This Time Isn’t Different

~1983 I read two articles that inspired me to change my career. One was by Ted Nelson on hypertext publishing, and the other by Doug Lenat on artificial intelligence. So I quit my U. of Chicago physics Ph.D. program and headed to Silicon Valley, for a job doing AI at Lockheed, and a hobby doing hypertext with Nelson’s Xanadu group.

A few years later, ~1986, I penned the following parable on AI research:


Once upon a time, in a kingdom nothing like our own, gold was very scarce, forcing jewelers to try and sell little tiny gold rings and bracelets. Then one day a PROSPECTOR came into the capitol sporting a large gold nugget he found in a hill to the west. As the word went out that there was “gold in them thar hills”, the king decided to take an active management role. He appointed a “gold task force” which one year later told the king “you must spend lots of money to find gold, lest your enemies get richer than you.”

So a “gold center” was formed, staffed with many spiffy looking Ph.D types who had recently published papers on gold (remarkably similar to their earlier papers on silver). Experienced prospectors had been interviewed, but they smelled and did not have a good grasp of gold theory.

The center bought a large number of state of the art bulldozers and took them to a large field they had found that was both easy to drive on and freeway accessible. After a week of sore rumps, getting dirty, and not finding anything, they decided they could best help the gold cause by researching better tools.

So they set up some demo sand hills in clear view of the king’s castle and stuffed them with nicely polished gold bars. Then they split into various research projects, such as “bigger diggers”, for handling gold boulders if they found any, and “timber-gold alloys’, for making houses from the stuff when gold eventually became plentiful.

After a while the town barons complained loud enough and also got some gold research money. The lion’s share was allocated to the most politically powerful barons, who assigned it to looking for gold in places where it would be very convenient to find it, such as in rich jewelers’ backyards. A few bulldozers, bought from smiling bulldozer salespeople wearing “Gold is the Future” buttons, were time shared across the land. Searchers who, in their alloted three days per month of bulldozer time, could just not find anything in the backyards of “gold committed” jewelers were admonished to search harder next month.

The smart money understood that bulldozers were the best digging tool, even though they were expensive and hard to use. Some backward prospector types, however, persisted in panning for gold in secluded streams. Though they did have some success, gold theorists knew that this was due to dumb luck and the incorporation of advanced bulldozer research ideas in later pan designs.

After many years of little success, the king got fed up and cut off all gold funding. The center people quickly unearthed their papers which had said so all along. The end.

P.S. There really was gold in them thar hills. Still is.

As you can see, I had become disillusioned on academic research, but still suffered youthful over-optimism on near-term A.I. prospects.

I’ve since learned that we’ve seen “booms” like the one I was caught up in then every few decades for centuries. In each boom many loudly declare high expectations and concern regarding rapid near-term progress in automation. “The machines are finally going to soon put everyone out of work!” Which of course they don’t. We’ve instead seen a pretty slow & steady rate of humans displaced by machines on jobs.

Today we are in another such boom. For example, David Brooks recently parroted Kevin Kelley saying this time is different because now we have cheaper hardware, better algorithms, and more data. But those facts were also true in most of the previous booms; nothing has fundamentally changed! In truth, we remain a very long way from being able to automate all jobs, and we should expect the slow steady rate of job displacement to long continue.

One way to understand this is in terms of the distribution over human jobs of how good machines need to be to displace humans. If this parameter is distributed somewhat evenly over many orders of magnitude, then continued steady exponential progress in machine abilities should continue to translate into only slow incremental displacement of human jobs. Yes machines are vastly better than they were before, but they must get far more vastly better to displace most human workers.

GD Star Rating
Tagged as: ,
Trackback URL:
  • e

    What’s your opinion on “AI” development in the near future? Siri, Cortana and co. seem to be moving steadily along (as natural-language parsing and response agents). Do you think in 15-some years we may have AI secretaries?

    • chaosmosis

      We have automated on-hold lines, is that terribly different?

      I think that real secretaries do more than answer the phones and schedule appointments, though. They do a lot of complicated social managing too.

    • Jimbo

      I don’t think natural language technology is as advanced as it seems. Machines cannot read and have no understanding of the text given to them. The advances reported, like IBM Watson winning jeopardy, spam detection, irony detection, the detection of tone of a document etc. is based on relatively simple statistical or probabilistic processes trained on very large numbers of documents. It is impressive, but the experience of the end user is a useful illusion of intelligence. It seems much more impressive than it is. You have to look under the hood to be underwhelmed.

