can't image how much it costs to keep tires on one with all that torque. 

Not my thing really for the money, but the people I know (just 3 people) that have them really love them.

[lost_in_the_endless_aisle]

I haven't heard much lately about how self-driving cars are doing out in the world.  Tesla seems to make news for lots of stuff lately but not accidents, thankfully, I guess.   Not only has another year passed but they've also sold probably 40-50 thousand more cars with 'auto-pilot' mode available. With people being people, I'm sure just as many idiots are ignoring the road and letting the car drive but somehow they must be surviving.    Maybe this means the tech is really getting closer to what we might consider fully self driving. 

I've ridden in a Telsa Model S once and it was kind of cool, sure.  Very powerful and stable feeling.  I can't image how much it costs to keep tires on one with all that torque. 

Not my thing really for the money, but the people I know (just 3 people) that have them really love them.

I think the analysis of Tesla safety is confounded by the fact early adopters were probably higher income, which would correlate to certain Big 5 personality traits associated with lower accident rates. I guess if we consider Model S cheap enough for the hoi polloi then maybe we are starting to see a real signal of the effectiveness of autopilot. I'm an AI pessimist in that I believe until we solve the general Hard Problem of Consciousness--which I unjustifiably think is a necessary precondition for achieving general intelligence--that automated systems will keep encountering endless edge-cases where they catastrophically fail (abusrdly extreme failure mode in current visual processing AI--single pixel attacks indicate neural networks are subject to extreme hallucinations on occasion, though maybe these problems can be smoothed over via GAN machine learning approaches). However, if I'm right, driving well in all circumstances is an AGI problem, not a narrow-AI problem. In controlled conditions (e.g. freeways) narrow-AI can work but I suspect it will have many intriguing failure modes when it encounters conditions one or two standard deviations outside of normal. The alternative is to convert all cars at once to reduce the possibility for failure but of course that is a coordination problem and we will probably have many different brands/flavors of self-driving algorithms instantiated at any given time. I'm going to stick with my '09 Civic for now.

[GuitarStv]

Once we invent cheap, attractive sex bots (and get over that uncanny valley issue) I fully expect the human race to die out in a single generation . . . maybe two at the most.

I could see men that are shut out of the sexual marketplace, as it were, going this route, but I can't see women doing this.  The reason for men having sex is to achieve orgasm, and a good enough sex robot could make this happen; actually, all that would need to happen is to have a device that could choke the chicken for him while the man watches porn, especially the live porn that sexy webcams produce.  Women, however, need to be validated that a real man of sufficiently good genetic stock, and typically also with a fat wallet, is laying the pipe; a sex robot cannot do this.

What it will do is greatly decrease the number of men that hang out at Tinder, or at the local nightclub, etc., but since most of these men have been getting shunned by women, it won't have much of an effect for younger women, but most definitely could lessen the number of men that are willing to "man up" to marry older women, so we'll probably see even more articles about how difficult it is for (older) women to find a husband.  Of course, since the most important thing to a woman is not to have the husband (finances aside), but rather a good genetic donor's sperm, we'll see even more sperm bank business as competition ramps up for the genetically desirable college students to give sperm samples.

I'll preface this by saying that I'm certainly not an expert on the minds of all human women as they pertain to sex . . . but in my short stay here on planet Earth I've found multiple women who enjoyed the act of sex just as much as men.  :P

[dougules]

I think the analysis of Tesla safety is confounded by the fact early adopters were probably higher income, which would correlate to certain Big 5 personality traits associated with lower accident rates. I guess if we consider Model S cheap enough for the hoi polloi then maybe we are starting to see a real signal of the effectiveness of autopilot. I'm an AI pessimist in that I believe until we solve the general Hard Problem of Consciousness--which I unjustifiably think is a necessary precondition for achieving general intelligence--that automated systems will keep encountering endless edge-cases where they catastrophically fail (abusrdly extreme failure mode in current visual processing AI--single pixel attacks indicate neural networks are subject to extreme hallucinations on occasion, though maybe these problems can be smoothed over via GAN machine learning approaches). However, if I'm right, driving well in all circumstances is an AGI problem, not a narrow-AI problem. In controlled conditions (e.g. freeways) narrow-AI can work but I suspect it will have many intriguing failure modes when it encounters conditions one or two standard deviations outside of normal. The alternative is to convert all cars at once to reduce the possibility for failure but of course that is a coordination problem and we will probably have many different brands/flavors of self-driving algorithms instantiated at any given time. I'm going to stick with my '09 Civic for now.

