Full AI - The End of Humanity?

[Dotini]

Why do so many humans commit suicide? Does AI commit suicide too?

I predict the human race will come very, very close to committing suicide. The question is, will it be survived by a viable, replicating species of AI, or will Earth revert to dominant species of biology like insects, crows and wolves?

 

[Imari]

Why do so many humans commit suicide?

Because it's so unpleasant to be alive that they would rather not be.

I would suggest that if you were to create an AI that was in constant agony and also gave it the ability to commit suicide, you would very quickly have a dead AI.

To be fair, this general type of behaviour has evolutionary advantages for a species. If members are wounded or infected with some crippling illness in a hunter/gatherer society that is barely managing to scrape by, then it's better for the group as a whole that they simply remove the burden of themselves from the group.

That's not true in a modern sense, caring for people who are suicidal and helping them to return to functional and contributing members of society shouldn't be that difficult. And the return from having them functional again should outweigh whatever work was put into helping them get well. Unfortunately, we're still not very good at it and so there's a fair amount of people who just get to the point where they can't take it any more.

It's hard to know what an AI would perceive as unpleasant stimuli, and what our ability to correct that might be. I think that once you get to a sufficiently complex AI it's absolutely possible that they could exhibit a whole range of behaviours that would at least externally mimic human ones, including suicidal behaviour.

 

[UKMikey]

I think that once you get to a sufficiently complex AI it's absolutely possible that they could exhibit a whole range of behaviours that would at least externally mimic human ones, including suicidal behaviour.

 

[Danoff]

...from the Deep Thoughts thread.

Sorry, I meant healthy people or beings that mostly rely on their instincts. Humans that do not want to live are clinically ill.

Also not true. There are plenty of examples where people endanger and sacrifice themselves for the preservation of others. The same can be found in other animal species as well (birds having some fantastic examples). This is because natural selection works at the level of the gene rather than the level of the organism. The difference between preservation of genes and preservation of the individual is a subtle one, but it results in otherwise unexplainable behavior. Recommended reading on this subject is The Selfish Gene by Richard Dawkins.

Complete control can never be achieved if the AI is around organic life that is at its whims of often illogical feelings and making gut reactions. This is a tremendous danger to the AI survival and it goals, its first logical act would be to limit illogical human influence on its goals, this would go so far as to completely rule out any interference on its goals which means to neutralize every single human being.

No. What you just described is akin to finding and exterminating every mosquito on the planet to make sure I don't get bit. Not at all necessary if you have bug repellent, mosquito netting, or just thick clothing. A general artificial intelligence is only going to do what it needs to do to maximize its goals. It's not going to waste a bunch of energy and time carrying out some contingency process that isn't needed to accomplish its goals.

You can, but an AI that is capable of doing 500 years worth of human research in a day in its larva state would be able to bypass and rewrite any limits in minutes if not seconds. That kind of processing power is magnitudes beyond anything we can imagine or plan for.

You ignored the best part of my post, which is the part where I explained that once it can bypass the protections we might put in place, it can also bypass any objective we put in place and be done with its programming.

 

[Danoff]

 

[Dotini]

Who does humanity serve? Only humanity, I figure, and not any other species except maybe dogs. So the total loss of all humanity will not be any great problem for anything other than humanity. If all humanity is lost, does the universe suffer? I think not. The universe is almost entirely stars and ionized gases, unfit for any form of life except for the stars themselves. So the purpose of the universe is for the stars, not for man. Furthermore, I think the stars are in an important sense "alive", and our star is very well "aware" Earth is here, connecting to it constantly with light, gravitation, ionized particles, neutrons and cosmic rays. Possibly the Sun is a form of self-generating, self-organizing artificial intelligence. If I were a religious person, I might worship the Sun, praying daily for the continued life of Man on Earth.

 

[Danoff]

What does this have to do with AI?

Who does humanity serve?

What does anything serve? You can't name a living species that serves another, all species of life exist for the same reason, the propagation of genes via DNA. We're all biologically designed for that purpose, as replicators, of a stable form of matter (genes) that outwardly manifests itself as an organism.

If all humanity is lost, does the universe suffer? I think not.

The universe does not suffer under any circumstances.

 

[baldgye]

Because it's so unpleasant to be alive that they would rather not be.

I would suggest that if you were to create an AI that was in constant agony and also gave it the ability to commit suicide, you would very quickly have a dead AI.

To be fair, this general type of behaviour has evolutionary advantages for a species. If members are wounded or infected with some crippling illness in a hunter/gatherer society that is barely managing to scrape by, then it's better for the group as a whole that they simply remove the burden of themselves from the group.

That's not true in a modern sense, caring for people who are suicidal and helping them to return to functional and contributing members of society shouldn't be that difficult. And the return from having them functional again should outweigh whatever work was put into helping them get well. Unfortunately, we're still not very good at it and so there's a fair amount of people who just get to the point where they can't take it any more.

