Skip to main content
SpringerLink
Log in

Data production and the coevolving AI trajectories: an attempted evolutionary model

  • Regular Article
  • Published:
Journal of Evolutionary Economics Aims and scope Submit manuscript

Abstract

This paper contributes to the understanding of the relationship between the nature of data and the artificial intelligence (AI) technological trajectories, on the one hand, and on the dynamic processes triggered by demand during the evolution of an industry, on the other hand. We develop an agent-based model in which firms are data producers that compete on the markets for data and AI. The model is enriched by a public sector that fuels the purchase of data and trains the scientists that will populate firms as workforce. Through several simulation experiments, we analyze the determinants of each market structure, the corresponding relationships with innovation attainments, the pattern followed by labor and data productivity, the quality of data traded in the economy, and in which forms demand does affect innovation and the dynamics of industries. We question the established view in the literature of industrial organization according to which technological imperatives are enough to experience divergent industrial dynamics on both the markets for data and AI blueprints. Although technical change behooves if any industry pattern is to emerge, the actual unfolding is not the outcome of a specific technological trajectory, but the result of the interplay between technology-related factors and the availability of data-complementary inputs such as labor and AI capital, the market size, preferences, and public policies.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Log in via an institution

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Institutional subscriptions

Fig. 1
Fig. 2
Fig. 3
Fig. 4
Fig. 5
Fig. 6
Fig. 7
Fig. 8

Similar content being viewed by others

Data Availability Statement

The pseudo-code for the simulation conducted in this paper is available upon request from the authors.

Notes

  1. This finding is in agreement with Acemoglu and Restrepo (2017), which find no negative relationship between aging and economic growth; by contrast, countries undergoing rapid demographic changes are more likely to adopt new automation technologies.

  2. Discussing all the relevant articles may divert our focus. For the sake of simplicity, we refer to Dosi (1982, 1988); Malerba and Orsenigo (1995, 1996b, 2002); Winter (1984); Malerba et al. (1999, 2007), and Silverberg and Verspagen (1994, 1995). Dosi and Nelson (2010) offer a literature review of uncommon clearness while Malerba (2007) discusses challenges yet to be explored.

  3. Focusing on China, Yu et al. (2021) investigate how data shape actors relations in data-driven innovation systems. Findings reveal that data are fundamentally different from conventional resources and controlling data impacts business value creation, knowledge development, as well as regulation formation.

  4. A corollary regards the actions data producers could take: they may in principle exploit the information contained in the data so as to negatively affect data subjects’ privacy. Concerns about privacy and security-related issues are in Chen et al. (2021, ch. 4).

  5. Arrieta-Ibarra et al. (2018) assert that the political economy of data should treat data as labor more than as capital, and considering the market for data like a labor market has the potential to constitute a significant fraction of national income to the benefit of the wider public. Looking at data as labor also implies reward mechanisms to individual users such that they are encouraged to enhance the quality and quantity of data they provide to AI companies and platforms. Furthermore, data as labor would envisage new business models, a new classes of data jobs, and a digital dignity, with noticeable productivity effects that are not easy to capture when the focus is only on data as capital.

  6. Accordingly, data accumulation can be observed as an investment in soft skills associated with creativity, whose nature makes them hardly automatable for these represent a form of tacit knowledge that humans accumulate and apply often spontaneously and unconsciously (Gnecco et al. 2022). We thank the reviewer who made this point.

  7. "[T]he returns to data may decline only gradually or there may even be increasing returns to data if more sophisticated tasks are disproportionately more valuable. This is consistent with the empirically observed dominance of the data economy by a few large firms" (Arrieta-Ibarra et al. 2018, p. 40). However, there is no consensus on the value of data: see Chen et al. (2021); Posner and Weyl (2019); Savona (2019) and Varian (2019) for differing opinions.

  8. We could also imagine a consumer for which we are interested in estimating its preference schedule. Generic data on this customer might be about age, gender, schooling, job, or leisure. These are very generic data. They become specialized when we start focusing on each specific subject. About leisure for instance, does the consumer prefer traveling or sporting? Why? How have her preferences changed over time? What are the reasons behind it? And so on so forth for whatever topic of interest. This leads us to argue that when firms compete on the market for generic data, their aim is to pile up large quantities of raw information about anything. When turning to specialized data, here conveying precise and accurate information on a given subject (hence quality data) makes the difference.