Just to note, humans are still constantly encountering endless edge-cases where they catastrophically fail.  It only has to be better than a human, and that's a really low bar. 

[scottish]

First, since when is it acceptable for automobiles to be just a bit better than a human?   

Second, we can fail on edge cases, but we can also slow down, think about it, figure out what's going on and make a good decision.

Your self driving Tesla can't do that.   If you're lucky it will slow down before doing anything irrevocable.    Thinking about a situation is completely off the table.   "AI" doesn't think.

Another problem nobody talks about - software bugs.    All the Teslas running the same software load will have the same software bugs.    Isn't that great?      They'll all be susceptible to the same problems...

[robartsd]

First, since when is it acceptable for automobiles to be just a bit better than a human?   

AI driven cars being safer than human drivers in 99%+ cases seems like a reasonable threshold for mass implementation to me. Not sure at what point it would be reasonable to ban human drivers though.

Another problem nobody talks about - software bugs.    All the Teslas running the same software load will have the same software bugs.    Isn't that great?      They'll all be susceptible to the same problems...

Yes there would be much more uniformity in performance of AI drivers than human drivers. And when a probem is found and corrected, the correction will be rolled out to the AI drivers making them all better.

[dougules]

Humans can slow down and think, but generally don't.  That kid texting is not going to slow down before doing something irrevocable.  Human reaction time is very slow, too.  AI may not be able to think about novel situations, but that ability isn't always much benefit for human drivers given that there's not always much time to think moving at 60mph.

Humans are very much susceptible to software bugs.  Mental health is a big problem.  I'm sure therapists could tell you that it's pretty hard to push updates to humans.

I'm sceptical of the limits of AI, too, but any flaws in the ability of software to drive should be compared to the alternative which is human drivers.  We feel better thinking we have control of the car, but the math shows that that is just an illusion.  Even then, statistics on traffic accidents implicitly take into account the fact that the designs of our roads have been tweaked for years to work with or mitigate the quirks of human cognition.  Painted lines, signs, speed limits, stop lights, etc. are not actually needed for a car to move down the road.  They’re solely there to help out a system that is not well designed to control a ton of metal rolling at 30m/sec.  Any discussion on the shortcomings of automated driving needs to be in comparison to the many equivalent shortcomings of humans.

[SauronHimself]

Humans can slow down and think, but generally don't.  That kid texting is not going to slow down before doing something irrevocable.  Human reaction time is very slow, too.  AI may not be able to think about novel situations, but that ability isn't always much benefit for human drivers given that there's not always much time to think moving at 60mph.

Humans are very much susceptible to software bugs.  Mental health is a big problem.  I'm sure therapists could tell you that it's pretty hard to push updates to humans.

I'm sceptical of the limits of AI, too, but any flaws in the ability of software to drive should be compared to the alternative which is human drivers.  We feel better thinking we have control of the car, but the math shows that that is just an illusion.  Even then, statistics on traffic accidents implicitly take into account the fact that the designs of our roads have been tweaked for years to work with or mitigate the quirks of human cognition.  Painted lines, signs, speed limits, stop lights, etc. are not actually needed for a car to move down the road.  They’re solely there to help out a system that is not well designed to control a ton of metal rolling at 30m/sec.  Any discussion on the shortcomings of automated driving needs to be in comparison to the many equivalent shortcomings of humans.

The NHTSA found that 94% of all crashes are due to human error.

