Friday, 24 March 2017

They Took Our Jobs

According to a report by PWC, millions of UK workers are at risk of being replaced by George Osborne. Sorry, by robots. This is a rehash of the Oxford University study from 2013 that suggested almost half of US jobs might be automated by 2030, with the addition of the OECD study from last year that suggested a more conservative estimate of only 10% of jobs at high risk of full automation. The great minds of PWC think the correct figure may be somewhere in the middle, say 30%. This latest report is little more than a marketing exercise to keep the mega-consultancy in the news, and the usual suspects have obliged by regurgitating suitably juicy gobbets, notably the headline figure of 10 million jobs at risk. Given that the Bank of England was warning in 2015 that the figure could be 15 million, you might think a more positive spin would be in order: "Robots to take fewer jobs than first thought", or similar. But this would be as naïve as wondering why Nigel Farage and Katie Hopkins foretell a race war because a bloke from Kent called Adrian went postal.

The PWC report opens with the now obligatory Luddite Fallacy Fallacy: "a backlash by skilled handloom weavers against the mechanisation of the British textile industry ... But, in the long run, not only were there still many (if, on average, less skilled) jobs in the new textile factories but, more importantly, the productivity gains from mechanisation created huge new wealth". The Luddites were complaining about the decline in skill, not the loss of jobs, which resulted in falling wages. The new wealth that accrued from increased productivity was captured overwhelmingly by capital, not by labour, while the subsequent expansion of the industry and creation of more jobs in the UK during the nineteenth century was dependent on the dismantling or degrading of textile manufacturing in India and other parts of the British Empire. As usual, the Luddites are deployed as a strawman to insist that technological change creates new jobs, and just as usual, the PWC report suggests that automation will free up labour for more human-centred work.

The heart of the analysis is a comparison between the Oxford and OECD studies, employing PWC's "own machine learning algorithm for identifying automation risk". This turns out to be nothing more than a weighted model applied to a merged dataset incorporating the two studies plus publicly-available census data. This is not "machine learning". The difference between the two studies is that Oxford assumed that manual or routine jobs with low educational requirements were highly automatable, while the OECD assumed automation would be task-oriented. The former biased towards substitution - workers replaced in whole by robots or AI - while the latter biased towards complementarity - parts of a job being automated but workers still having a role. To give a real world example of this from history, containers substituted for some dock workers by replacing their role in the transfer of goods from ship to shore, but fork-lifts and cranes complemented other dock workers in the packing and onward distribution of those goods.

The problem with the genre of job-stealing robot studies is the ideological emphasis on "technological revolution", which has been repackaged since the millennium as "disruption". While the pace of change certainly speeds up and slows down, the long-term reality is incremental and steady progress, often over decades. This is obvious enough from history - dock workers did not disappear overnight - but the nature of tech-boosting and the appetite of the media means that sudden lurches are always more prominent in popular debate, hence the current predictions of an "economic tsunami" as truck-drivers are replaced by autonomous vehicles. In reality, the impact of technology on work is heavily-mediated by existing institutions (state regulation, employment rights etc) and by social absorption - i.e. the way we informally accommodate new tools and the cultural norms this gives rise to. For example, the mass adoption of email has undoubtedly led to productivity gains, but it has also produced productivity losses through distraction and poor practice.

Many of these studies reflect historic social assumptions and give insufficient weight to political imperatives. For example, they usually assume that education will remain a sector at low risk of automation because of its traditional human-centred nature, despite the obvious manufacturing paradigm that has been evident in its organisation since the late 19th century. However, the political drive in the UK and elsewhere is towards the further commoditisation of mass education, which drives both automation (greater reliance on online tools) and deskilling (classroom assistants). It is possible that qualified teachers will increasingly take on more of a pastoral than pedagogic role in the future, reflecting how state concerns about value formation have shifted from the preparation of workers (learning self-discipline and obedience) to the moulding of conformant citizens (countering "radicalisation" and facilitating the emergence of sanctioned identities). Mr Chips would be proud.

One of the more amusing assumptions in the PWC study emerges when it compares the difference in the number of jobs that might be automated by sector between different countries, noting that finance and insurance scores 61% at risk in the US but only 32% in the UK. "[T]he key difference is related to the average education levels of finance professionals being significantly higher in the UK than the US. This may reflect the greater weight in the UK of City of London finance professionals working in international markets, whereas in the US there is more focus on the domestic retail market and many more workers who do not need to have the same educational levels". A more honest comparison of the two countries is that the UK has a lopsided finance sector, dominated by rent-seeking rather than particularly sophisticated services, in which education (where you went as much as what you learnt) serves the purpose of social sorting (along with the right kind of shoes). As a business consultancy, PWC is part of the same social milieu and thus subject to the same norms and biases.

Likewise, construction in the UK is seen to be less at risk of automation than the same sector in Germany because "there is a greater proportion of time spent on management tasks in the UK, such as planning and consulting others, and those that require social skills such as negotiating". The implication that restrictive planning laws impose a greater burden should be resisted: Germany doesn't have a planning free-for-all, its builders are simply more efficient than their British counterparts. Another way of putting this is that we don't actually build enough houses in the UK, with the result that the industry can support greater levels of rent-seeking and the distribution of profits via supernumerary roles. This obliviousness to actual social and political factors in the analysis of job automation means that little reliance can be placed on speculative estimates. It also means that we should stop thinking of automation as an irresistible force of nature and recognise that it is a political matter.

