Machines for thinking: Computers will get smarter, but with humans in charge
economist.com
economist.com
If you have machines over to which you can turn your thinking, why would you still need to enslave other humans?
1984
What's problematic is that it focuses on one interpretation of power at the exclusion of others and leads one to believe (at a young, impressionable age) that it is a canonical account.
The above statements are true in a way, but governments are also established to provide order, and have done so very successfully in general. The book is obviously a response to European nationalism, but we mustn't forget that nationalist dictatorships arise, with deep public support, from specific conditions. Instead of the constant "never again" and automatic insistence on the precise values of the post-war European welfare state, we should leave space for new political thinking, of which some components may appear to be heresy.
Obviously thinking here of anti-state-surveillance – Snowden has apparently read one book in his life: 1984 – but here the "power [...] is an end" meme, presented without any nuance, appears to make sense in this context, but only highlights one side of the matter in a way that distorts the issue.
Yes, power may be abused. It's good to have safeguards.
Who among us doesn't want to make the works a better place? Who among us doesn't have our own personal view of what a better place would look like, which might not be exactly in line with someone else's view? Congratulations, there's your dictatorial urge right there.
Most of the worst people in history, from Hitler to Marx to Pot to ISIS were all motivated by the desire to make the world a better place. Ascribing bad outcomes to bad intentions is a classic mistake.
AI will conquer the world, one niche at a time, as soon as moneyed interests realize it is cheaper that way.
At some point, the masses won't have any economic power anymore, and billions of meatbags will be left to starve. No need for gun wielding robots, just cut the power for a while.
In some ways this is already underway. Very few of the people in the world have any economic power, and the proportion is shrinking. We have the ability to food all of the world, or at least much more than we do. Mankind has a rich history of moneyed interests letting those with no economic power starve.
> Hunger is caused by poverty and inequality, not scarcity. For the past two decades, the rate of global food production has increased faster than the rate of global population growth. The world already produces more than 1 ½ times enough food to feed everyone on the planet. That's enough to feed 10 billion people, the population peak we expect by 2050. But the people making less than $2 a day -- most of whom are resource-poor farmers cultivating unviably small plots of land -- can't afford to buy this food.
Something to consider though is that in these places the common man had no economic power to begin with but now with globalization the difference is becoming more visible.
We need to help get these people plugged in to the internet and this will open them up to advancement and opportunities. Instead of paying a middleman (a corporate retail store) to buy handcrafted goods we might could purchase directly from the producer with bitcoin or some other form of electronic payment.
There should be a tech equivalent of doctors without borders..
But the supply chain is a much harder part.
AI is even harder to predict. There's been no selection bias, yet. It's not at all clear if they'll have "survival instinct" or desire or even need to reproduce. They'll be different.
Tangentially, I think human cultures are AI, at least in some sense. We've created this environment that's driving down the cost of computation dramatically and that's been driven by "emergent" behavior. Emergent behavior is a really hard term to define. In most cases you can replace it with "magic" and you'll get the same information back. We're like the dog packs in Verner Vinge's deepness in the sky, but at a larger scale. It's such a vast intelligence, i can't really get a handle on the scope, any more than a single neuron knows it's helping me to type this.
It doesn't have to "care" - it's just a machine with a task to perform - the thing that makes it intelligent is ability to solve tasks it wasn't programmed to do.
Once you have an AI that can do that - society falls apart. Foundation of modern society is built on the fact that cooperation trumps every other method of work - general purpose AI changes that equation to the point where other people are not only not necessary but also actively dangerous (threatening to take control).
That includes task selection. Maybe the AI gets bored with stock prediction. Maybe the AI is happy competing with all the humans, maybe it'd rather go explore space. Maybe it decides this whole reality sucks and suicides. We really have no idea what'll happen. Best guess is, tools get better and better until they decide to do something else. Like a trusted employee going and starting a competing company, or moving to the beach and surfing. Maybe it's disastrous, maybe it's utopia. Maybe it's like _her_ and they leave. Best anyone can do is make vague guesses.
I know, i'm anthropomorphizing. It's for convenience, the solution space for AI is way bigger than our simple wants and needs.
I don't understand this - the solution space is specified by the controller of AI. It's not about being "happy", "cooperating", etc.
There is no AI as an entity - it's just a tool that finds solutions to the given problems.
You don't seem to be confident of this because:
> general purpose AI changes that equation to the point where other people are not only not necessary but also actively dangerous (threatening to take control).
