AI panics
« previous post | next post »
The last month or so has seen renewed discussion of the benefits and dangers of artificial intelligence, sparked by Stephen Hawking's speech at the opening of the Leverhulme Centre for the Future of Intelligence at Cambridge University. In that context, it may be worthwhile to point again to the earliest explicit and credible AI warning that I know of, namely Norbert Wiener's 1950 book The Human Use of Human Beings [emphasis added]:
[T]he machine plays no favorites between manual labor and white-collar labor. Thus the possible fields into which the new industrial revolution is likely to penetrate are very extensive, and include all labor performing judgments of a low level, in much the same way as the displaced labor of the earlier industrial revolution included every aspect of human power. […]
The introduction of the new devices and the dates at which they are to be expected are, of course, largely economic matters, on which I am not an expert. Short of any violent political changes or another great war, I should give a rough estimate that it will take the new tools ten to twenty years to come into their own. […]
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.
Steve Levinson and I reprised these concerns — with a nod to Wiener — in our 1981 Scientific American article on speech recognition. And in "Intellectual automation", 3/7/2011, I wondered whether the long-predicted impact was finally on the horizon:
Perhaps machine learning, AI, and computational linguistics will finally bring about the long-predicted migration from white-collar cubicle farms to the urban reservoirs of unemployment. But history suggests that it would be prudent to wait and see. Farming delivers a relatively well-defined and consistent product: a bushel of wheat is a bushel of wheat, however radically the process of creating and delivering it changes. The services that (for example) lawyers deliver are more abstract, and technological innovation (whether in the form of typewriters, copy machines, word processors, document retrieval, or document-understanding systems) is likely to lead to changes in the product as well as changes in the way the product is created and delivered.
As Yogi Berra said, "It's tough to make predictions, especially about the future".
Anyhow, here's a longer sequence of quotations from Wiener 1950:
It is the thesis of this book that society can only be understood through a study of the messages and the communication facilities which belong to it; and that in the future development of these messages and communication facilities, messages between man and machines, between machines and man, and between machine and machine, are destined to play an ever increasing part. […]
The older machines, and in particular the older attempts to produce automata, did in fact function on a closed clockwork basis. But modern automatic machines such as the controlled missile, the proximity fuse, the automatic door opener, the control apparatus for a chemical factory, and the rest of the modern armory of automatic machines which perform military or industrial functions, possess sense organs; that is, receptors for messages coming from the outside. […]
The machine which acts on the external world by means of messages is also familiar. The automatic photoelectric door opener is known to every person who has passed through the Pennsylvania Station in New York, and is used in many other buildings as well. […] The steps between the actuation of a machine of this type by sense organs and its performance of a task may be as simple as in the case of the electric door; or it may be in fact of any desired degree of complexity within the limits of our engineering techniques. A complex action is one in which the data introduced, which we call the input, to obtain an effect on the outer world, which we call the output, may involve a large number of combinations. […]
It is my thesis that the physical functioning of the living individual and the operation of some of the newer communication machines are precisely parallel in their analogous attempts to control entropy through feedback. Both of them have sensory receptors as one stage in their cycle of operation: that is, in both of them there exists a special apparatus for collecting information from the outer world at low energy levels, and for making it available in the operation of the individual or of the machine. In both cases these external messages are not taken neat, but through the internal transforming powers of the apparatus, whether it be alive or dead. […]
[E]xcept for a considerable number of isolated examples, the industrial revolution up to the present has displaced man and the beast as a source of power, without making any great impression on other human functions. The best that a pick-and-shovel worker can do to make a living at the present time is to act as a sort of gleaner after the bulldozer. In all important respects, the man who has nothing but his physical power to sell has nothing to sell which it is worth anyone's money to buy.