  9. Traub et al. (2019) is an interesting examination of AI components.

  10. We suppose a firm as integrated, such that it builds its own required capital stock. The cost of capital is assumed away from the analysis: it can be either considered as negligible or as embodied in the wage rate, such that entrepreneurs hire workers endowed with physical capital.

  11. If the economy’s potential does not meet aggregate demand, the economy accumulates backlogs that will be added to next period demand. Moreover, at this point of the analysis, somebody might envisage a contradiction in our reasoning: on the one hand, we assume no inventories, on the other hand we speak about a firm’s expectations and the possibility to not satisfy demand. This contradiction is only illusory. The production of data and the subsequent sale does not mean that the seller transfers the good to the buyer and loses it. In fact, the \(i^{th}\) firm keeps part of its production on a repository. What is stored there is in fact an inventory, but it does represent an inventory if inputs, not of output.

  12. The process is similar to the Hicksian production function with capital and labor, in which capital does not substitute for labor, but allows the latter to produce. In the present case, we have the opposite mechanism: the \(i^{th}\) entrepreneur hires scientists to build enough capacity as to transform data in new data, for labor produces the algorithm and related machines.

  13. As a matter of clarification, the constraints on the improvements of the algorithm have a systemic nature (Vannuccini and Prytkova 2023): when the algorithm and the hardware are complements, the evolution of the former is the result of a strategic choice in which the developers design the feature of superior algorithms on the basis of their current hardware production plans, and vice-versa.

  14. The production of new data might hence enlarge the quality and the productivity of the whole ensemble. There is an inherent difference between the productivity of a bundle of data and its quality. The first is a characteristic that benefits data providers with enhanced efficiency in production through better AI technologies; thus, it is something which stays inside the firm. Conversely, the quality of the data concerns to the benefits in terms of information content the users draw from the single data.

  15. The empirical evidence on the pecking order in the capital structure of firms (Fama and French 2002) justifies this assumption. Myers (1984) suggests that firms finance investments with retained earnings to minimize asymmetric information costs and other financing costs. Only as a subsequent step, firms fund investments through safe debt, risky debt, and equity.

  16. Our take entails the belief that novelties, new ideas, innovation, are brought into the firms by newcomers, i.e., newly hired scientists. Equation (20) highlights a further mechanism: entrepreneurs hire workers to build their own AI means of production and these workers build the same vintage of AI through time, unless newcomers arrive at the firm, possibly with better organizational ideas on how to improve technical vintage and labor productivity. Therefore, when technical change unfolds, the elder generation of employees have to learn and adapt to new vintages.

  17. For the sake of simplicity, achievements in quality are commonly shared among stored data.

  18. In some extent, Eq. (27) presents scientists as funds à la Georgescu–Roegen, for the hiring of scientists consists of investments a firm does to ameliorate its capabilities and gain further market power.

  19. It is worth noting that the wage rate may differ between types of scientists within the same firm.

  20. Obviously, different (social) technologies differ in the way they set up the division of labor and the coordination of the many tasks. These differences stand both between and within the borders of any firm. Moreover, such differences may prove to be more, or less, efficient as circumstances change, reflecting variations in opportunities and contexts (Nelson and Sampat 2001).

  21. To clarify, the demand from D and PS is allocated through a replicator mechanism as usual.

  22. For example, we could have pharmaceutical and health organizations, which are interested in both data and AI technology. This industry requires better performing techniques to process medical records, and visual recognition systems for a precise detection of human cancers. We might also have firms building autonomous vehicles, which need an ample collection of sensor readings and actions taken by expert drivers, so as to develop security control systems; such producers might need highly performing algorithms too to analyze data on a car’s energy consumption and to devise more efficient batteries. Still, we could simply have data on customers’ preferences: sellers will be hungry to know these preferences to tailor their products accordingly. The related matching between supply and demand would be further enhanced.