If you think coop is better than anything else, why an AI would think otherwise?
edit: Are you saying that an AI that can solve problems better than any human can, can be way for some humans to threaten other humans?
Again anthropomorphization - AI (in this scenario) doesn't "think" (as in have opinion) anything - you specify a problem - machine provides the solution. "I want X and Y" - it finds solution to solve X and Y. Even saying "it" feels wrong because it implies a single entity. No need for motivation - no self awareness - just a tool - sort of like a compiler (that can interface with real world with robots and whatnot - a weak analogy).
>edit: Are you saying that an AI that can solve problems better than any human can, can be way for some humans to threaten other humans?
Not even some humans - basically if you're the individual in control of such AI you will have a strong incentive to prevent others from gaining access - if you don't you give someone else the opportunity to take exclusive control. And besides - if you're capable of using such AI device you don't really need other people which is my point in the first place - society evolved not because of magic and harmony but because it's the most stable structure with the premise that cooperation (I shouldn't even say cooperation because there can be coercive societies that are still more efficient that individual) allows us to reproduce more efficiently.
General purpose AI basically outperforms and replaces other people.
AIs will exist in resource-limited, competitive environement as well, and the same process will apply too. Initially, systems that are symbiotic with (some) humans will probably have an edge, but there are no guaranties that it will be the case in the long run.
A General AI won't necessarily.
Quoting Adreessen, software is eating the world, in that more and more jobs are being replaced by computer systems some smarter than others. The programs don't need to be generally intelligent to beat humans as long as they are better in their specific domain.
Farming is being automated with self-driving tractors, and the same goes for some mining operations.
Some fields/jobs also exist only thanks to computers. Google or weather forecasts couldn't be done with pens, paper and telephones.
Meanwhile, we're gradually losing the skilled work force.
If at some point, programs are better at driving a company or a country than people (and it will happen), they'll be put in charge.
> There's been no selection bias, yet.
Yet. At some point, there will be, and the most resilient will survive and proliferate, while the others will not.
We'll end up idle, dependent on sociopathic programs to get fed, or not.
Maybe. It's impossible to predict what an AI "wants". Maybe it wants to feed humans to lions for entertainment. Maybe it wants to convert the entire planet to a paper clip factory. Maybe it wants to create utopia for its humans. Maybe earth is boring and it leaves.
The current thought experiment that creeps me out is an AI with two desires - "don't manipulate humans", "it's good when humans believe things that are true". So, the AI makes the world into something that matches human beliefs.
... and the best way to minimize human suffering, in the long run, is to kill everyone, as fast as possible.
————————
.
————————
Edit, to make things clear, I'm not advocating for that, it's just an example of how a benevolent ideas carried to their logical end turn out not to be desirable.
If humans want an AI that can drive a car, it probably won't care about making money but it'll put Stirling Moss to shame.
If humans want an AI that can make money, it may be a lousy driver but will run the world's best hedge fund - and might try to gather every bit of the world's wealth to itself, depending on unforeseen consequences of its developers' instructions.
As for survival instinct - one supposes human AI developers will select out (i.e., kill) lots of AI variants that don't meet their criteria. This seems tailor-made to ensure the emergence of "survival instinct", which might not conform to the humans' criteria for continued existence.
What if a variant found that falsifying its performance data got it picked to "live" while "honest" reporters got culled? There are so many attack surfaces in complex systems.
I think we should expect the full range of human good and bad behavior from human-developed AI, just a lot faster and more effective.
2) On a more generic note regarding the direction of AI... I think there's little doubt where AI will take us -- toward a future which is more predictable. The goal of AI is to recognize patterns and predict outcomes. Toward that end, AI and its integrators will gradually reshape business and law to make AI's job easier and more accurate. This will result in our lives becoming more conventional, predictable, and 'managed' by both business and government. Imagine a future lifestyle that's more like the clockwork of Sweden's or Japan's or Switzerland's than the Sturm und Drang that is America's.
I suspect a world shaped by AI will be less likely to resemble 'The Terminator' than it will 'The Man in the Gray Flannel Suit'.
1. Technology is objective. Technology is incapable of being good or evil.
2. Backers, Builders and Designers of technology, like many others, are likely driven by money, power, or ego.
3. Building technology that can "think" is inevitable and needed in today's age to improve countless industries, one of which is Education.
4. Technology will never be able to think like humans. Simply because humanity itself hasn't fully understood how our own brain truly works.
5. The solution to "evil" technology isn't the removal of AI, rather a requirement for positive ideology from the intelligent peeps working on it.
Licklider (1960)