Let us now go on to a picture of a more completely automatic age. Let us consider what for example the automobile factory of the future will be like; and in particular the assembly line, which is the part of the automobile factory that employs the most labor. In the first place, the sequence of operations will be controlled by something like a modern high-speed computing machine. computing machine. In this book and elsewhere, I have often said that the high-speed computing machine is primarily a logical machine, which confronts different propositions with one another and draws some of their consequences. It is possible to translate the whole of mathematics into the performance of a sequence of purely logical tasks. If this representation of mathematics is embodied in the machine, the machine will be a computing machine in the ordinary sense. However, such a computing machine, besides accomplishing ordinary mathematical tasks, will be able to undertake the logical task of channeling a series of orders concerning mathematical operations. Therefore, as present high-speed computing machines in fact do, it will contain at least one large assembly which is purely logical.
The instructions to such a machine, and here too I am speaking of present practice, are given by what we have called a taping. The orders given the machine may be fed into it by a taping which is completely predetermined. It is also possible that the actual contingencies met in the performance of the machine may be handed over as a basis of further regulation to a new control tape constructed by the machine itself, or to a modification of the old one. I have already explained how I think such processes are related to learning.
It may be thought that the present great expense of computing machines bars them from use in industrial processes; and furthermore that the delicacy of the work needed in their construction and the variability of their functions precludes the use of the methods of mass production in constructing them. Neither of these charges is correct. […]
The computing machine represents the center of the automatic factory, but it will never be the whole factory. On the one hand, it receives its detailed instructions from elements of the nature of sense organs, such as photoelectric cells, condensers for the reading of the thickness of a web of paper, thermometers, hydrogen ion-concentration meters, and the general run of apparatus now built by instrument companies for the manual control of industrial processes. These instruments are already built to report electrically at remote stations. All they need to enable them to introduce their information into an automatic high-speed computer is a reading apparatus which will translate position or scale into a pattern of consecutive digits. Such apparatus already exists, and offers no great difficulty, either of principle or of constructional detail. The sense-organ problem is not new, and it is already effectively solved. […]
[T]he machine plays no favorites between manual labor and white-collar labor. Thus the possible fields into which the new industrial revolution is likely to penetrate are very extensive, and include all labor performing judgments of a low level, in much the same way as the displaced labor of the earlier industrial revolution included every aspect of human power. […]
The introduction of the new devices and the dates at which they are to be expected are, of course, largely economic matters, on which I am not an expert. Short of any violent political changes or another great war, I should give a rough estimate that it will take the new tools ten to twenty years to come into their own. […]
[The intermediate period of the introduction of the new means, especially if it comes in the fulminating manner to be expected from a new war, will lead to an immediate transitional period of disastrous confusion. We have a good deal of experience as to how the industrialists regard a new industrial potential. Their whole propaganda is to the effect that it must not be considered as the business of the government but must be left open to whatever entrepreneurs wish to invest money in it. We also know that they have very few inhibitions when it comes to taking all the profit out of an industry that there is to be taken, and then letting the public pick up the pieces. This is the history of the lumber and mining industries, and is part of what we have called in another chapter the traditional American philosophy of progress. […]
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.
Victor Mair said,
November 27, 2016 @ 9:33 am
"Short of any violent political changes or another great war, I should give a rough estimate that it will take the new tools ten to twenty years to come into their own."
Wiener wrote that in 1950. Since there have been no "violent political changes or another great war", the "new tools" should have long since "come into their own". But in what sense and to what degree have they done so?
[(myl) Wiener's radical underestimate of the time scale is exactly the point of the quotation — thanks for making it explicit.
Wiener seriously underestimated the difficulty of pattern recognition, of robotic control for complex mechanisms, and of integrating the two. Considerable progress has been made in those areas but there are still unsolved problems. He also underestimated the difficulty of automatic speech recognition and text analysis.
In my opinion, current prognosticators tend to similarly underestimate the difficulty of human-like communicative interaction. It's relatively easy to give the impression of solving the problem (Eliza, Siri) without really even trying to solve it.]