It's hard to know what an AI would perceive as unpleasant stimuli, and what our ability to correct that might be. I think that once you get to a sufficiently complex AI it's absolutely possible that they could exhibit a whole range of behaviours that would at least externally mimic human ones, including suicidal behaviour.

Reading that reminded me about the AI from I Have No Mouth But I Must Scream

 

[Dotini]

This is an interesting statistic.

"At the moment, humans collectively drive for at least eight million hours on average before misidentifying something that leads to an accident. Currently, AVs can only manage 10,000-30,000 hours."

https://www.bbc.com/news/business-49420570

 

[TenEightyOne]

"At the moment, humans collectively drive for at least eight million hours on average before misidentifying something that leads to an accident.

Something's a bit off with that comparison. 8,000,000 hours would be a thousand years of continuous driving. If the figure's collective then arguably there are a lot more human drivers. With a million deaths in road crashes every year we're clearly not making the collective 1,000 years on average.

 

[Dotini]

Something's a bit off with that comparison. 8,000,000 hours would be a thousand years of continuous driving. If the figure's collective then arguably there are a lot more human drivers. With a million deaths in road crashes every year we're clearly not making the collective 1,000 years on average.

Well, the stat seems to come from BBC, so it's on them. But I'll guess it has to do with the proportion of crashes due to "misidentifying some thing".

 

[Danoff]

This is an interesting statistic.

"At the moment, humans collectively drive for at least eight million hours on average before misidentifying something that leads to an accident. Currently, AVs can only manage 10,000-30,000 hours."

https://www.bbc.com/news/business-49420570

What really matters is whether that experience can be shared. The human figure can't, but the AI figure can. What's more, AI doesn't have to actually drive on the road to log driving hours. Driving can be simulated at faster than real time without ever putting a car on the road.

 

[Dotini]

What really matters is whether that experience can be shared. The human figure can't, but the AI figure can. What's more, AI doesn't have to actually drive on the road to log driving hours. Driving can be simulated at faster than real time without ever putting a car on the road.

Another key take from that article is that AV is only really at its best when "guided by six cameras, two radars, 18 ultrasound sensors, and five lidar [light detection and ranging] sensors, and the onboard computer processor is capable of 24 trillion operations a second." If I read the article right, it may do even better with flashing laser light beams!

 

[Danoff]

Another key take from that article is that AV is only really at its best when "guided by six cameras, two radars, 18 ultrasound sensors, and five lidar [light detection and ranging] sensors, and the onboard computer processor is capable of 24 trillion operations a second." If I read the article right, it may do even better with flashing laser light beams!

Is that, a bad thing?

 

[Dotini]

Is that, a bad thing?

No, not necessarily, but at least for the moment it is an expensive thing. And I don't know about flashing laser lights, either.

 

[Wolfe]

No, not necessarily, but at least for the moment it is an expensive thing. And I don't know about flashing laser lights, either.

It's shocking how mindblowingly expensive it is, in collisions where human-powered sensors and the fleshy operator they belong to may be unharmed and require no hospital bills (thanks to more reasonable passive safety technology, it's worth noting).

As a proactive driver, I can't help but see the whole undertaking as awfully wasteful and impractical. Mountains of cash invested into desperately attempting to teach a computer to mimic the fatty lump between my ears -- mostly because we've never bothered to train drivers better, and would rather scapegoat distractions instead of teaching people to overcome them.

 

[Danoff]

It's shocking how mindblowingly expensive it is, in collisions where human-powered sensors and the fleshy operator they belong to may be unharmed and require no hospital bills (thanks to more reasonable passive safety technology, it's worth noting).

As a proactive driver, I can't help but see the whole undertaking as awfully wasteful and impractical. Mountains of cash invested into desperately attempting to teach a computer to mimic the fatty lump between my ears -- mostly because we've never bothered to train drivers better, and would rather scapegoat distractions instead of teaching people to overcome them.

It literally says it's not just the sensors that are increasing repair costs, but lightweight construction materials as well.

No amount of training is going to eliminate road deaths. People are still going to die. Many drivers simply aren't interested in driving, and they tend to make the worst drivers. You're not going to be able to train them up to be super vigilant or enthusiastic, they're just not going to be interested.

What autonomous vehicles represent is an enormous increase in standard of living for all people. Shipping costs gobs of money in part because you have to pay drivers to do it. Taxis cost gobs of money in part because you have pay to drivers to do it. Commuting sucks productivity and enjoyment out of the lives of countless people every day. You might enjoy commuting, I might enjoy commuting, but not everyone would spend their time that way. Some of them would rather be getting things done or spending more time with their kids.