[maizefolk]

A big part of the problem is that we tend to over-estimate how good of drivers we, personally are. This statistic is a bit dated and based on an small sample size, but 93% of americans reported assess their driving skills as above average.

The average driver is not very good and a significant number of people are worse than average. We (as a population) text while driving. We eat while driving. We get behind the wheel after anywhere from 1-20 beers. We get in arguments with passengers in the car. We get lost and are too busy looking for the sign for the next turn to focus on the stoplight that just changed color. We can be turn our heads to look at a deer that just ran across the road and so do not see the second deer following it across the road and in front of our car.  We get lost daydreaming about the girl/boy we dated in high school on I-80 in Wyoming where nothing changes for hours.

[robartsd]

A big part of the problem is that we tend to over-estimate how good of drivers we, personally are. This statistic is a bit dated and based on an small sample size, but 93% of americans reported assess their driving skills as above average.

I think part of this may be that we notice most of the bad drivers that we come across, but fail to notice how many good drivers are around as well.

[dougules]

A big part of the problem is that we tend to over-estimate how good of drivers we, personally are. This statistic is a bit dated and based on an small sample size, but 93% of americans reported assess their driving skills as above average.

I think part of this may be that we notice most of the bad drivers that we come across, but fail to notice how many good drivers are around as well.

Even the best drivers have significant limitations just from being human.  Reaction time is very slow for even the quickest person. 

[dougules]

The NHTSA found that 94% of all crashes are due to human error.

Honestly I'm a little surprised that it's that low. 

[scottish]

The NHTSA found that 94% of all crashes are due to human error.

Honestly I'm a little surprised that it's that low.

Yeah, what are the other crashes?    It's hard to think how a collision could not be due to human error at some level.

Could one of you school me in how you go about measuring the safety of autonomous cars as compared to people?

[TempusFugit]

A big part of the problem is that we tend to over-estimate how good of drivers we, personally are. This statistic is a bit dated and based on an small sample size, but 93% of americans reported assess their driving skills as above average.

The average driver is not very good and a significant number of people are worse than average. We (as a population) text while driving. We eat while driving. We get behind the wheel after anywhere from 1-20 beers. We get in arguments with passengers in the car. We get lost and are too busy looking for the sign for the next turn to focus on the stoplight that just changed color. We can be turn our heads to look at a deer that just ran across the road and so do not see the second deer following it across the road and in front of our car.  We get lost daydreaming about the girl/boy we dated in high school on I-80 in Wyoming where nothing changes for hours.

I think that driving is something that most of us learned to do at an early age and we’ve become so accustomed to it that we fail to realize just how complex it really is.  Not so much the mechanics of turning the wheel and using the pedals but the shear amount of information that we need to process in order to drive (mostly) safely in different environments and circumstances.  That leads us to underestimate just how hard it will be to make a machine that can do it.  Its relatively simple most of the time, but we have to deal with all the time. 

I think they'll eventually crack it but not so quickly as some think. But its exponential, so who knows.  I wouldn't have thought they could land the first stage rockets on a barge reliably, but they seem to have worked that one out. 

[maizefolk]

It's hard to think how a collision could not be due to human error at some level.

I'm assuming by human error they really mean driver error. A crash cased by brakes failing or a tire exploding is still an human error on some level in terms of maintaining the machine, but the driver could do everything right and still end up in a collision.

Or a cow walk into the road too close for a perfectly maintained car to stop. Still reflects a human error somewhere along the chain of events (how did the cow get loose?) but the driver may not be able to do anything to avoid it.

Maybe if we make the cow a moose no human error would be involved?

[TheAnonOne]

It's hard to think how a collision could not be due to human error at some level.

I'm assuming by human error they really mean driver error. A crash cased by brakes failing or a tire exploding is still an human error on some level in terms of maintaining the machine, but the driver could do everything right and still end up in a collision.

Or a cow walk into the road too close for a perfectly maintained car to stop. Still reflects a human error somewhere along the chain of events (how did the cow get loose?) but the driver may not be able to do anything to avoid it.

