One thing I often see repeated is that every new industrial technology initially destroyed some jobs but eventually created a lot of new ones: the cotton gin, the steam engine, the car, etc. That's because these things only did one thing, and by doing that one thing really well they opened a lot of side-niches. It took a lot of time to invent and scale out new machines, so for long periods of time these side-niches would be available to people.
I completely agree with Sam here. I think the fallacy in the above argument is that there is a qualitative difference between Turing-complete machines and special-purpose machines. Turing-complete mechanization is broad and endlessly adaptable.
Programmable machines aren't machines. They're machine-machines, and can be adapted to new tasks in short linear time by small numbers of people with little capital. That makes it "different this time."
I also think that malicious AI is already kind of here, but in a hybrid "cyborg" form. It's the corporation. Corporations that destroy human livelihoods and abuse human beings in general to maximize per-quarter shareholder returns are a bit like "paperclip maximizers."
The danger is not in some Terminator-like AI apocalypse, but that incremental advances in AI will make these things progressively less and less human and more and more machine. I can imagine a future almost-entirely-silicon financial corporation that uses its speed and superior analytical intellect (at least in the financial ___domain) to lay waste to entire national economies in order to maximize shareholder value... i.e. paperclips. Since this would likely be found in the hedge fund world, nearly all of this siphoned-off wealth would be captured by a small number of already very rich people.
Nightmare AI wouldn't be much like Skynet -- a new being pursuing its own self-interest. It would be more like a very, very smart dog helping its elite owners "hunt" the rest of us in the financial sphere. This could fuel even more massive consolidation of financial wealth. We are already seeing the beginning of this with algorithmic quant finance.
In a thread on Twitter I also heard someone bring up "AI assisted demagoguery," a notion I found to be total nightmare fuel. Imagine a Hitler wannabe with a massive text-comprehending propaganda-churning apparatus able to leverage the massive data sets available via things like the Twitter and Facebook feeds to engage in high-resolution persuasion of millions and millions of people. The thing that makes this scary is that populist demagoguery gets more appealing when you have things like massive wealth inequality.
You can make counter-arguments here, but I also agree with Sam that it is foolish to just hand-wave these kinds of possibilities away. We should be thinking about them, and about how -- as he puts it -- we can find ways to channel this trend in more positive directions.
Agree that earlier advancements created a lot of new ones. But do you think that lack of globalization in previous instances had a very significant part to play? Especially where and how the cost savings achieved through automation were invested back? (honest question)
I do think that lack of globalization made it easier for societies to achieve good resolutions to internal labor disputes-- globalization prevents employers from going outside a nation's socioeconomic framework to break the negotiating power of employees. But I think this is a linear term in the equation, not an exponential one. Exponential effects always dominate linear ones.
I'd argue from the stance that globalization has played a role in every major revolution, from agriculture to the silk road to colonization to industrialization to the computer age. They were all enabled by global trade, driving demand for foreign goods, prompting responses that produced innovations that swept whomever was at the forefront of the world in decades.
The size of the world changes - and gets broader - but the mechanism at play, that the concentrated powers of the world fuel each others innovations and those innovations catapult advancement but also introduce huge power vacuums between those nations adopting the new and those nations outside the sphere of influence - something I would not say was missing in the computer revolution, since the adoption of computer technology happened first and is still only pervasive in first world nations. The third world is still late to the party and comes in with fractured infrastructure and access, where systemic access and ubiquity enabled the Internet and a lot of the current revolutions in the first place.
In previous instances, the cost savings made the nations that had them superpowers in their times. The cotton gin is a huge part of why America went from an English vassal to a world power - between the innovations and the raw resources of the Americas, it could propagate empire.
But those profits just made men rich. The cotton, tobacco, industrial, automotive, oil, etc barons were the kings of their times through the innovations and automations of their industries. That has never really changed, those ruling over the industries being modernized always reap unfathomable wealth and power from the enterprise. Their wealth made their country rich, but the laborers still had to find something else to do every time, and up until now we have always had some unskilled menial and physical thing to have most people do. In actuality, we ran into that wall probably seventy years ago in the aftermath of World War 2 - as the US at least rapidly adopted women and minorities into the whole workforce the huge surge of productive labor combined with the green revolution, the reforms of the late industrial giving workers reasonable hours, unions, and power over their lives, plus the fledgling technological revolution that had already produced a lot of wonders (consumer refrigeration, microwaves, clothes washers and dryers, etc) had already crippled the low skill labor market and we collectively adapted by organically injecting superfluous bureaucracy in almost every business and part of life to make up for the work shortage.
Problem is that we did that, became a "service" economy, and are now faced with the obsolescence of busy office work. I can just remember CGPGrey talking about it in Humans Need Not Apply, in how the prime target for automation is not the McDonalds burger flipper but every middle income office worker who can be replaced by software. After bureaucracy, where do we inject the overflow labor of humanity? Or do we finally admit we don't need everyone laboring?
I completely agree with Sam here. I think the fallacy in the above argument is that there is a qualitative difference between Turing-complete machines and special-purpose machines. Turing-complete mechanization is broad and endlessly adaptable.
Programmable machines aren't machines. They're machine-machines, and can be adapted to new tasks in short linear time by small numbers of people with little capital. That makes it "different this time."
I also think that malicious AI is already kind of here, but in a hybrid "cyborg" form. It's the corporation. Corporations that destroy human livelihoods and abuse human beings in general to maximize per-quarter shareholder returns are a bit like "paperclip maximizers."
http://wiki.lesswrong.com/wiki/Paperclip_maximizer
The danger is not in some Terminator-like AI apocalypse, but that incremental advances in AI will make these things progressively less and less human and more and more machine. I can imagine a future almost-entirely-silicon financial corporation that uses its speed and superior analytical intellect (at least in the financial ___domain) to lay waste to entire national economies in order to maximize shareholder value... i.e. paperclips. Since this would likely be found in the hedge fund world, nearly all of this siphoned-off wealth would be captured by a small number of already very rich people.
Nightmare AI wouldn't be much like Skynet -- a new being pursuing its own self-interest. It would be more like a very, very smart dog helping its elite owners "hunt" the rest of us in the financial sphere. This could fuel even more massive consolidation of financial wealth. We are already seeing the beginning of this with algorithmic quant finance.
In a thread on Twitter I also heard someone bring up "AI assisted demagoguery," a notion I found to be total nightmare fuel. Imagine a Hitler wannabe with a massive text-comprehending propaganda-churning apparatus able to leverage the massive data sets available via things like the Twitter and Facebook feeds to engage in high-resolution persuasion of millions and millions of people. The thing that makes this scary is that populist demagoguery gets more appealing when you have things like massive wealth inequality.
You can make counter-arguments here, but I also agree with Sam that it is foolish to just hand-wave these kinds of possibilities away. We should be thinking about them, and about how -- as he puts it -- we can find ways to channel this trend in more positive directions.