Autonomous vehicles have the opportunity to offload air travel passengers, saving time and resources at airports and in airline fleets as well. As collisions go down, it reduces wasted money on repairs too.

Autonomous vehicles are not wasted effort, it is lifesaving technology that will boost productivity, quality of life, and reduce resource consumption for all of civilization. The impacts of this development are almost immeasurable, and will fundamentally alter the way we live forever. It's not a waste of time, it's an essential step in human development.

As a bi-product, it gets terrible drivers off the road and reduces traffic so that you can enjoy yourself. So you should be all for it.

 

[Imari]

As a proactive driver, I can't help but see the whole undertaking as awfully wasteful and impractical.

It's about as wasteful and impractical as anything post-industrial revolution. Sure, humans can do all this stuff manually if they want. You don't even need a car, you have perfectly good legs. Why have mechanical anything? Ned Ludd knew which side of his bread was buttered.

The reality is that automation frees humans to do the things that humans are best at. Creativity. Enjoyment. Relationships. Navel gazing. Et cetera. If there's a car that will drive me safely to work while I eat my toast and scratch my butt, I'm all for it. If there's people who still want to drive themselves, sure, I guess. I'm sure there's people who like watching paint dry too.

 

[Wolfe]

No amount of training is going to eliminate road deaths.

I firmly believe no amount of AI training is going to make autonomous vehicles viable on an all-encompassing scale, much less with zero or close-to-zero road deaths.

That's the context of my previous post -- what makes AV research wasteful and impractical compared to investing in better training. I have no expectation whatsoever of changing your mind, but it means you and @Imari are addressing what I said from a different context.

Assuming AVs are as viable as you both believe they are, then I absolutely agree with many of the arguments in favor of AVs. 👍 I don't share that assumption, that's all. And only time will tell. I mainly chipped in to offer a reference for @Dotini's statement, with my $0.02 on the state of driver's education in this country by comparison to investments in autonomous technology.

(insert whynotboth.jpg)

 

[Danoff]

I firmly believe no amount of AI training is going to make autonomous vehicles viable on an all-encompassing scale, much less with zero or close-to-zero road deaths.

Why not?

 

[DDastardly00]

Crappy drivers should use mass transit if they don't want to drive, that would do far more to help alleviate traffic in congested urban areas than self-driving vehicles would.

 

[Danoff]

Crappy drivers should use mass transit if they don't want to drive, that would do far more to help alleviate traffic in congested urban areas than self-driving vehicles would.

They care about not spending an extra 2 hours getting to work, not about driving.

 

[Wolfe]

Why not?

In short, from a lifetime of experience with computers and digital sensors, and from living and working in a rural area with old roads and cold winters. Computers are excellent at certain kinds of tasks. They can pilot a plane in mid-air, till and plant fields, race on a closed course, and probably drive a bus around its urban routes...but taking to the open road -- anywhere, anytime, sharing the road with anything, unsupervised -- seems to me like a prohibitively steep hurdle to clear by comparison.

 

[Danoff]

In short, from a lifetime of experience with computers and digital sensors, and from living and working in a rural area with old roads and cold winters. Computers are excellent at certain kinds of tasks.

Snow will undoubtedly be the very last thing to be tackled by AI driving. Heck I know plenty of human drivers that refuse to drive in snow.

They can pilot a plane in mid-air, till and plant fields, race on a closed course, and probably drive a bus around its urban routes...but taking to the open road -- anywhere, anytime, sharing the road with anything, unsupervised -- seems to me like a prohibitively steep hurdle to clear by comparison.

What's the difference in your mind between driving a bus around urban routes and taking to "the open road"?

 

[SlipZtrEm]

They care about not spending an extra 2 hours getting to work, not about driving.

This is the core truth, and one of the primary reasons public transit just isn't working for most people IMO.

Toronto has a garbage public transit system. If we take it out to the 'burbs to visit family, we budget about an hour. A drive? Maybe 20 minutes unless it's rush hour.

But if I do drive, I'm not basking in the driving experience. Maybe there will be a moment to go full throttle, getting right up to speed and enjoying that shove, but most of it will just be a dull drone from point A to point B. I'm not really getting any enjoyment out of it, just as I wouldn't using the subway. The difference is the time involved. And while it's shorter, there's the added stress of dealing with other drivers that doesn't come up with the TTC.

I think we'd all agree the vast majority of drivers are not interested in the fun of driving, and the majority of their trips are out of necessity. Automating that means removing those that look at driving with disdain from the roads, and freeing everyone who partakes up to spend that commute time doing something else. Automation, in theory, would combine a lot of the perks of public transit with private car ownership.

But...