Maybe if we make the cow a moose no human error would be involved?

Deer crashes are a norm in forested areas. These are not human caused. Unless you think it is reasonable to go 20MPH near trees.

[maizefolk]

Good point! Much better example than the moose.

[dougules]

Deer crashes are a norm in forested areas. These are not human caused. Unless you think it is reasonable to go 20MPH near trees.

I didn't think about animals.  I've had a very near miss with a deer driving myself and been in another near miss as a passenger. 

Predicting what animals are going to do is going to be hard for automation to account for, but I don't think people are that good at it either.  With automation, you might actually be able to bring animal behavioral experts into the design process. 

[scottish]

Deer crashes are a norm in forested areas. These are not human caused. Unless you think it is reasonable to go 20MPH near trees.

I didn't think about animals.  I've had a very near miss with a deer driving myself and been in another near miss as a passenger. 

Predicting what animals are going to do is going to be hard for automation to account for, but I don't think people are that good at it either.  With automation, you might actually be able to bring animal behavioral experts into the design process.

Autonomous vehicles can drive at safer speeds (20 mph?)  at twilight when wildlife is active.    People are too impatient to do this.

[Leisured]

Just to note, humans are still constantly encountering endless edge-cases where they catastrophically fail.  It only has to be better than a human, and that's a really low bar.

+1

[Leisured]

In Australia, kangaroos are notorious for crossing a road at night whenever they feel like it. Headlights do not matter. Around dusk in Kangaroo country we reduce speed for that reason.

[lost_in_the_endless_aisle]

I think the analysis of Tesla safety is confounded by the fact early adopters were probably higher income, which would correlate to certain Big 5 personality traits associated with lower accident rates. I guess if we consider Model S cheap enough for the hoi polloi then maybe we are starting to see a real signal of the effectiveness of autopilot. I'm an AI pessimist in that I believe until we solve the general Hard Problem of Consciousness--which I unjustifiably think is a necessary precondition for achieving general intelligence--that automated systems will keep encountering endless edge-cases where they catastrophically fail (abusrdly extreme failure mode in current visual processing AI--single pixel attacks indicate neural networks are subject to extreme hallucinations on occasion, though maybe these problems can be smoothed over via GAN machine learning approaches). However, if I'm right, driving well in all circumstances is an AGI problem, not a narrow-AI problem. In controlled conditions (e.g. freeways) narrow-AI can work but I suspect it will have many intriguing failure modes when it encounters conditions one or two standard deviations outside of normal. The alternative is to convert all cars at once to reduce the possibility for failure but of course that is a coordination problem and we will probably have many different brands/flavors of self-driving algorithms instantiated at any given time. I'm going to stick with my '09 Civic for now.

Just to note, humans are still constantly encountering endless edge-cases where they catastrophically fail.  It only has to be better than a human, and that's a really low bar.

In a more rational world, I would agree this would be the hurdle to clear. In our actual world, automated systems will have to be 10-1,000x safer than humans. The two alternatives will not be measured on equal footing for several reasons:

1) liability issues for the companies that produce autonomous vehicles
2) media amplification of relatively rare accidents that are novel in that they involve automated driving systems and resulting public misconception of relative risk
3) problems with security and hacking (speculative, though it seems you can hack smart coffeemakers with ransomware)

I'm long-term optimistic about technology but short-term pessimistic on how rationally people will judge the tradeoffs when automated systems are only modestly better than humans.

[ender]

1) liability issues for the companies that produce autonomous vehicles
2) media amplification of relatively rare accidents that are novel in that they involve automated driving systems and resulting public misconception of relative risk
3) problems with security and hacking (speculative, though it seems you can hack smart coffeemakers with ransomware)

I'm long-term optimistic about technology but short-term pessimistic on how rationally people will judge the tradeoffs when automated systems are only modestly better than humans.

This is actually why I think automated cars are going to die.

