Why Is Our Sci-Fi So Glum About A.I.?
nytimes.com
nytimes.com
To maintain a post-scarcity society, humans have turned over control of to the Minds, "hyperintelligent machines originally built by biological species which have evolved, redesigned themselves, and become many times more intelligent than their original creators."
Minds manage everything from resource allocation to war planning. They also control all of the technical, day-to-day operations. Still, "the essentially benevolent intentions of Minds towards other Culture citizens is never in question. More than any other beings in the Culture, Minds are the ones faced with interesting ethical dilemmas."
I think much of the explanation is that benevolent A.I. makes for a (often boring) character, whereas hostile A.I. creates conflict.
JARVIS is a sidekick AI. A story where the AI is the main character is either unrealistic (because the AI is too human) or lacks identification (because an AI is not human-like). The first is ok for Hollywood et al. The latter makes for bad stories and I know no examples.
An example that jumps to mind is the sphere from Sphere. And it could be an intelligence, the viewer is left simply not knowing.
I enjoyed Terminator 2 with Arnie as a good AI. Whether it is unrealistic we won't really know till Skynet takes over.
Actually I recommend everything alexanderwales has written. He has a Frozen fanfic, which basically is about the technological singularity via magic: https://www.fanfiction.net/s/10327510/1/A-Bluer-Shade-of-Whi...
I consider the suggestion that Terminator is realistic as trolling. So I will refrain from arguing against it, because there is nobody to convince.
The remaining fields — where no changes to the seed had yet been tested, and no predictions could be made — were classified as indeterminate. Here, the one tried code, the known landmark, was shown as grey against white: a mountain peak protruding through a band of clouds which concealed everything to the east or west of it. No more detail could be resolved from afar; whatever lay beneath the clouds could only be discovered firsthand.
Whenever the conceptory created an orphan, it set all the benignly mutable trait fields to valid codes chosen at random, since there were no parents to mimic or please. Then it selected a thousand indeterminate fields, and treated them in much the same fashion: throwing a thousand quantum dice to choose a random path through terra incognita. Every orphan was an explorer, sent to map uncharted territory.
And every orphan was the uncharted territory itself
http://gregegan.customer.netspace.net.au/DIASPORA/01/Orphano...
The most interesting and evil manifestation of AI I think has to be the Shrike from Dan Simmons' Hyperion and by extension the TechnoCore which created it.
These aren't my own thoughts -- they've come up in a discussion of _Flowers for Algernon_ which I read somewhere and don't seem to be able to cite now. In _Flowers_, the author gives momentum to his eventually super-intelligent protagonist by starting off with severe mental handicapping, then spending a majority of the story at normal intelligence before allowing us to imagine his next state. It is a great writing technique, giving direction and then leaving things to the imagination. Imagination is like a variable that scales the amazingness to fit each reader. I think movies often fail to leave enough space for imagination -- when they're left spelling everything out to the least imaginative member of the audience they leave everyone feeling limited and dissatisfied, glum.
If you want upbeat AI Sci Fi, read a book. Read Banks's _Culture_ series, James P. Hogan's _Two Faces of Tomorrow_, or in a gentler vein, Bradbury's _I Sing the Body Electric_.
He writes a lot, but this seems a good start: http://yudkowsky.net/singularity/intro/
Oy gevalt.
For all the article's comments about the "mindboggling" potential of Moore's law, my word processor looks about the same as it did nineteen years ago, and computers still suck at simple games like Go that, unlike chess, can't be brute forced. I'm grateful to Google for making finding information that bit easier, but I'm even more grateful for humans for creating or curating the content in the first place.
[1]Compare "Do Androids Dream of Electric Sheep" with Edward Bellamy's utopian "Looking Backward". Both versions of the early second millenium are pretty far from the mark (indeed even the technology of the latter book is more arguably more accurate despite it being written in the nineteenth century) but only one of these books is considered to be a riveting read that says profound things about human nature. Similarly, there's a reason some of HG Wells' work is lauded and some of it's laughed at.
[2]Deus Ex Machinas suck and "luckily the AI figured it out" is definitely a loss of imagination and nerve.
I agree with you that conflict is the source of much of compelling story telling, but the question over here is why isn't the AI an ally as opposed to a malevolent entity? This in of itself is very revealing about the nature of our culture. If you contrast the work from the early 1900s to the mid 1900s, as personified by Arthur C. Clarke and Isaac Asimov, then a lot of the stories were based around conflict, but the transgressor was in fact a complex chain of events or a third party entity and the AI was a factor as opposed to something evil that needed to be taken down.
