Global judgements and ideas.
Some notes on machinery working and measuring work
Research on the capability of machinery to carry out human-like activities started in the 1950s when a researcher at Dartmouth College at a summer conference first coined the term artificial intelligence, saying that progress can be made in getting machines to ‘solve kinds of problems now reserved for humans’ (McCarthy et al, 1956). Also in the 1950s, Ted F. Silvey, National Headquarters department of education staff of the Congress of Industrial Organizations (CIO) and American Federation of Labour (AFL) pointed out that ‘machines and instruments can do almost everything except buy what they make!’ (Silvey, 1958). Silvey had earlier noted (1957):
Instruments substitute for man’s mind, just as the rest of the machine takes the place of his muscles. Machines are acquiring the skill of human beings, but they must work faster and more accurately than anything of flesh and blood—and they never tire.
In this context, unions were concerned in the early Western era of manufacturing that the technologies involved in mass production worked to ‘trivialize’ man by its ‘repetitive performance of bits. His craft skills, his creativeness, his human dignity, his uniqueness were, at best ignored and at worst, stomped on’, causing the ‘destruction of the workers’ dignity as people’ (Silvey, 1956, p. 3). This same trade unionist also optimistically pointed out the possibilities that mechanization would reduce the work week and work year, ‘both with full wage or salary income’ (1958) and claimed that ‘automation promises a time when a comparative handful of people will have to work in factories at the dull, repetitive tasks demanded by mass production’ (1957, p. 30). Silvey’s optimism as well as pragmatism are remarkable: he states that ‘in the long run, automation will make more jobs… but the challenge is to solve the problem in the short run, to give immediate aid to the worker whose fingers are caught in the door when it is slammed shut’ (1957, p. 29). Shortly afterward, Braverman (1974) hinted at the origins of algorithmic processes as a feature of the development of machinery, indicating that ‘when the tool and/or the work are given a fixed motion path by the structure of the machine into that machinery in the modern sense begins to develop’ (130). The machine’s ability to run itself has become almost accepted in contemporary life, but what happens when humans begin to make decisions based on specific aspects of the machine’s operations with little or no external interference?
Today, the machinery question is ‘back with a vengeance’ (Economist, 2016, p. 3) because this question is infiltrating professional workplaces where all kinds of work can be increasingly automated and new technologies begin to measure work as it happens. In 2012, the ImageNet Challenge set people to programme computers to recognize images. These ‘challenges’, or contests, coordinated by top researchers and corporations and became a measure of success in the field, contributed to the rapid improvement in what is called ‘deep learning’, and the computer’s ability to recognise images has now surpassed humans. This and other experiments are bringing about the realisation that tasks once considered the exclusive remit of humans are now at risk of automation, mechanisation and digitalisation, and Frey and Osborne’s more recent report (2016) demonstrates that a both repetitive and non-repetitive jobs are now susceptible. Tele-marketers, tax preparers, insurance underwriters and library technicians are at a high risk, at .99 probability (1 = certain). Work in the professions as healthcare and social work (.0035) and recreational therapy (.0028) are also under threat. Much legal casework research can now be done by computers using deep learning algorithms (Ford, 2015). Non-routine work such as driving and deciphering handwriting is now being made possible by machines. So aspects of our work are automated, as logging that workers once may have done on clipboards and in filing cabints or what I call ‘ana-logging’ as a reworking of the word ‘analogue’, can now be done by machines (whether that is done more easily/more efficiently is another debate, given the issues surrounding incessant technological failure we have all experienced and corporations build entire infrastructures to avoid).
In any case, the threat of automation in factories has been updated by new patterns of labour selection in new work design models such as the ‘sharing economy’, facilitated by a new methods of work selection and distribution called people analytics (PwC, 2015), which facilitate a process of identity management (Ajana, 2013) evident in new online platforms in the demand economy (AFL-CIO 2016) where people buy and sell labour. The sharing economy or work in the ‘human cloud’ includes such platforms as Upwork, ODesk, Guru, Amazon Mechanical Turks, Uber, Deliveroo and Handy which are called ‘online platforms’ in the Digital Single Market European Commission terminology. Huws (2015) and Cherry (2011) label this type of exchange and work as ‘crowdsourcing’ and Huws defines it as ‘paid work organized through online labour exchanges’ (2015, p. 1). Crowdsourcing has facilitated companies’ outsourcing of labour as well as introduced new platforms for freelance and self-employed work. The Office of National Statistics in the United Kingdom reported in July 2016 that the number of self-employed workers increased from 3.8 million in 2008 to 4.6 million in 2015 (ONS, 2016). The platform economy relies on self-employed contracted labour in both the UK and internationally and, as such, have no access to regular employment benefits such as health care or maternity leave. Workers have very little legal protection either, and platforms are designed to reduce employer liability.