Not because the tech is better/safer than humans, but because the liability aspects amplified by the media are going to make it practically impossible to have the tech widely accepted by people who can't do basic statistics.

Nevermind all the stupid drivers that cause deaths now, if there's a few with automated cars it's going to be all over the media. People eat that type of news up. And the media is turning into clickbait anyways which "Automated car kills innocent pedestrian" will turn into a frenzy.

[maizefolk]

You think so? I agree the unequal standard applied to reporting the tens of thousands of people killed in regular car accidents vs those where the car might have been self driving could doom adoption. But at the same time I'm starting to wonder if the various fatal collisions with Teslas are starting to produce "automated car accident" story fatigue (just like regular car accidents rarely make the news anymore), even though at the moment Telsa usually states the driver was in control of the vehicle.

[lost_in_the_endless_aisle]

You think so? I agree the unequal standard applied to reporting the tens of thousands of people killed in regular car accidents vs those where the car might have been self driving could doom adoption. But at the same time I'm starting to wonder if the various fatal collisions with Teslas are starting to produce "automated car accident" story fatigue (just like regular car accidents rarely make the news anymore), even though at the moment Telsa usually states the driver was in control of the vehicle.

Well in my case I am uncertain at what the level of bias against automated tech failures will be, hence my massive 10-1000x range. I definitely do not think it is 1:1 but if at the lower end, maybe that is reachable on a short time horizon. However, consider that airplanes still have 2 pilots when there are probably reasonable arguments for 1 or 0 given current technology*. Of course, air travel is strange since people demand that it be 100,000x safer per passenger mile than ground transit. If having just 1 pilot brings that down to 10,000x safer I see that being a hard sell. The irrationality of people extends over many dimensions and its not clear how the specific case of automated driving will manifest.

*I have skimmed arguments for this a long time ago but am no expert and could be completely wrong

[maizefolk]

Air travel is an interesting analogy. I think one difference is that I haven't heard anyone argue that having pilots in the cockpit makes flying less safe.

The other big difference is that whether or not a plane is self flying doesn't change anything for me as a passenger. Maybe the price of a ticket goes down $5-10? Self driving cars, on the other hand, would make life appreciably better for a lot of people. Commuters get hours of their lives back every week to do other things. The elderly and children become far more mobile and independent. Alcoholics will be less likely to put their own lives and those of others at risk, without having to spring for the price of an uber.

So once safety is at or beyond parity, my guess would be that we'll see actual advocacy from people in those groups to legalize self driving, while self flying planes don't have the same built in constituencies.

[lost_in_the_endless_aisle]

Air travel is an interesting analogy. I think one difference is that I haven't heard anyone argue that having pilots in the cockpit makes flying less safe.

The other big difference is that whether or not a plane is self flying doesn't change anything for me as a passenger. Maybe the price of a ticket goes down $5-10? Self driving cars, on the other hand, would make life appreciably better for a lot of people. Commuters get hours of their lives back every week to do other things. The elderly and children become far more mobile and independent. Alcoholics will be less likely to put their own lives and those of others at risk, without having to spring for the price of an uber.

So once safety is at or beyond parity, my guess would be that we'll see actual advocacy from people in those groups to legalize self driving, while self flying planes don't have the same built in constituencies.

The argument that self-flying would be less dangerous would require a more detailed analysis of historical air accidents and incidents. It seems 50% of all accidents are caused by pilot error so there is plausibly a case to be made. Making flying meaningfully safer, however, is not easy since the fatality rate is already so low.

The higher scrutiny on air travel safety is probably a function of the fewer, larger accidents that do occur when something does go wrong. Just looking at US ground transportation accidents, 20 died in this one and 23 in this bus accident. Outside of these rare cases, almost every other ground transportation death involves one or several people at a time, which barely gets mentioned on the news or in a newspaper. If the mainstream media wasn't largely innumerate, I would be more confident that self-driving would be put into the proper context once the technology begins to be deployed at scale. On the other hand, Americans have become used to covid deaths so maybe the learning curve will be easier.