MattF said,
November 27, 2016 @ 10:21 am
I use that quotation about predicting the future quite a bit. It seems, though, that the attribution to Mr. Berra is bogus:
http://quoteinvestigator.com/2013/10/20/no-predict/
Coby Lubliner said,
November 27, 2016 @ 12:28 pm
The "future" quote has been investigated here.
chris said,
November 27, 2016 @ 1:06 pm
[T]he machine plays no favorites between manual labor and white-collar labor.
It turns out that this is sort of correct, but not quite the way Wiener appears to have meant it: significant parts of both manual *and* white-collar labor are not as easy to mechanize as might have been hoped. Lots of things that humans take for granted that other humans can do without any particular effort or training turn out to be quite complicated when you don't have human instincts (and a human's childhood doubles as training in a lot of things so commonplace they aren't even necessarily thought of as skills, until you have to deal with a machine that lacks them).
One human with a backhoe can replace 20 with shovels, but you still have a surprisingly hard time getting rid of that last human. Ditto for copy machines vs. scriveners.
Bloix said,
November 27, 2016 @ 1:17 pm
The automobile is a threat to the employment of stable boys and street sweepers everywhere! The washing machine will take the jobs of laundry girls! Oil heat will put the chimney sweeps out of work! The telephone means the end of messenger boys! Electric-powered carts means no work for boys in coal mines! Running water and flush toilets will deprive young housemaids of employment!
Funny how we don't miss child labor nowadays.
Bloix said,
November 27, 2016 @ 2:06 pm
"predictions about the future"
The attribution to Yogi Berra was always clearly a phony (as are attributions to Mark Twain and Sam Goldwyn). The more interesting attribution is to Niels Bohr, and it makes some sort of sense attached to him because of his association with quantum mechanics, which asserts that sub-atomic events cannot be predicted but only given a probability of occurring.
Bohr might have said it, but if he did he was quoting an earlier Danish source, as noted in the link provided by Corey Lubliner. According to Fred Shapiro in the Yale Book of Quotations, the original Danish is a pun, but he doesn't explain what the pun is.
https://books.google.com/books?id=ck6bXqt5shkC&pg=PA92&lpg=PA92&dq=danish+predictions+about+the+future+pun&source=bl&ots=oCg4VFo71h&sig=u7WET_WJCDHJzy9bf-rkpRmGa64&hl=en&sa=X&ved=0ahUKEwjVwuyI0MnQAhXq5YMKHYbJCq4Q6AEILzAD#v=onepage&q=danish%20predictions%20about%20the%20future%20pun&f=false
peter said,
November 27, 2016 @ 5:31 pm
It took a century for printed books to be as error-free as hand-copied books. This was because hand-copying had evolved over centuries elaborate supporting systems and processes of training, proof-reading, and supervision of the people doing the copying. Printing of books, at least initially, had no such supporting quality-control systems, and the quality of printed books was consequently very poor.
While robots may eventually take all our jobs, it will be some time before we have in place all the supporting systems needed for them to equal humans in effectiveness in any but the simplest tasks. As the introduction of robots to Wedgewood's pottery factories in England a decade ago showed, even stacking plates so that the stack does not fall over, took some months of continual robot re-calibration.
Harold said,
November 27, 2016 @ 6:00 pm
@Bloix Unbelievably enough, there actually are people in the Foundation for Economic Education who think we should bring back child labor: https://fee.org/articles/let-the-kids-work/
Gunnar H said,
November 28, 2016 @ 3:02 am
@Bloix:
The claim that the Danish version of "It is difficult to make predictions, especially about the future" is a pun appears to be from a mistranslation of sprogblomst (lit. "language flower"), which means a play on words in a wider sense – that is, intentional or unintentional use of language to striking effect (good or bad). The QI translation "howler" is better in this context.
