"The Real Threat of AI"
« previous post | next post »
Kai-Fu Lee has an interesting opinion piece in yesterday's NYT: –"The Real Threat of Artificial Intelligence":
What worries you about the coming world of artificial intelligence?
Too often the answer to this question resembles the plot of a sci-fi thriller. People worry that developments in A.I. will bring about the “singularity” — that point in history when A.I. surpasses human intelligence, leading to an unimaginable revolution in human affairs. Or they wonder whether instead of our controlling artificial intelligence, it will control us, turning us, in effect, into cyborgs.
These are interesting issues to contemplate, but they are not pressing. They concern situations that may not arise for hundreds of years, if ever. […]
This doesn’t mean we have nothing to worry about. On the contrary, the A.I. products that now exist are improving faster than most people realize and promise to radically transform our world, not always for the better. They are only tools, not a competing form of intelligence. But they will reshape what work means and how wealth is created, leading to unprecedented economic inequalities and even altering the global balance of power.
Read the whole thing — and then compare it to Norbert Wiener's expression of very similar concerns in 1950, discussed in "AI panics", 11/27/2016, and "Intellectual automation", 3/7/2011.
Wiener's warnings were certainly premature. Kai-Fu Lee has a more plausible case to make, though it's possible that the climb is going to be somewhat steeper and slower than he suggests it will be.
But as both Wiener and Lee explain, the eventual social and political consequences will be profound. Kai-Fu is conditionally optimistic, though his meta-Keynesian prescriptions may strike some as naive:
One way or another, we are going to have to start thinking about how to minimize the looming A.I.-fueled gap between the haves and the have-nots, both within and between nations. Or to put the matter more optimistically: A.I. is presenting us with an opportunity to rethink economic inequality on a global scale. These challenges are too far-ranging in their effects for any nation to isolate itself from the rest of the world.
It's not clear why the global 0.01% should be any more benevolent than it it now — Norbert Wiener put it this way:
Let us remember that the automatic machine, whatever we think of any feelings it may have or may not have, is the precise economic equivalent of slave labor. Any labor which competes with slave labor must accept the economic conditions of slave labor. It is perfectly clear that this will produce an unemployment situation, in comparison with which the present recession and even the depression of the thirties will seem a pleasant joke. This depression will ruin many industries-possibly even the industries which have taken advantage of the new potentialities. However, there is nothing in the industrial tradition which forbids an industrialist to make a sure and quick profit, and to get out before the crash touches him personally.
Thus the new industrial revolution is a two-edged sword. It may be used for the benefit of humanity, but only if humanity survives long enough to enter a period in which such a benefit is possible. It may also be used to destroy humanity, and if it is not used intelligently it can go very far in that direction.
And both current events and recent history suggest that large groups of motivated people with simple weapons are not easily overcome by superior technology. So whatever the ultimate outcome, we may get there after we relive years like 1811, 1848, 1871, 1905, 1917, 1949, 1965, …
bks said,
June 26, 2017 @ 8:40 am
When janitors and fast-food workers are replaced by AI, I'll believe the threat. These are low-skill, hightly repetitive tasks that involve real-world manipulation. Why we still have stockbrokers is a complete mystery, however.
[(myl) Um. And also um.]
David L said,
June 26, 2017 @ 9:29 am
There used to be a strain of science fiction in which the future was imagined as a sparsely populated world, with robots taking care of basic human needs and doing all the dirty work, leaving people to spend their time debating philosophy, playing tennis, making artisanal cheeses, and so on. The human population in these worlds were evidently descendants of today's chardonnay-swilling elites, and what these stories never explained was what happened to the bus drivers, the garbagemen, the soccer hooligans, etc. I think the only writer to tackle this issue was Douglas Adams, who had the telephone sanitizers shipped off to another world.
Mark P said,
June 26, 2017 @ 10:05 am
We saw a report on 60 Minutes on AI just this Sunday. They mentioned a company that had a Go-playing AI. They said it learned the game without any instruction. I thought, "No, it didn't." I'm not aware of any computer program that doesn't consist mostly of a set of instructions. I'm also not aware of any program that does not give deterministic results, at least within the limits of roundoff error and assuming no random number generator (where random=hard to predict). I have run large, complex programs that sometimes give surprising results, but never a result that I could no trace through the logic. Maybe one day a program can be devised that can give truly human-like results, but I suspect that even human behavior is deterministic at some level we are not presently able to detect and understand.
Reporters and companies selling AI systems generally focus on the Wizard but ignore the man behind the curtain. They talk about the amazing things AI can do, but what impresses me is that somewhere there is a group of very clever programmers who enabled the program to do what it does.
