Allyn Young's concept of increasing returns (not to be confused with static, equilibrium constructs of economies of scale and increasing returns to scale) is applied to analyse how and why increasing returns arise in the production (generation) and use (application) of knowledge and big data, thereby driving economic growth and progress. Knowledge is chosen as our focus because it is said to be our most powerful engine of production, and big data are included to make the analysis more complete and recent. We analyse four mechanisms or sources of increasing returns in the production of knowledge, and four in the use of knowledge. Turning to big data, we analyse increasing returns in the functioning of digital platforms and increasing returns in machine learning from gigantic amounts of training data. Concluding remarks concern some key differences between big data and knowledge, some policy implications, and some of the social negative impacts from the ways in which big data are being used.
‘Whether it be the older literature on research and development or the modern New Growth Theory, the mainstream account runs along the following lines. Knowledge is produced privately using a sausage-machine called research and development that takes in inputs and gives off technological knowledge, which then immediately augments the production function for other goods’ (Langlois, 1999, p.249). This characterization applies to Paul Romer's path-breaking paper (1990) in which knowledge and new designs are generated in the R&D department of firms, and the spillover of such knowledge to all other firms is reflected by including the total stock of knowledge in the economy in the production function of the representive firm (see also Kurz, 2012, pp.95–7). The growing literature on ‘non-R&D’ sources of knowledge production and/or innovation (Barge-Gil, Nieto and Santamaria., 2011; Lee and Walsh, 2016), together with the difficulty of drawing the line, within firms, between R&D, design, engineering, prototyping and scaling up from pilot plants (Freeman and Soete, 2009), should put question marks over such an approach.
Economic growth is quantitative, progress qualitative.
According to Loasby (1999, p.135), 'The division of labour is the primary means of increasing the division of knowledge, and thereby of promoting the growth of knowledge. Knowledge grows by division: each of us can increase our knowledge only by accepting limits on what we can know.' According to Metcalfe (2014, p.17), 'The division of labour is a division of knowing and, moreover, the division of labour applies to the development of knowledge as well as to its application.'
Joan Robinson (1979, p.58) repeatedly pointed out that ‘a confusion between comparisons of imagined equilibrium positions and a process of accumulation going on through history’ was ‘an error in methodology’ on the part of neoclassical economists.
The knowledge management literature identifies at least four kinds of knowledge processes at the organizational level – knowledge creation, knowledge application, knowledge integration and knowledge retention (Kraaijenbrink, 2012). The sociology of knowledge literature distinguishes between processes of knowledge production – knowledge organization, dissemination-distribution, and storage-retrieval – and knowledge application (Holzner and Marx, 1979). The history of knowledge identifies at least 32 processes that can be grouped under the four main stages of knowledge gathering, analysing, disseminating and employing (Burke, 2016).
Pavitt (2005) identified the production of scientific and technological knowledge and the transformation of knowledge into working artefacts as two of the three key sub-processes in the process of innovation.
They are: the communication theory approach, the probabilistic approach, the modal approach, the systemic approach, the inferential approach, and the semantic approach.
This was quite a departure from the tradition in epistemology which defines knowledge as justified true belief.
Shapiro and Varian (1999, p.3) define information as anything that can be digitized (encoded as a stream of bits).
Paul Romer's paper, ‘Endogenous technological change’, was not published until October 1990.
It is for this reason that we do not adopt the expression ‘economies of specialization and division of labour’ put forward by Yang and Ng (1998, p.8).
One of the founders of economic growth theory wrote more than 70 years ago: ‘Problems arising from a onceover change can, I believe, be satisfactorily handled by the apparatus of static theory. It is when we come to a steadily continuing change that we have to consider a different technique … Dynamics will specifically be concerned with the effects of continuing changes and with rates of change’ (Harrod, 1948, pp.7, 8).
The Gospel of Matthew (chapter 25, verse 29) says that 'For unto every one that hath shall be given, and he shall have abundance: but from him that hath not shall be taken away even that which he hath.' The term 'Matthew effect' was introduced by sociologist Robert Merton to describe the reward and communication systems of science (Merton, 1968).
The law, named after Gordon Moore of Intel Corp., has been expressed in several ways: that the number of transistors incorporated in an electronic chip would approximately double every two years, that the performance of microchips would double every 18 months, or that the price of integrated circuits halves as the number of transistors therein doubles.
Strictly speaking, negative feedback processes lead to a successive reduction in the amplitude of the deviations towards equilibrium, whereas vicious circles are the symmetrical opposite of virtuous circles.
Division of labour and specialization are synonymous except in the case where the division of labour is increased to the point of greatly simplifying the job to be done, which nullifies the element of job-specific skills or specialization (see Morroni, 1992).
There used to be a productivity paradox in the US, where between 1973 and 1995 labour productivity growth was poor (1.3% per annum) despite the huge investments that had been made in information technology. This could have been attributable in part to mismeasurement of outputs and inputs, and also time lags in learning and adjustment (Brynjolfsson, 1993). ‘The impact of the computer revolution became apparent in the productivity statistics beginning around 1995. Having grown slowly during the 1973–1995 period, labour productivity then surged ahead at 2.6% per year from 1995 to 2008‘ (Samuelson and Nordhaus, 2010, p.233).
‘In 2017, 24 hyperscale firms operated 320 data centres with anywhere between thousands and millions of servers’ (Zuboff, 2019, p.501). Google's warehouse-sized data centres, in 15 locations, had an estimated 2.5 million servers on four continents in 2016 (Zuboff, 2019, p.188).
Network externality means that the value of connecting to a network depends on the number of other people already connected to it. A tenfold increase in the size of the network leads to a hundredfold increase in its value (see Shapiro and Varian, 1999).
Although Google is primarily a search engine, most of its revenues are derived from advertisements. In 2016, 89% of the revenues of Google's parent company (Alphabet) derived from Google's targeted advertising programs (Zuboff, 2019, p.93).
The concept of ownership of knowledge is well-established in intellectual property rights legislation, whereas ownership of data is not yet firmly established in law. The European Union's General Data Protection Regulation (GDPR), which came into effect in May 2018, is a first step towards establishing that personal data should be owned by the persons from whom the data are collected.