When these kinds of platforms were first introduced, workers used them to top up incomes and the work was mostly in more advanced countries. However, over time, workers with no other incomes in both the north and south have become heavily reliant on these spaces. They facilitated outsourcing of work to the global south where labour prices are lower and labour market regulation localized (Bergvall-Kåreborn and Howcroft, 2014). But now, even professional tasks on these platforms are being .broken down by their least common denominator…. and ‘the way that tasks and human capital is being viewed and handled is… one that almost serves to dehumanize workers’. (Cherry, 2011, p. 30). Companies have tended to follow minimum standards particularly in Third World Countries (Estreicher and Cherry, 2008) and often adopt corporate social responsibility models alone which affects how outsourced labour occurs.
On some platforms, such as Mechanical Turks and Upwork, people place available job contracts online and workers contact clients to pick up work. Work is often distributed in a piece meal fashion to various workers as part of outsourced labour. The work from such platforms offered ranges from graphic design to programming, but communication between the worker and client is usually very limited, leading to a distinct lack of transparency. This can raise ethical questions as ‘workers are unable to make judgments about the moral valence of their work’ (Zittrain, 2008, cited in Howcroft, p. 218).
‘People analytics’ is is the dominant selection method in the sharing economy and is made possible by reduced costs of data and information processing and is of interest to business because it is seen to reduce costs in service-provider selection and/or workplace reorganization and restructuring. Intensified worker reputation self-management is essential in the online labour market as freelancers seek work and as employers or clients actively profile employees with the use of new technologies (Pasquale 2010; Gandini 2016; Bodie et al, 2016). The chasing and utilising of social capital to enhance and further careers and to find work and employment is not itself new, but the type of reputation formed that allows freelancers and the like to find work on online platforms is ‘based on algorithmic-based third party elaboration that translated the opinions of others into reputation proxy’ (Gandini et al, 2016). So reputations are acquired through the number of tasks a worker took on board and ratings by customers. For example, Uber drivers report that if they receive customer ‘star’ rankings below 5.6 or 4.5 they can be fired, despite some aspects of a journey can have nothing to do with a driver’s performance such as traffic and 3G or the soiling of the car by another passenger. Drivers receive no help from the firm for related issues and often receive much less income than they were promised upon becoming drivers (Brownstone, 2015). Nonetheless, the paradox and fiction of algorithms is that they are ‘absent’ of ‘human bias’ (Frey and Osborne, 2013, p. 18). Bodie et al (2016) point out that ‘workers want to be treated as people, not ranked as fungible data sets or assessed as cost centres’ (p. 75).
The profiling of human behaviour and resulting data, allows management to make judgments about ‘who people are’ as well as to predict future behaviour. Computer generated data is expected to be reliable, neutral and to help with forecasting (Amoore, 2013; Cheney-Lippold, 2011). The assumed neutrality and utility of data for these purposes is what is at stake in workplace power relations, whether the workplace is one of a freelance worker or a full time employee. Workers are increasingly easily selected and discarded; replaced and disposable in this ‘profane’ referencing system (Gandini et al, 2016). Reputation in the online labour market has become incredibly important for work that happens in digital spaces, so-called ‘virtual work’ (Huws, 2014, 2013; Holts, 2013) and ‘digital labour’ (Fuchs, 2014).
Virtual work has already been proven to perpetuate precarity and pressures people to overwork (Huws, 2014; Moore and Robinson, 2015). The problems that arise are the obstruction of newcomers and difficulties in entrance points to what becomes a closed loop of client/worker relationships. The logic of algorithmic reputation acquisition penalises non-standard workers and people who have other responsibilities such as child and elderly care. Irregular career patterns can result from time out of work for reproductive domestic labour, maternity leave, physical illness and mental health issues. Trade unions are taking steps to address some of these issues and there is more work to be done.
These notes are taken from a paper I have written with Pav Akhtar, UNI Global Union. To cite please email me pvm dot doc at gmail dot com.