Zeppelin said,
June 26, 2017 @ 12:09 pm
Mark P: I'd assume the AI was a neural network type setup? When you let those train themselves they end up so far from human-readable that they're pretty much black boxes. You have to experiment on them afterwards to figure out how they do what they do. That's been my layman's impression, anyway.
Jerry Friedman said,
June 26, 2017 @ 12:17 pm
David L: You might be interested in Frederik Pohl's "Midas World" stories and C. M. Kornbluth's "The Marching Morons".
Mark P: In addition to "hard to predict", you can get true random number generators based on quantum randomness.
Sergey said,
June 26, 2017 @ 12:18 pm
The last four don't seem to be examples of "with simple weapons". Well, maybe the Russian revolution of 1905 was to some degree, although it also included uprisings in the army and navy (such as on the ironclad "Potyomkin"). But the last 3 all had the state-of-the art weapons. The Russian socialist revolution of 1917 was based on the turned military units quartered in St. Petersburg and navy ships in its port. The signal to start the fighting was a shot fired from the cruiser Aurora. The further fighting included plenty of WWI technology in machine guns, armored trains, tanks, airplanes. The Chinese and Vietnamese were supplied with both the latest weapons and the military specialists (from the aircraft pilots and anti-aircraft missile system operators to the infantry officers) by the Soviet Union.
[(myl) There were no VC or North Vietnamese planes or helicopters in the south during the American phases of the war, and no tanks that I'm aware of. During most of the war, the Americans were mostly opposed by local guerrilla forces armed who were relatively lightly armed.
And anyhow in the (hopefully imaginary) future war of the excluded against the AI masters, there will be plenty of semi-advanced weapons lying around in various corners of the world, and it would be imprudent to depend on autonomous drones and tanks to carry the day against them.]
Margaret Wilson said,
June 26, 2017 @ 3:09 pm
I recently re-read Douglas Adams, and was struck by how hateful and snobbish that bit was. As if meaningless mid-level jobs were the invention of the people who end up in those jobs.
Mark S said,
June 26, 2017 @ 7:55 pm
Jerry Friedman: I was thinking of Pohl's story, too. It's actually called 'The Midas Plague', and you can read it here: https://archive.org/stream/galaxymagazine-1954-04/Galaxy_1954_04#page/n7/mode/2up
Jerry Friedman said,
June 26, 2017 @ 9:32 pm
Mark S: Amazon tells me that Midas World is a collection of stories set in that future, including "The Midas Plague". Apparently not all the stories problem take place in the time of limitless production, though.
Andrew Usher said,
June 26, 2017 @ 9:42 pm
David L:
Such a world is very plausible, and you are asking how we'd get from here to there. Well the SF writers that wrote of such utopias may not have wanted to think about it, but clearly at some point there must have been a complete redistribution of wealth. Douglas Adams is of course not entirely serious.
I argue that such redistribution of wealth is now a moral imperative, and that the only reason people don't see it is that minds are clouded by a mess of capitalist ideas that were never entirely defensible, and certainly aren't now with computers (already) able to replace much of human labor, and about to become even more able.- it is foolish and ignorant to imagine that we are 'safe' because one particular path of AI developement doesn't seem promising.
flow said,
June 27, 2017 @ 6:08 am
@Mark P—I think your narrative of what a computer, what a program constitutes is an oversimplification to the point where it becomes 'somewhat thoroughly wrong'. If I had to put my finger on the crucial missing part, I'd say it's what we call 'emergence'; it is also present in the Hegelian discussion of quantity vs quality (https://en.wikipedia.org/wiki/Dialectic#Hegelian_dialectic).
A so-called random number in a modern computer is often the outcome of what is more properly called a PRNG, a pseudo-random number generator; that is essentially a deterministic computation receipe, and as such repeatable and deterministic. As long as you do know all the inputs, which often you don't. It is also commonplace nowadays to feed random generators with figures from truly non-deterministic sources, such as thermal noise or radioactive delay.
But, crucially, even if you do not introduce such sources, the modern computer with its millions upon millions of parts that are engaged in constant interaction presents the programmer with lots of opportunities for the unanticipated, seemingly Heisenbergian incident to occur.
Even more to the point, not only has the mathematics of chaos theory and fractals shown that systems with a sufficient number of interacting parts can cause surprisingly complex behavior, we also have been knowing for quite a while that already the Newtonian three-body problem is not solvable in an analytic fashion, that the double pendulum swings in unpredictable ways, and recently Mr Wolfram was able to show that his utterly simple rule 110 automaton is Turing-complete, from which he draws the conclusion that rather simple molecular-chemical assemblies may indeed perform complex, open-ended calculations.