The PWC analysis does address the social and political dimension, but in predictably abstract and biased ways. Thus we are told that "the government, working with employers and education providers, should invest more in the types of education and training that will be most useful to people in this increasingly automated world". Substitute "capital" for "people" and you reveal the true thinking. The hoary old 'climate for business growth' also gets a shout-out: "Central and local government bodies also needs to support digital sectors that can generate new jobs, for example through place-based strategies centred around university research centres, science parks and other enablers of business growth". To show they're up with the zeitgeist, the report authors even suggest the need to consider a basic income, but they promptly disavow it in terms that will please the government: "For the moment, the need to reduce the UK budget deficit may be a significant barrier to any such scheme on a national level, as well as concerns about the social acceptability of giving people ‘money for nothing’", proving that the language of self-interest and class contempt has not changed much since the days of the Luddites.


  1. Likewise, I'm sceptical. And I think there's a lot of confusion about what we're actually talking about. There's a lot of hype about AI, but that gets confused with big data and number/behaviour crunching, and that's not the same thing. Even with data that can be crunched, there would be a lot of problems with the quality of the data, and the compatibility of different datasets (as dsquared often mentions). And that's even before we start designing the software.

    Much of the hype has speculated on the automation of a lot of white collar jobs - accounting, solicitors, etc. While large manufacturers have made enormous strides in computerising their production and accounting systems, these have tightly defined scopes and are easily extensible across different manufacturers. I'm not sure that contract law, for example, would be handled so easily. I would have thought that even conveyancing, which on the surface is a routine operation, would have a vast number of potential inputs and exceptions - although maybe I should leave commenting too much on that to somebody in the industry.

    The one thing that IDS was apparently aware of with the implementation of Universal Credit is that it involves business transformation. I wonder how easy such transformations would be in the areas being speculated about.

    1. The fact that journalism on the subject of automation treats robots and AI as interchangeable tells you that they are not interested in the particular impact of these technologies on the means of production but see them simply as the latest iteration of the general demand directed at labour: you must change to suit the needs of capital.

    2. Yes, I hadn't though that this might just be a reflection of wanting to make a statement of position, of attitude, rather than be seriously discussion the issues.

      In that sense, it's rather like Rudd, who transparently knows nothing about encryption, striking a pose over WhatsApp. Mind you, it was also transparent that the editors of last nights BBC TV news didn't either. The "no safe space for terrorists" line had me morphing into Scotty shouting "it's against the laws of physics".

  2. Ben Philliskirk25 March 2017 at 21:09

    Given that a majority of jobs are probably superfluous anyway, I suppose they'll just invent some more to replace the ones that are automated out of existence.

    I suspect a 20-hour week is perfectly plausible now if people really wanted it. Or alternatively, to paraphrase the Jameson quote from your recent tweet, it is easier to imagine the end of the world rather than the reduction in the hours of wage labour....

  3. Herbie Kills Children29 March 2017 at 17:34

    It never ceases to amaze me that computers can replace endless manufacturing jobs, even postmen, yet the number of accountants employed continues to rise.

    If computers can do one thing really really well it is easily adding things up across multiple dimensions!

    So the puzzle is why are there any accountants left?

    The more jobs that can be replaced by robots the better.

    The interesting thing is when watching the news they prattle endlessly on about people living longer therefore health provision under threat but never ever bring in the other side of the equation, rise in productivity, improvement in technology. The improvement in technology should have more than compensated for people living longer.

    The news never bring in this aspect because that would point to systematic failure, and the news are simply and nothing more than a brainwashing unit at the service of the ruling class

    1. The problem isn't accountants per se - i.e. people who put together statutory accounts - but the wider ecosystem of corporate finance. SMEs don't use any more accountancy that they ever did, despite claims about red tape. It used to be said that the British Empire was a vast system of outdoor relief for the upper classes. In the postwar era, that function shifted to big business with finance as one of the chief vectors for the creation of supernumerary roles.

      The percentage of CFOs who attended private schools (70%) is much higher than that for CEOs (50%). This is ultimately because CFOs for PLCs need to "culturally fit" with banking and investment people in the City, which inevitably privileges people from a similar social background. This then filters down to next level roles within finance and to medium-scale businesses who want to be flotation-ready. Like hires like.

      As a self-appointed business aristocracy, it is in the interests of the finance crowd to both expand their numbers and to further the complexity of their roles in order to drive up remuneration. Though this is a drag on productivity (and explains in part the UK's poor record relative to others), it is tolerated by most business owners and institutional investors because they are part of the same milieu.

      Far from being a way of reducing the power of finance, IT was quickly absorbed into the empire. Early EDP functions were usually under the control of the CFO because they did number-crunching, and many business have stuck with this model. This has had the pernicious effect of channelling systems design into a finance-friendly paradigm, hence clunky ERP systems and rubbish BI.

      Pensions are just another form of wealth distribution in which the social bias is obscured by the temporal shift. We imagine we're saving (or paying tax) for our own pensions, but we're usually doing it for others.