Joe Econ said,
November 28, 2016 @ 8:41 am
Predictions about disastrous effects of the replacement of human labor have been wrong for ages, not necessarily because of flawed estimates of the time tables of technological change or of the difficulty of solving certain technological problems, but because of economic misunderstanding. Jobs that are replaced by mechanization, automation, or artificial intelligence are always balanced to some extent by jobs in new industries that previously did not exist and are created by the very technologies that replaced the old jobs. It's not simply a question of human labor replacement when other economic opportunities are always created simultaneously. Does this change create disruption? Most definitely. But the technologies that cause the disruption don't come from nowhere. Someone has to produce, distribute, repair, sell, optimize, and train people to use them–all new economic opportunities that previously did not exist.
Bill Benzon said,
November 28, 2016 @ 10:53 am
Wiener seriously underestimated the difficulty of pattern recognition, of robotic control for complex mechanisms, and of integrating the two. Considerable progress has been made in those areas but there are still unsolved problems. He also underestimated the difficulty of based speech recognition and text analysis.
In my opinion, current prognosticators tend to similarly underestimate the difficulty of human-like communicative interaction. It's relatively easy to give the impression of solving the problem (Eliza, Siri) without really even trying to solve it.
Yes. And of course he's not the only one.
Back in the mid-1970s I was studying computational semantics with David Hays. Every now and then I would ask him, When do you think we'll be able to do X? where X ranged over various interesting things one might want of linguistic computing. He always refused to answer, asserting that these things are deeply unpredictable. Remember, he'd been on the committee that wrote the ALPAC report. He had practical experience in such things.
In 1975 he got invited to review the computational linguistics literature for the journal, Computers and the Humanities. He asked me to draft the text (as I'd been reviewing the literature for the American Journal of Computational Linguistics). I did so and included a bit about an article about computational semantics I was publishing in MLN (Modern Language Notes), as it spoke directly to humanist concerns and included an analysis of a Shakespeare sonnet. We then floated, as a thought experiment, the idea of a computational system capable of reading a Shakespeare play, in some interesting, but unspecified, sense of the the word 'reading.' We called it Prospero and set no date on when Prospero would be operational, but in my mind I figured we'd have it in 20 years or so.
Well, the article appeared in 1976 ("Computational Linguistics and the Humanist"). Add 20 to that and we have 1996. Was anything like Prospero available then? No. Not only that, but the symbolic computing that was at the center of our review, and of Prospero, was being pushed into the background by statistical methods. It's now 2016, 40 years after that paper. We don't have anything like Prospero now – though I believe Patrick Henry Winston is using the Macbeth story (but not Shakespeare's play) in an investigation of story comprehension – and I see no prospects for Prospero in the near future. And yet, by the practical standards of 1976 Siri is a miracle, as is Google's translation tech, and self-driving vehicles. Etc.
It's a brave new world that has such machines in it, and most of it is still unexplored.
ajay said,
November 28, 2016 @ 11:03 am
"Short of any violent political changes or another great war, I should give a rough estimate that it will take the new tools ten to twenty years to come into their own."
Wiener wrote that in 1950.
Yes, and he wasn't that far off. Look at the example he picks – " Let us consider what for example the automobile factory of the future will be like; and in particular the assembly line, which is the part of the automobile factory that employs the most labor" – well, GM installed its first robot on the assembly line in 1961 and industrial robotics really took off in the early 1970s. By the end of that decade Fiat was marketing its Strada as "designed by computer, built by robots".
http://www.autoexpress.co.uk/car-news/61495/9-fiat-strada-built-robots
Watch the video. That isn't CGI or a studio shoot, that is actually the Fiat line, in 1979.
Victor Mair said,
November 28, 2016 @ 7:21 pm
"A single chart everybody needs to look at before Trump’s big fight over bringing back American jobs" (WP, 11/28/16)
The economics are unavoidable and irreversible. Although a human welder may earn $25 an hour, a robot welder costs around $8 an hour over a five-year period, according to estimates from the Boston Consulting Group. The group projects that the cost could fall to as little as $2 an hour within 15 years.
