icm2re logo. icm2:re (I Changed My Mind Reviewing Everything) is an  ongoing web column  by Brunella Longo

This column deals with some aspects of change management processes experienced almost in any industry impacted by the digital revolution: how to select, create, gather, manage, interpret, share data and information either because of internal and usually incremental scope - such learning, educational and re-engineering processes - or because of external forces, like mergers and acquisitions, restructuring goals, new regulations or disruptive technologies.

The title - I Changed My Mind Reviewing Everything - is a tribute to authors and scientists from different disciplinary fields that have illuminated my understanding of intentional change and decision making processes during the last thirty years, explaining how we think - or how we think about the way we think. The logo is a bit of a divertissement, from the latin divertere that means turn in separate ways.


Chronological Index | Subject Index

A stitch in time saves nine

About the future of machine learning for micromanufacturing

How to cite this article?
Longo, Brunella (2019). A stitch in time saves nine. About the future of machine learning for micromanufacturing. icm2re [I Changed my Mind Reviewing Everything ISSN 2059-688X (Print)], 8.11 (November).

How to cite this article?
Longo, Brunella (2019). A stitch in time saves nine. About the future of machine learning for micromanufacturing . icm2re [I Changed my Mind Reviewing Everything ISSN 2059-688X (Online)], 8.11 (November).

London, 15 October 2019 - While the public discourse around applications of artificial intelligence revolves around machine learning in the utilities sectors or personal data in health and finance, or focuses on spectacular innovations like driverless cars or cleaning robots (good luck with both!), little we know about the incremental changes constantly sought and experimented through automation of repetitive processes in manufacturing, particularly in the textiles and apparel industry, and how these incremental changes interact with the whole of the society and the environment.

I refer here to the type of changes that usually cause the elimination of low paid labour and look forward to major transformative progress in the supply chain - most of the times making two steps forward and one backward until the innovation finds accommodating patterns of social acceptance for all the stakeholders.

The consumer side of fashion and clothing is, of course, overwhelmingly more important for marketing and media communications than the productive one. The risk of associations of luxurious brands with crowded sweatshops and cheap labour is always over the corner. This is one of the many reasons why there is poor media coverage on innovations in manufacturing. Another one is that you do not usually engage large audiences talking on how applied sciences make the modern world or - even worse - on how public policies impact the adoption of new technologies: with few exceptions that leverage on personalities and community engagement around easy or topical ideas, attempts in this direction have proved to be pathetic.

A couple of generations ago, in many Western countries almost every family owned a sewing machine. Will the future of manufacturing in the fashion and apparel sector resemble that past, with a myriad of family, small and micro businesses replacing the mammoth and mostly unsustainable factories that have proliferated in Asia for the last two or three decades?

It may be. These are really exciting times to think about change management for technical and scientific communication. Business, industry and company history offers examples of extraordinary failures, inventions, dynamics of such a mission: shall we say there are several ways to tell economic history or history of science and technology once we choose this or that narrator?

Take the invention of the sewing machine, for instance. Singer was the greatest of all the inventors of his generation in that: he used a number of others' patented ideas to make his machine, he partnered with other entrepreneurs and paid royalties on others' patents in order to achieve his goals. He wanted to sell a very innovative solution to the problem and practice of sewing, for which he also revolutionised sales, offering instalments plans: there were hundreds of entrepreneurs and models of sewing machines both in USA and in the UK in the 1850s and 1860s, very proud of their amazing designs. But the absolute majority of them did not create a multinational like Singer did, still a major brand and global business player 160 years later.

The Age of Information versus the Age of Automation

Some authors would be totally unknown and forgotten forever if public libraries and digitisation projects did not exist. It is the case of the American Don Bissell, author of an amazing story of the Singer Sewing Machine Company covering 145 years (The first conglomerate. Audendreed Press, 1999). In his extraordinarily documented though inevitably apologetic reconstruction of the company history, from the beginnings to the late 1990s, he suggests that the workplace has evolved mainly by producing and placing small appliances into the world’s households. Of all the goods produced during the entire Industrial Revolution, none have been more influential than the first mass marketed home appliance, the sewing machine. I tend to agree.