Even for the events in 2001, Clarke painfully explains over the course of the second novel that the AI had been driven insane by conflicting goals imposed by human authorities who should have known better, but didn't. The AI lashed out in madness not because it had malevolent intent. In fact, much of the themes of those books could be summed up in, my friend - this really awesome and earnest robot - and me are out in the vast cosmic unknown and it's desolate, lonely, and I'm one step away from going looney, but luckily my friend - the robot - is here for me.
>>> When it comes to assessing how people might react when blessed with superaugmented intelligences, there are plenty of cautionary examples of people with natural but notable extremes of intelligence who've been tripped up by crippling vulnerabilities. <<<
Except, if you examine this belief then it is a somewhat recent thing. If you go back a few decades, then there was marked optimism about how humanity was going to evolve beyond our very humdrum beginnings into something like the star child Dave Bowman becomes at the end of 2001 - a gift that's bestowed upon him by benevolent alien entities working with a machine.
There are actual surveys which show the rise of this distrust, it is very much an artefact of modern culture as opposed to being something inherent to technology itself.
http://www.ecumenicalnews.com/article/new-study-reveals-a-st...
http://motherboard.vice.com/en_ca/read/it-doesnt-matter-that...
>>> For all the article's comments about the "mindboggling" potential of Moore's law, my word processor looks about the same as it did nineteen years ago, and computers still suck at simple games like Go that, unlike chess, can't be brute forced. I'm grateful to Google for making finding information that bit easier, but I'm even more grateful for humans for creating or curating the content in the first place. <<<
Except now you have this one organisation that has written this magical piece of code that executes in parts across millions of individual computers to search trillions of documents and gives you the reply to any question within milliseconds. This allows you to access all of the world's information on a slab of glass and plastic that is portable and can help you communicate with most of humanity.
At the other end, we finally have control systems that can turn an entire rocket into a robot - allowing it to perfectly hover in mid air and land on its own. A feat that would have been deemed impossible if not magical a few decades ago.
Sure, there are so many things that need to be done, but where we are going is fundamentally beautiful and wonderful. Technology continues to be simply magical, but just not the ways we want it to be. This in of itself says nothing about the technology, just about how we're focussing our attention as a society and as a civilisation.
We can still make magical nuclear reactors and go to mars. There's nothing stopping us, but us. All of those challenges can be surmounted if we push and develop the promise of our tools and technology. We can indeed do another Apollo and the fact that such a thing doesn't exist isn't a fact about our tools, but it's a fact about us.
So all in all colour me puzzled over why technological pessimism is on the rise.
This is a critical point. The danger posed by AI is not that it will actively take over, but that we will put the machines in charge ourselves. The fact that HAL was running the ship in 2001 was a decision made by humans. HAL would not have murdered Frank and locked Dave out of the pod bay, I would argue, if it had not been programmed, at some level of abstraction, to consider itself responsible for the mission. But HAL was a machine, and not capable of taking such responsibility.
Our AIs are already controlling planes and boats. Cars and trucks are at the next corner, we there is little movement for stopping it - hell, it's even hard to find somebody that thinks it's a bad thing (unless they go into "the machines are stealing our jobs" line).
AIs are piloting our planes and boats, and soon our cars and trucks, so that they will go where we want them to go without demanding so much effort and attention from us. They implement the mechanism, but we still set the policy -- and we are still responsible for the consequences.
Now read 'austinl's comment at the top of this page, about a fictional world that cedes its policy decisions to AIs. That, I would argue, is a fundamentally different thing.
I think there is a temptation in human psychology to want to be taken care of, to want someone (or something) else to make the hard decisions. And I think there will be points where we're tempted to turn our power and responsibility over to machines.
I also think that the concern about machines taking over is misplaced: machines, though they may one day be unimaginably more sophisticated than our current ones, will never be conscious or make conscious choices. We will always be able to pull the plug, unless we ourselves set things up so that pulling the plug is too risky.
So the situation I see us in is where concern about a "singularity" that I don't think will ever happen is somewhat distracting us from more important, if rather more mundane, questions about exactly which decisions we want our machines making, and which we should reserve to ourselves. The stock market is a good example here.
The story provides the same situation to both the crew and the computer. The crew sees HAL acting erratically, considers it a threat to the mission and their lives and decide to disable it. HAL sees the crew acting erratically, considers them a threat to the mission and a threat to his life, and decides to disable the crew.