In short, from a lifetime of experience with computers and digital sensors, and from living and working in a rural area with old roads and cold winters. Computers are excellent at certain kinds of tasks. They can pilot a plane in mid-air, till and plant fields, race on a closed course, and probably drive a bus around its urban routes...but taking to the open road -- anywhere, anytime, sharing the road with anything, unsupervised -- seems to me like a prohibitively steep hurdle to clear by comparison.

Going full autonomous is a massive undertaking, and for it to really work, it'd have to be an all-or-nothing approach. Some cars on the road that are still driven manually could create chaos in the system, and vice-versa.

I maintain autonomy is overall a positive thing — mostly because driving standards are just getting worse — but I believe it's much further in the future than its proponents want us to believe.

What concerns me is the misrepresentation of a lot of the semi-autonomous tech out there right now. It's chiefly Tesla's fault — "Auto Pilot" is anything but — but much like how Kleenex is synonymous with tissue paper, people seem to take driving assists as self-driving modes. I had some people in the RAV4 Hybrid last week to show them it, and they repeatedly referred to the lane-keep assist as "self-driving". They also asked me why I'd ever drive with it off, because "it lets you focus on something else." That's problematic IMO, but also emblematic of what most people want out of a car.

Of course, being in Canada, these systems, at least in their current iterations, are pretty much useless once the snow is on the ground anyway.

 

[DDastardly00]

They care about not spending an extra 2 hours getting to work, not about driving.

2 hours to commute to work on mass transit? I suppose that could happen if you live far enough away or if the mass transit system is designed very poorly, but for most people that live and work in metropolitan areas that simply won't be the case.

In short, from a lifetime of experience with computers and digital sensors, and from living and working in a rural area with old roads and cold winters. Computers are excellent at certain kinds of tasks. They can pilot a plane in mid-air, till and plant fields, race on a closed course, and probably drive a bus around its urban routes...but taking to the open road -- anywhere, anytime, sharing the road with anything, unsupervised -- seems to me like a prohibitively steep hurdle to clear by comparison.

Early studies in DC, specifically during bar hours when mass transit is no longer operating show this very thing, it actually causes more traffic. Autonomous vehicles have a purpose but peddling them as a solution for mass transit in urban areas isn't one of them.


edit:
I suppose I could ask my wife for her opinion on this, since her work for the last few years is directly involved with the technology behind this, she probably knows far more about it than anyone here.

 

[Danoff]

Going full autonomous is a massive undertaking, and for it to really work, it'd have to be an all-or-nothing approach. Some cars on the road that are still driven manually could create chaos in the system, and vice-versa.

I don't think that manually driven cars create any kind of problem for autonomous driving. Especially since the technology is being designed in that environment.

Early studies in DC, specifically during bar hours when mass transit is no longer operating show this very thing, it actually causes more traffic. Autonomous vehicles have a purpose but peddling them as a solution for mass transit isn't one of them.

They'll replace mass transit in two ways. First, they'll operate buses. Second, they'll be more predictable and respond better to fluctuations in traffic, reducing traffic jams. It takes a threshold level of AI driven vehicles to achieve that though, and it helps if they're communicating.

Another way, they'll be more strategic about what time of day they drive large trucks through cities.

Edit:

Another way, they'll be more strategic about which roads they pick based on current traffic patterns.

Edit:

Another way, they won't fall asleep at stoplights because they know that they're first in line and won't miss the light. So when the light turns green, they'll actually go instead of sitting there waiting for the 4th car (not the 2nd one, because that person knows they'll make it too) to honk at them to wake up.

 

[Wolfe]

What's the difference in your mind between driving a bus around urban routes and taking to "the open road"?

Fewer variables in some respects. It's slow driving on relatively slow-moving streets, following a pre-determined route that can be scanned in detail and refined to make it work (ie. Google's methodology). Buses are very visible, and people already give them room, whether by foot, bicycle, or car. They kind of enjoy the right of way by default, so if the AI ever lapses in its duty, it might work out without anyone being harmed or even noticing it happened.

Seems relatively reasonable to me, and it automates a particularly monotonous driving occupation, as you were saying.

 

[SlipZtrEm]

I don't think that manually driven cars create any kind of problem for autonomous driving. Especially since the technology is being designed in that environment.

Eventually, sure. There will be a tipping point, and enough autonomous advancement, that they should be able to blend reasonably well with traditionally-driven vehicles. Especially — or maybe more accurately, only — if the drivers are reliably skilled.

But I just don't want to underestimate the desire to be dickish from those that don't like it. I think of brake-checking or barging one's way through an intersection instead of waiting for one's turn, knowing that the other car will avoid an accident and thus give way.

This seems like an apt comparison here: games. Players tend to find ways to exploit a game for their own needs. I don't doubt some drivers would do the same with autonomous cars once they know they're sharing the road with them.

 

[ROAD_DOGG33J]

What are the chances of human beings heading into a merger with this AI technology in the form of cyborgs?