By comparison - Bissell continues - our way of life owes much to the sewing industry’s evolution and comparatively little to the recent advent of the computer. [...] The computer’s popularity, when viewed statistically, only now begins to rival that of the sewing machine. [...] Further, today’s self-styled Information Age is no more than the predicted outgrowth of automation. (Singer company executives foretold of this future outpouring of data processing technology during the early 1950s.) How do modern editorialists leap over this inescapable historical perspective? Mostly, they assert that present-day changes in social and business models represent an incipient and unique era in world history. Such assertions, which often take the form of “a new world order is upon us,” rely on a supposed reformation ongoing in world commerce and in the global workplace. However, these models of business behavior are not new. The Singer company history discloses that today’s changing business methods merely embellish its own historical models. [...] The computer is no more than the latest tool or logical enhancement occupying a brief sidelight in the saga of industrial history. The study of economic history reveals that a country’s fortune and way of life will lead or lag according to its willingness to control emerging technologies. [...] The economic revolution began in France’s textile industry and then prospered under Napoleon’s protectionist trade policies and his well documented support of scientific research and development. Unfortunately, France, which brought forward the textile industry’s automatic machines, lost itself in internal politics, betrayed Napoleon’s legacy, and did not benefit from its countrymen’s scientific genius. Instead, England took hold of and reinvigorated automation’s tools. In England, new technologies, like spinning jennies and power looms, flooded its fabric industries. When the technological mantel next crossed the Atlantic, Yankee ingenuity gained control over the economic revolution. Then, after years of affluence, America – like her many historical counterparts – decided to send her leading edge technologies onto foreign shores. The American news press has little criticized this latest transfer of technology into the international public domain. But, even before passing automation’s mantel to a new generation of upstart foreigners, even before giving computer technology away, America paved the road by giving sewing machine technology away.

I reported here this long paragraph from Bissell's introduction to his book because it reminded me, with quite literally the same words, not only the discussion on the long wave of Information and Communication Technologies I considered last month (see icm2re 8.10 but also the debates I joined few years ago about the transfer of the control of the Internet main governance body, ICANN, from the USA original NTIA to various international multistakeholders communities and other organisations more representatives of worldwide interests (for more on this see icm2re 3.7). And so on and so forth. If you want to make an exercise on the rhetoric of change management and technology policies try to apply the quote to any other subject in which experts talk of a "new world order upon us": it is likely to work. "America paved the road by giving [some] technology away."

The future of the factory beyond the behemoth era

Many descriptions of the Asian clothing and smartphones manufacturing plants recall the image of the typical giant automotive plants of one century ago. But why do export-oriented factories in China and Vietnam need to be so big? This is a fascinating questions Joshua B. Freeman, author of a history of the factory and the making of the modern world, has tried to answer in a book suggestively entitled Behemoth (W.W.Norton & Company, 2018).

In these enormous manufacturing factories, seen with favour by the Chinese government up to the point that is not trivial to see a State vested interest in their existence, there is no assembly line with dozens or hundreds of workers one would imagine. On the contrary, masses of workers are sharing the same roof while their work is actually done either individually or in small groups, assembling pre-cut pieces, with almost no interaction with each other. There is no actual need for them to be physically sharing such huge spaces but, the author explains, it is in the interest of an economy of scale that pleases the retailers, not the manufacturers that these masses of cheap workers congregate in physical spaces where they can be managed and controlled fast.

This reality is not very different from what happened in the mills of the first industrial revolution where weavers or spinners stood side by side doing individual tasks, Freeman notes.

We can also add, in my view, that this is pretty much what happens to the work done on many digital platforms by underpaid, voluntary or unintentional workers: networks of Facebook friends as well as the journalists that populate with their profiles and portfolios the BBC recruitment portal, do not need to share a physical location in order to perform single tasks or produce digital contents that will then circulate and become available for the "surveillance capitalism" economy (more on the subject with the next article), be aggregated, transformed and exploited as "user generated" contributions to big conversations, packed with commercial solicitations or reframed within propaganda discourses nobody but the machines seem to have a remit to control. What does matters is, in the end, the user experience design and the brand of the final product or service it carries forward.

The invisible plant, non even physically existent but for digital connections among people and machines, seems therefore the next evolutionary step in the economy of scale of these cyber-physical gigantic asiatic plants that manufacture the world's goods.

Will the physical gigantism of asian factories last for long or will it transform itself into some sort of new micro-manufacturing and distributed model, making labour more sustainably managed and aggregated? On this idea, I did not find hooks in Freeman's book.

Nonetheless, he seems incline to believe that the dinosaurs of modern manufacturing are going to face the music of more automation on one side and the pressure to modernise working conditions on the other - that means to me expansion of the number of people working in production as self employed and, al least in nominal terms, with a more independent status.

Freeman's history of factory production interestingly links past and present in the global evolution of the industrial plants. It shows how, from the XIX Century onwards, few manufacturers established their dominance by selling their products under brand names and controlling distribution networks. In the United States, the Lowell mills pioneered this approach, which was adopted by such iconic companies as the McCormick Harvesting Machine company. The Singer Manufacturing Company extended the model to a global scale, as its salesmen and distribution agents sold sewing machines across Europe and the Americas, largely produced in just two factories. The big automobile manufacturers used the model as well, selling cars that they branded.

The thousands of Chinese workers that assembly iPhones on a work shift of twelve hours are given brain-washing induction on how to comply with the factory rules but they do not need any training at all in the actual production tasks that require very little know-how: these tasks could be performed anywhere.

A rash of suicides has brought to the international attention the theme of the working condition in the Behemoth plants but apart from these extreme news the media tend to ignore the reality of the asian mega factories.

Neither the issues they raise would be easily categorised as part of a digital slavery or modern slavery phenomenon so far. These giant plants are just the extreme version, in some sense the perfect one, of the factories invented in the West applying theories and techniques of scientific management production.