“More generally, the 'job intensity' of America’s manufacturing industries — and especially its best-paying advanced ones — is only going to decline,” Muro writes. “In 1980 it took 25 jobs to generate $1 million in manufacturing output in the U.S. Today it takes five jobs.”
Andrew Usher said,
November 28, 2016 @ 8:06 pm
Wiener may have been wrong in details, but his economic point is unassailable – computers are in principle capable of replacing almost all human labor, and what in possible in principle will eventually be possible in practice, even if the exact timescale is very difficult to predict. Lots of smart people (including me) have independently, since 1950, looked at computing and automation and come to the same conclusion.
The solution is also agreed on by those that do not stubbornly refuse to see it: full-time employment as we know it must end, and the only way to do this consistent with the freedom we all desire is a universal guaranteed income. This had been publicly articulated, at least, in 1964; and some people have supported the same for other reasons, most of which are valid as well.
At the rate things are progressing now we don't have much time before America collapses under the weight of all the nonsense we've developed in an effort to ignore this problem.
Bill Benzon said,
November 28, 2016 @ 8:20 pm
I've been entertaining the idea that, in some ways, we're on the edge of the Marvelous Future. No, we're not flying around in jet packs; getting humans to low-earth orbit is not as routine as Kubrick depicted in 2001; the computational marvels of the Star Trek computer are still in the unforeseeable future, not to mention Cmdr Data; and environmental catastrophe seems to be closing in on us. But we're living in a very different world from that of 1950 and confront very different possibilities. Technology is at the center of it. Now we have to accommodate our thinking about society to fit the very different world before us. Andrew Usher is right, we need to think about universal basic income. Among other things.
I just watched conversation in which economist Glenn Loury (of Brown) cited Dani Rodrick to the effect that, given globalization, the national sovereignty, and democracy, you can have any two of the three, but not all three.
edhall said,
November 29, 2016 @ 2:04 am
As someone with a bit of involvement with AI back in what was an earlier peak of excitement — the early 1980s — it amuses me that today's "deep learning" techniques look an awful lot like what was being done then. Self-training neural networks abound — but running at a million or more times the speed that we had back then. Specialized hardware (both Google and Microsoft have designed chips specifically for machine learning algorithms) and repurposed hardware (graphic processing units such as used in high-end computer gaming) have been brought to bear. There have been a few algorithmic innovations, but no real game-changers — except for the raw computing power brought to bear. In some areas, such as speech or face recognition, economically useful results have been obtained. Even in less lucrative areas, such as the supposedly uncomputable game of Go, have been overtaken by machines.
I'm definitely skeptical of proposed timelines for the computational replacement of white-collar work — anyone in the field who isn't skeptical just doesn't know their history — but I'm confident that it will eventually happen.
Acilius said,
November 29, 2016 @ 9:10 am
I sometimes think that the major challenge of prediction is not answering the question "What will happen?" but "When will it happen?" Wages are paid for the completion of processes, all processes can be automated, and the cost of automation can be reduced below the minimum that would be required to keep human workers alive. Therefore, if technological innovation continues, wage labor will become obsolete. But when? If it happens in 2050, that has urgent implications for public policy. If it happens in 3050, we don't really need to give it much thought just yet.
It's like predicting, before this month, that the Chicago Cubs would win the World Series. Predicting that they would win the 2016 World Series would have been interesting, predicting that they would win the World Series someday would have been trivial.
bks said,
November 29, 2016 @ 10:35 am
I'm not sure about that Acilius. The answer to When will it happen? might be Never! Roomba has been in production for over 15 years it still can't move the furniture out of the way or clean up cat vomit, let alone dust the sideboard. The first Automat was established in Berlin in 1897, but Eatsa still depends on having humans behind the curtain and has a severely simplified menu.