Nobody thinks twice about the crew's decision, but HAL's choice is viewed as insane because of human bias towards the meatbags in the movie.
> human bias towards the meatbags
Interesting phrase. Do you really think people and machines have the same value?
Because SF is always about the present, and most of the population has precious little optimism at the moment.
The interesting thing for me about 50s and 60s SF was how much of it was about humans developing psychic and intellectual superpowers. Through 2001, Dune, and hundreds of short stories you had writers redefining what human minds were capable of.
That changed in the 80s, with smart transcendent machines taking the place of smart transcendent humans.
There were a few evil overlord AIs, but a lot of SF AI went in a transcendent direction, while SF humans were evicted into a universe of parasitic aliens and noir dystopias filled with comic book ultra-violence.
As tech gets closer to real AI it has become more obvious that there's a huge difference between Artificial Intelligence in the traditional comp sci sense, and Artificial Personality, which is something else entirely.
The idea seems to have been that if you get one you'll get the other for free. I think it's becoming clear now that practical AI won't work like that.
Thou shalt not make a machine in the likeness of a human mind! -- the Orange Catholic Bible
I like to think the Butlerian Jihad actually involved a real war against a real UFAI.
Do they really? I'd be very surprised if probabilistic AI methods hadn't made a dent in Go, even a little one.
Side note: there is a really detailed blog about the typography in Moon which was on HN a while back: https://news.ycombinator.com/item?id=7236411
The machine was not the protagonist property, not created by him, and not programmed by him. Why did such a malevolent corporation program the machine to be that benevolent?
Also, I had a lot of sympathy for Samantha in the movie Her. She was such a wonderful companion to him and he was a dick to her. That movie makes me wonder if a strong AI might decide after meeting us that maybe they would be better off without us.
I love stories that upend this myth, especially if they manage to do it without just mindlessly inverting it. Simple inversion usually gives you a story that vilifies the poor and he disadvantaged, whether intentional or not.
Memory-improvement techniques (http://mt.artofmemory.com/wiki/Main_Page) are better used for transdisciplinary creativity, applying specialist knowledge to new usage scenarios.
> A few dedicated hours of dredging the depths, Googling names and then Googling the names those names mentioned led me to a hard clutch of source texts representing the precise gaps in my knowledge I hoped to fill.
Google's search of the public web seems like a magical form of consciousness ... until you use information retrieval software with non-web datasets, e.g. proprietary research, or books.
When collaborating with humans, do we restrict our collaboration to a single human? If not, why restrict our digital collaboration to a single dataset & algorithm?
So I think it's structural.
Until there is a reliable definition of intelligence which addresses the substance that seems to be behind it, which science refuses to address (the soul), there won't be much progress from the sci-fi .. since the solution in the sci-fi world to the problems of all of sciences hole, is thus: there is more to us than intelligence. That's not something that a lot of folks can deal with, alas ..
On the other hand, every new chat AI got instantly challenged and this very place is littered with chat logs anecdotally proving individuals are still more clever than an AI. I mean we must be close, because asserting my parrot is clever provokes nothing similar. I'm sure a real AI wouldn't admit to be more intelligent than humans, because that - currently - doesn't sound like a survival strategy.
Probably it is all about Bayes and whoever accumulated the right information has the advantage.
Sometimes I wonder if our conscious mind is just a puppet and rationalization mechanism for our unconscious decision making process. This makes conscious decision making a farce.
2001 is another case of a movie that was very realistic but most people consider downright boring and confusing. I think its a great movie and while it started the trend of evil AI this article talks about it isn't even close to the worst offender.
And of course this also sells tickets, nobody wants to watch a movie about an AI solving all the problems of the world because in that case nothing happens. A deranged AI with power over UAVs is a great set-up for a summer blockbuster, Watson curing cancer is not.
(Disclaimer: Frozen fan fiction)
Goddamnit LessWrong, just write one damn story with a properly Friendly AI, just for once, just to show it can be done.
That being said, maybe "Friendship is optimal" fanfic (http://www.fimfiction.net/story/62074/friendship-is-optimal) is something that you have in mind, as it does have a properly friendly AI. I'm not just sure if it's a good enough story, for the reasons I stated above.