The future of machine learning

History does not only show the backstage and the rationale behind socio-technical innovations through business, company and industry chronicles. It also documents the scientific, technical and engineering struggle that makes the things what they are.

Fascinating pieces for an archeology of information and machine learning design can be found in the specifications for programmable machines used in manufacturing and services from the 1970s onwards. One of these pioneering machines was The Centurion, a programmable industrial sewing machine designed and developed by electric engineers at the Singer Company in the late 1970s.

The machine was supposed to learn a sewing operation by monitoring a trained operator performing the operation once manually: the sewing speed and length of each seam in the operation are memorised and saved in a sewing-program library in battery-protected CMOS RAM memory. Doesn't it sound familiar to modern machine learning designers? This is what Jack V. Landau wrote in February 1979 in an article published by the IEEE's "Computer" magazine, documenting what was at the time a revolutionary approach. Thereafter he continued a less-experienced operator can quickly achieve high-quality work by simply guiding the fabric under the needle while the machine automatically controls the sewing speed and seam length it learned originally. As the trainee becomes more proficient at guiding the fabric, the sewing speed can be increased.

Massive outsourcing of production to countries like China, financial downsizing and corporate restructuring have deeply transformed the entire fashion, apparel and textile industry not less than automation programmes.

In the USA, in Europe, in Japan various public funded programmes have now tackled the problematic issue of dealing with automation of manufacturing of flexible materials, achieving some interesting results. These materials must be cut assembled arranged or sewed with an infinite variation of patterns and parameters, easily managed by skilled labour forces, but not so bravely done by machines.

Workers handling material pieces during and between the sewing operations have been progressively offered support, aid and actual substitutions of manual operations by robotic conveyors, with grippers, rollers, pickers. Robotic parts of a gigantic dressmaker can be mathematically joint and become more precise and faster than skilled labourers in dealing with materials and surfaces that are vibrating, freezing, rolling, sliding, destacking, aligning and so on in predictable terms.

Progenitors of the smart sensing sewing machines introduced in the 2010s, the automatic controls and programmable units of the previous two decades were often aiming at replacing the workers. Intelligent sewing machines back in the 1990s have implemented neural network and fuzzy logic approach to calculate the best ways to handle, to move, to bend, to cut and to stitch limp materials in a complex jigsaw of efficiency that can be unachievable for human operators. But the question often unanswered has remained in the air since then: it is worth the effort?

Today modern automation in the textile factory follows an industry 4.0 paradigm that looks at the integration of different systems into optimised cyber-physical spaces and workflows. The core of the automation sought is now on the prevention of mistakes and errors: the idea of replacing tout court human labour with software routines and networks of sensors and robotic hands and limbs is not socially viable. Instead, the paradigm of Industry 4.0 tends to make more culturally and technically acceptable the brutal effects of automation on the employability of the current generation of workers while blinking at the next one. The future has always had room for answers we cannot easily find by the end of the day. Manana. Tomorrow.

Once again, the whole of the technical and digital revolution seems revolving around the idea of automata performing better than human beings. How to achieve this in the smart sewing factory? I am not sure I have got all the sources on the state of the affairs (neither I can reach the end of many of these technical papers without regretting I have embarked for unachievable understanding of technical complexity) but it looks like robots need to focus at the stitch level, trying to support, optimise or perform minuscule and very often extremely versatile operations. In sum they should try to imitate, as the grandad The Centurion, the best workers.

In fact, it is on the accuracy of execution of these tiny bits of work, mostly carried by human labourers until now, that the entire system of sensed data and controls relies (I will not return here on the matter I considered in icm2re 6.4, Decluttering machine learning through accuracy). Automated systems for the Industry 4.0 textile plant should be not only designed with obsessive care for details but also made extremely flexible: in a certain sense they need to be redesigned again and again any time a request for change becomes a business priority. Once again, is it worth?

Requests for changes in the assembling, cutting or finishing of a product can come from the retailers and the fashionistas, from the businesses involved in the supply chain or from the final consumers: will the machines be able to satisfy such demand better than human labourers, on time and on budget?

The future of the cyber-physical system of the smart sewing factory seems a fascinating concept with strong foundations: it builds upon the specifications of the Centurion, it includes human factors and the human being, it aims at excluding errors, mistakes and misbehaviours and not just at replacing humans with machines. All in all, everything sounds good. But it looks at present quite aspirational.

New routes of development can be sought, looking at how to make both the consumers and the machines more flexible and responsive to the need of a world that is changing approach to sustainability and environmental issues, bypassing the same idea of the mammoth factory altogether.

Optimisation of production cycles may suddenly not be at all the first priority, neither for the consumer nor for the regulators.

Think of small changes we are going to witness for long decades to come in materials, in the volumes traded, in the chemistry of the fashion and apparel industry as well as in the automotive or in the food processing sectors: the cyber physical system of Industry 4.0 can actually face the fact that information design, controls design and management of change and maintenance are important matters, at least as much as the tension towards optimisation and efficiency, and require experience and know how.

That is a lot to learn. A stitch in time saves nine.