  • kurt9


    As an automation/control system engineer I concur. Robotics and automation is not going to make 50% of all current jobs redundant in the next 20 years. We don’t even have decent machine vision, which is necessary for really useful robotics. Maybe 10% of all human jobs will be displaced by automation over the next 20 years.

    In other ways, the great automation displacement has already occurred in manufacturing. Since Chinese wages have increase lately, some textile and other manufacturing has returned to the U.S. A new textile mill was build in South Carolina. It employs 150 people. The same mill built in 1980 would have employed 2,000 people. Automation is a slowly progressing phenomenon.

    • Jimbo

      Many big AI problems could turn out to be AI-complete (like visual perception, for instance). If so, we’d need to solve the problem of general AI before we have machines capable of automating a lot of occupations.

  • Yvain

    “One way to understand this is in terms of the distribution over human jobs of how good machines need to be to displace humans. If this parameter is distributed somewhat evenly over many orders of magnitude, then continued steady exponential progress in machine abilities should continue to translate into only slow incremental displacement of human jobs. Yes machines are vastly better than they were before, but they must be much far vastly better still to displace most human workers.”

    Why can’t jobs be arranged in a series of bottlenecks, such that solving the bottleneck opens a very large category of things to automation?

    For example, human level visuospatial movement is a plausible near-term advance, and might allow the automation of waiters, janitors, tour guides, taxi drivers, et cetera.

    Human level natural language recognition (like Watson) is another, and potentially hurts secretaries, call center employees, et cetera.

    More generally, if you imagine jobs as stratified by IQ (with people able to substitute any job for any other job up to their IQ level), you see no unemployment once machines reach IQ 10, still no unemployment at IQ 20, still none at IQ 30, and then once you reach 80 or so suddenly a few more steps put nearly everyone out of work.

    If you imagine machines improving one IQ point per year, then you can have the first eighty years of machine progress with minimal unemployment, and then the next thirty put almost everyone out of work.

    Obviously IQ isn’t a good way to model machine progress, but there could still be a similar pattern of relatively low unemployment until machines approach “human level” by some metric, and then very sudden unemployment of non-intellectual tasks.

    Since a lot of the hard work is done by programming the machines and giving them lookup table type things for specific situations, I could imagine this happening well before the machines reach anything like general intelligence.

    • Most jobs from two centuries ago have been automated by now. So we’ve already seen a LOT of automation. Yet we haven’t seen much of your hypothesized bottleneck burst scenario.

      • Michael Vassar

        I think we have seen exactly that. The jobs that were fundamentally about the application of physical force all got automated away fairly quickly. So did most of the agricultural jobs, which had largely been bottlenecked by harvesting, leading to dust bowls and socially disruptive flooding of the cities, and massive changes in the nature of government to accommodate.

        It seems to me, Robin, that you look at history and see little abrupt technological change, while I see lots of abrupt technological change, very locally, and political resistance spreading it out and mitigating its impact. People who say that something must be done are saying that such political solutions will again be needed, as they have been in the past, not that nothing will be done so DOOM.

      • You have any data to support this claim? Or even particular places and time periods you claim these bottleneck bursts happened during?

      • HumansNeedNotApply

        I don’t see any bottlenecks but we are currently in a steady fall. Labor’s share of income has been falling around the world:

        And it can’t be blamed on offshoring since it is also happening in places like Mexico and China.

    • skeptic

      Sure. But there’s no reason to believe the knee of the curve happens to be near us now.

  • TheBrett

    I’ve since learned that we’ve seen “booms” like the one I was caught up in then every few decades for centuries. –

    Faster than every few decades. I’d say it comes about once every ten years now, usually during an economic downturn with high unemployment. I remember Jeremy Rifkin had the misfortune to pen The End of Work about this topic in 1995, just as things were getting better in terms of employment, growth, and wages across the board in the US.

    Yes machines are vastly better than they were before, but they must get far more vastly better to displace most human workers. –

    I could see them replacing many of the existing jobs today eventually (key word being “eventually”), although I can’t see them really displacing human work in general* – especially since the definition of “work” can change. For more than half of free Americans before 1860, “work” meant “farm/small business/professional/artisan/etc”. We live in an “employee” society now, but it may not stay that way forever.

    Plus, it seems like you could always figure out ways to put humans in a managerial/supervisory role over ever increasing layers of machinery underneath their control.


    This discussion is meaningless without agreeing on the defintion of “work” or “automation”. Surely automation has replaced many, many jobs in the past, I’d even go as far to say a majority of jobs that existed in 1900 have been replaced by automation. By that standard “the machines” have already taken our jobs. But so many new jobs have been invented that we aren’t all unemployed. That trend may continue for a while even after machines become capable enough to do most of today’s jobs. If you’re talking about an era when automation proceeds faster than the creation of new jobs (and should we also compensate for working hours and percentage of lifetime spent in employment?) over a long period then yes, that indeed won’t happen anytime soon, but people should be clear about what they mean exactly in this discussion.