And an outcome of Friendly AI is all humanity 'living happily ever after' - if we avoid the details of what exactly would 'living happily' look like (since we likely are unable to define that before actually implementing FAI and/or CEV, and the exact result could also extremely vary depending on flavor and degree of [un]Friendliness that is achieved, and none of us know if that will involve ponies), then 'living happily' would quite likely include converting the entire universe to an environment where the 'living happily' happens nonstop for the rest of eternity.
And, unless there's the rather implausible 100% agreement of everyone or an ability to create arbitrary numbers of new, separate universes, yes, Friendly AI would still involve either some people being 'forced' to join or those people being left behind in isolated world[s] by almost everyone else. I see no reason to assume that the maximum possible level of Friendliness can make 100% of people 100% happy 100% of time. Choosing universe-future A prevents universe-future B from happening, and anyone who prefers universe-future B will be 'forced' to accept universe-future A if that is the most Friendly choice by whatever exact definition of Friendliness that happens to get implemented.
The only situation where I think that could be avoided was if we were quick to grant them rights as individuals. And still, we have enough problems getting along with people of other (skin color, age, gender, religion, political denomination, geographic location) that it's very unlikely we'll avoid conflict.
I'd love to be proven wrong, though.
At the moment, what I'm seeing are a lot stories of how we managed to survive being ourselves, and that just barely.
Sci fi itself seems to be a little low on hope.
I find myself sympathizing.
Realistically AI is far more dangerous than presented in any movie. The author did nothing to address real concerns about AI. I suggest starting here: http://intelligence.org/ie-faq/
It's going to be about assimilation. One's life is most affected by one's livelihood and peers. Machine intelligence will obviously change both.
+1 for "Superintelligence: Paths, Dangers, Strategies" by Nick Bostrom.
Palantir is a fantastic example for this, if you go beyond their entire CIA hunting down people origins, the function of Palantir is at its core co-operative. It connects some patterns and offers a search based interaction with a numerous set of databases, and in turn the relies upon a human being to work with and match semantic patterns.
>>> Peter Thiel: Without trying to start a fistfight, we’ll ask Bob: Why is the correct intelligence augmentation, not strong AI?
Bob McGrew: Most successes in AI haven’t been things that pass Turing tests. They’ve been solutions to discrete problems. The self-driving car, for instance, is really cool. But it’s not generally intelligent. Other successes, in things like translation or image processing, have involved enabling people to specify increasingly complex models for the world and then having computers optimize them. In other words, the big successes have all come from gains from trade. People are better than computers at some things, and vice versa.
Intelligence augmentation works because it focuses on conceptual understanding. If there is no existing model for a problem, you have to come up with a concept. Computers are really bad at that. It’d be a terrible idea to build an AI that just finds terrorists. You’d have to make a machine think like a terrorist. We’re probably 20 years away from that. But computers are good at data processing and pattern matching. And people are good at developing conceptual understandings. Put those pieces together and you get the augmentation approach, where gains from trade let you solve problems vertical by vertical.
- http://blakemasters.com/post/24464587112/peter-thiels-cs183-... <<<
Have you read the Bostrom book?
Because people enjoy signaling high status and masculinity derived from personal independence by throwing out cheap put-downs at the very concepts of companionship or family. After all, Aristotle said that one who does not need friends is a beast or a god.
Of course, the beast half of the fork is what usually comes through, and if you claimed godhood I'd have to ask what sort of divinity involves total apathy towards others. Likewise, if you said that you are putting your work towards the cause of human brotherhood or whatnot, I would call it extremely idealistic.
Tell that to the Native Americans after the arrival of Columbus. Or to the North American megafauna after the arrival of humans. Or to the millions of species driven extinct in the Anthropocene.
There's hardly anything more correlated than less-advanced species getting wiped out by a more-advanced newcomer.
> Needless killing and destruction doesn't correlate with intelligence.
Tell that to the Europeans after the arrival of the Mongols. Or to the North American megafauna after the arrival of humans. Or to the millions of species driven extinct in the Anthropocene.
There's hardly anything more correlated than less-advanced species getting wiped out by a more-advanced newcomer.
—Eliezer Yudkowsky, Global Catastrophic Risks p. 333.[1]
Apparently Nick Bostrom's Superintelligence: Paths, Dangers, Strategies[2] does a better job of highlighting the dangers of AI, though I haven't read it yet.
1. http://www.amazon.com/Global-Catastrophic-Risks-Nick-Bostrom...
2. http://www.amazon.com/Superintelligence-Dangers-Strategies-N...