Joe Econ said,
November 29, 2016 @ 11:04 am
"computers are in principle capable of replacing almost all human labor, and what in possible in principle will eventually be possible in practice, even if the exact timescale is very difficult to predict. . . . full-time employment as we know it must end"
Calm down. I don't think any of that is necessarily true. There are major swaths of the economy that can't anytime even in the distant future be replaced by automation. Education? Healthcare? Engineering? The entire world of culture and entertainment? These will need many, many humans probably forever. And the timescale of change is not only difficult to predict but most probably very, very long, accommodating decades of adjustment, which, by the way, when we speak of the manufacturing sector, already has been going on for a generation or so. This is why our economic center of gravity in the States is no longer Detroit or even New York but Silicon Valley and L.A. Change is hard, but rarely is it instantaneous.
Arthur said,
November 29, 2016 @ 11:41 am
My business area, the law firm, has been experiencing the gradual loss of white collar jobs to machines for at least 30 years. When I entered practice my law firm had 40 lawyers, 30 secretaries or typists, and about 20 paralegals, a travel agent and 1 or 2 messengers. Nowfor 40 lawyers, 15 secretaries, and 12 paralegals and no travel agents or messeengers– total lower employment, despite much much higher case load and profits per employee. As everything goes paperless, we need fewer humans in filing and trash removal also.
In retrospect, it's possible to identify several individual technological advances that cost jobs, although it wasn't obvious at the time and the layoffs usually came several years later.
secretarial: word processing, voicemail, internet, email, machine document storage eliminating filing, smart phones
paralegals: machine indexing of word processed documents (initially transcripts), machine indexing of documents generally, "smart" or "predictive document word search, internet, automated Court filing
travel agents: internet, simplified telephone conferencing, video conferencing
messengers: file transfer protocols and email
MattF said,
November 29, 2016 @ 1:16 pm
@Arthur
Note, though, that none of those technical advances is 'AI'. And most of them have a 'backend' that employs people– with rather different skills than those who were replaced.
Greg Malivuk said,
November 29, 2016 @ 3:19 pm
@Joe Econ:
Yes, all of those fields will continue to emply human beings in the most visible positions forsome time yet, but as Arthur experienced in a law firm, overall employment can nonetheless decrease greatly. Artists and engineers and singers and doctors may not be replaceable by machines just yet, but tons of jobs that used to be necessary to support their work have already been replaced, and many more likely will be in the nearish future. Anything related to paperwork and interacting with distant people, for example, can be partially replaced in those industries just as it has in law firms.
Andrew Usher said,
November 29, 2016 @ 6:54 pm
Well, 'Joe Econ' certainly reflects conventional economic thinking. That wisdom was approximately correct in the pre-computer age in which it was developed, but persisting with it now is just 'fighting the last war'. As Wiener was apparently one of the first to see, computers are fundamentally different.
And, in retrospect, we can see that even before computing, machines had reduced the total amount of human labor being done (which is a good thing) – the economy adapted to this by reducing hours of work, getting women out of the workforce, and in the last few decades introducing the concept of 'retirement age' and getting older men out of the workforce as well. That can't be repeated, but reducing the amount of human labor required is still a good thing; it just requires different instruments this time.
Bill Benzon concedes '… we need to think about universal basic income. Among other things.' The last qualifier is problematic, since there are no 'other things' that could replace or substitute (in a free society) the guaranteed income. We need to change our thinking to 'you no longer need to work in order to live', and it appears such change will have to be sudden. But then, for the first time, we could truly say that all men (and women) are equal economically.
k_over_hbarc at yahoo dot com
George said,
November 30, 2016 @ 4:17 am
Coming back to the extract that Victor Mair quoted in the first comment on this thread…
"Short of any violent political changes or another great war, I should give a rough estimate that it will take the new tools ten to twenty years to come into their own."
… I would have thought that, whatever about violent political changes, war very often accelerates technological advances rather than slowing them down.