Closer to home, and more picturesque, are the practical demonstrations of enthusiasts who have maxed-out on that old Computer Science classic, Conway's Game of Life (see e.g. https://www.youtube.com/results?search_query=Conway%27s+Game+of+Life). Here we see demonstrations how an utterly simple cellular automaton that could be easily cast into hardware can produce arbitrarily, highly organized patterns; for example, one guy https://www.youtube.com/watch?v=3NDAZ5g4EuU built a digital clock with a 'macroscopic' Arabic digits display out of gazillions of tiny on/off cells that all operate on a purely local basis, changing state only when their immediate neighbors change state.
Why is that important? Because the double pendulum and Conway's Game of Life are proof that you can build systems that, although composed of simple parts that obey a few straightforward laws, behave in complex, surprising, and hard-to-predict ways. In the case of the Game of Life, you can zoom into and out of the picture and re-play the essentials on a scrap of paper, something hardly possibly nowadays with so-called neural networks (or at least now way I'm aware of).
flow said,
June 27, 2017 @ 6:12 am
oops, 'radioactive delay' -> 'radioactive decay' (although what is measured is really the delay of the decay…). 'now way' -> 'no way'.
*as* said,
June 28, 2017 @ 8:16 am
Zeppelin: "Mark P: I'd assume the AI was a neural network type setup? When you let those train themselves they end up so far from human-readable that they're pretty much black boxes. You have to experiment on them afterwards to figure out how they do what they do. That's been my layman's impression, anyway."
Here's a pertinent (and nicely illustrated) article in MIT Technology Review:
https://www.technologyreview.com/s/604087/the-dark-secret-at-the-heart-of-ai/
("No one really knows how the most advanced algorithms do what they do. That could be a problem.")
Mark P said,
June 28, 2017 @ 9:57 am
I extracted one, simple issue with AI, that computers are inherently intended to produce deterministic results in the absence of some kind of randomness. I sort of equated randomness with unpredictability, and sort of assumed that one characteristic of human intelligence is unpredictability. I was wondering, if that is the case, can a completely predictable, deterministic process produce something we would call intelligence. I understand that a deterministic process can produce results that are very hard to reproduce by following the logic of a program, in the same way that writing down every possible game of chess might be possible but undoable in actuality. But, in principle, it can be done. However, I'm also not certain that unpredictability is a defining characteristic of intelligence. In fact, the direct opposite might be the case. In any event, it's too easy to get off into the weeds in a discussion like this.
Mark P said,
June 28, 2017 @ 9:57 am
I extracted one, simple issue with AI, that computers are inherently intended to produce deterministic results in the absence of some kind of randomness. I sort of equated randomness with unpredictability, and sort of assumed that one characteristic of human intelligence is unpredictability. I was wondering, if that is the case, can a completely predictable, deterministic process produce something we would call intelligence. I understand that a deterministic process can produce results that are very hard to reproduce by following the logic of a program, in the same way that writing down every possible game of chess might be possible but undoable in actuality. But, in principle, it can be done. However, I'm also not certain that unpredictability is a defining characteristic of intelligence. In fact, the direct opposite might be the case. In any event, it's too easy to get off into the weeds in a discussion like this.
Rodger C said,
June 28, 2017 @ 11:58 am
@Mark P: The blog you want for that is Conscious Entities.
Terry Hunt said,
June 28, 2017 @ 10:03 pm
@ Margaret Wilson: I think you missed some of Douglas's point – the Golgafrinchams who shipped off what they thought was the "entire useless third of their population" in the "B" Ark, the telephone sanatizers and so on, were subsequently themselves "suddenly wiped out by a virulent disease contracted from a dirty telephone."
Barbara Phillips Long said,
June 29, 2017 @ 9:48 pm
Based on today's struggles with my iPad, none of the science fiction cited plausibly predicts the future. Right now, I would expect AI to work toward ever more effective baiting about "you have won a free x," countered by AI that is designing filters to screen out bogus free offers, existing in parallel with AI trying to find entertainment, factual information, patentable procedures, or political advantage for particular audiences. I have no clue what currency will be used. I find it believable, however, that parents might take up the practice of assigning a name to each child that is whispered in their ear but not documented because identity theft becomes so common that privacy requires frequent resets of official documentation, probably done through AI.
ajay said,
June 30, 2017 @ 4:46 am
I was wondering, if that is the case, can a completely predictable, deterministic process produce something we would call intelligence. I understand that a deterministic process can produce results that are very hard to reproduce by following the logic of a program, in the same way that writing down every possible game of chess might be possible but undoable in actuality. But, in principle, it can be done.
I am finding it hard to think of a reason why, in that sense, a human brain isn't predictable and deterministic but a highly sophisticated computer is.
bks said,
June 30, 2017 @ 8:44 am
We are still waiting for the robot revolution
https://www.ft.com/content/4423a404-5c0f-11e7-9bc8-8055f264aa8b