  • vaniver

    I worked on a project once for a warehouse management software company, which sold software and hardware to make human laborers more effective. Their competitive advantage was that they were very good at simulation and optimization; they could take actual order data that the warehouse had received over the last six months and say “this is how we would have arranged your flow, and it would have saved you 17%” whereas the robot companies would say “we think it’ll save you 20%, maybe,” and the non-engineers who didn’t get the toy joy out of buying new robots would price the risk at much more than 3% and go with the simulation company.

    The thing I found fascinating about it, though, is that warehouse workers had three relevant organs: eyes, hands, and feet. It’s very difficult to replace the eyes and hands–humans are much better at grabbing one box of pens from the big box of pens than a robot–but it’s not as difficult to replace the feet (these are the Kiva robots that pick up and move shelves to bring them to the picker, and they were bought by Amazon around the time I was working on this project).

    Turns out, about 80% of a picker’s time is spent walking. So as the picker job transitions from walking through lines to fill boxes to standing in place and filling boxes, about 80% of picker jobs go to the Kiva robots. It seems to me that this is the visible low-tech version of the general human-computer synthesis argument; as computers make humans better at their jobs by reducing the tedious work, we need less humans to do those jobs (both because there’s less tedious work to do, *and* we move from the best 10 pickers to the best 2 pickers, who probably have higher average productivity).

    (It also makes it obvious how a drastic local transition–one warehouse switches and fires 80% of its picker workforce–can become a smooth global transition, as the warehouses switch over one by one.)

  • HumansNeedNotApply

    I think it is true that we are a very long way from automating all professions. But we can automate a very significant percentage of jobs (>20% for example) while not automating a majority of professions.

    Professional drivers alone are between 5 and 6 million jobs in the US. But it’s just one profession. Only a handful of professions automated like that could result in an unemployment rate higher than during the great depression.

  • Mark Bahner


    Here are some of the most common jobs in the U.S. in 2012, with the number of people, in millions:

    #1 Retail salespeople = 4.2 m

    #2 Cashiers = 3.4 m
    #4 Fast food preparation and service = 2.7 m
    #9 Janitors = 2.1 m
    #12 Bookkeeping, accounting = 1.8 m
    #14 Tractor trailer drivers = 1.6 m
    #15 Elementary school teachers = 1.5 m

    Robin of 1984 might have predicted that many of them would be gone due to AI in 2014. But they’re obviously not. But is this time different? What are peoples’ predictions for what the number of people in the U.S. employed in those sectors in 2024, 2034, and 2044?

    I’ve already typed my predictions into a blog post on my blog. I’ll hit the publish when some folks (hopefully especially Robin) have made predictions here.

    P.S. A potential format for the predictions would be the value in 2012 in millions (just a repeat of the numbers above), followed by the predicted values in 2024, 2034, and 2044. For example, using hypothetical numbers only:

    #1 Retail salespeople = 4.2, 4.3, 4.4, 4.5.

    #2 Cashiers = 3.4, 3.6, 3.5, 3.4.

  • Cahokia

    It’s weird that public sentiment simultaneously holds that mass unemployment due to automation is a serious danger and that we need to reduce spending on health care, one of the most labor intensive of industries.

    • MarkBahner

      I think if U.S. spending on health care was increasing life expectancy at birth by 6 months per year, or even better one year per year, people would be a lot happier with the increasing health care expenditures.

      • Cahokia

        Ok, but here’s a question for you:

        Do you suppose that using the federal government as a battering ram to reduce health care spending will improve current rates of life expectancy increase or not?

      • Richard Haven

        Seems to work in other places

  • MarkBahner


    I made my predictions:

    I predict that by 2044, employment in what were the top 15 industries in 2012 will be reduced by more than 50 percent. The biggest percentage losers I expect to be tractor trailer drivers, cashiers, and material movers (such as loading dock workers). I see all three declining by about 90 percent.

    But I don’t think the drops will be particularly noticeable for another ~10 years.

  • Pingback: Overcoming Bias : AI Boom Bet Offer()

  • Pingback: Overcoming Bias : Missing Engagement()

  • Pingback: Overcoming Bias : World Basic Income()

  • Pingback: Overcoming Bias : Brains Simpler Than Brain Cells?()

  • Pingback: Overcoming Bias : This AI Boom Will Also Bust()