  23. For the sake of clarity, we define \(id^m_{i,t} = \max \left[ 0; Y^{d,m}_{i,t}-Y^{P,m}_{i,t}\right] \).

  24. Looking at single simulations, we found that often a batch of firms emerged as benefiting from higher productivity standards. However, benefits from increased productivity in terms of market shares led to higher mark-ups and higher wage rates. These matters counterbalance the gain in competitiveness originating from productivity standards greater than average, on the one hand, and allow for a reallocation of market demand to other firms. Competition is then restored. Related figures are available upon request.

  25. On technological imperatives, we refer to the literature on history-friendly models, e.g., Landini et al. (2020); Malerba et al. (2001); Malerba and Orsenigo (2002).

  26. We should spend a few words on average labor productivity: the significant decrease corresponding to \(e_1=0.3\) is correlated to the monopolistic structure envisaged in the market for AI systems. A reduced amount of revenues to invest in R &D resulted in no process innovations from the laggards, strengthening the monopolistic leader on the one hand, but diminishing the occurrence of innovation on the other.

  27. The survival of some form of competition through innovation in every market explains the increase in the inverse Herfindahl index referring to aggregate demand, revenues and R &D. At the same time, the lower the elasticity with respect to quality, the lower the variation in the price and mark-up applied to specialized data, when significantly different to the benchmark.

  28. Though statistically significant from the baseline for most of the cases, we do not find any remarkable pattern for what concerns to the coefficients of variation in prices and mark-ups after changes to \(g_{PS}\).

  29. Figures available upon request.

  30. Looking at sample simulations, we noticed that in the market for generic data there is a turnover between full competition and leap-frogging, and full competition again. The point can be related to the fact that whenever a firm gains a prominent position in the market because of the introduction of a novelty, the problem of labor shortage does not allow for an ever-lasting monopolistic position in the market. In order to satisfy the demand for data, that firm must buy from others, thus re-establishing competition.

References

  • Acemoglu D, Autor D, Hazell J, Restrepo P (2022) Artificial intelligence and jobs: evidence from online vacancies. J Labor Econ, 40(S1):S293–S340. (Publisher: The University of Chicago Press Chicago, IL)

  • Acemoglu D, Restrepo P (2017) Secular stagnation? The effect of aging on economic growth in the age of automation. Am Econ Rev, 107(5):174–79. https://doi.org/10.1257/aer.p20171101

    Article  Google Scholar 

  • Acemoglu D, Restrepo P (2019) Automation and new tasks: How technology displaces and reinstates labor. J Econ Perspect, 33(2):3–30. https://doi.org/10.1257/jep.33.2.3

    Article  Google Scholar 

  • Aghion P, Jones BF, Jones CI (2019) 9. Artificial intelligence and economic growth. University of Chicago Press

  • Agrawal A, McHale J, Oettl A (2019) 5. Finding Needles in Haystacks: Artificial intelligence and recombinant growth. University of Chicago Press

  • Agrawal A, McHale J, Oettl A (2022) Superhuman science: How artificial intelligence may impact innovation. Publisher, Brookings Institution

    Google Scholar 

  • Ahmed N, Wahed M (2020) The De-democratization of AI: deep learning and the compute divide in artificial intelligence research. https://doi.org/10.48550/arXiv.2010.15581

  • Almudi I, Fatas-Villafranca F, Izquierdo LR (2012) Innovation, catch-up, and leadership in science-based industries. Ind Corp Chang, 21(2):345–375. https://doi.org/10.1093/icc/dtr041

    Article  Google Scholar 

  • Almudi I, Fatas-Villafranca F, Izquierdo LR (2013) Industry dynamics, technological regimes and the role of demand. J Evol Econ, 23:1073–1098

    Article  Google Scholar 

  • Annoni A, Benczur P, Bertoldi P, Delipetrev B, De Prato G, Feijoo C, Junklewitz H (2018) Artificial intelligence: A European perspective

  • Armstrong S, Bostrom N, Shulman C (2016) Racing to the precipice: a model of artificial intelligence development. AI Soc, 31(2):201–206. https://doi.org/10.1007/s00146-015-0590-y

    Article  Google Scholar 

  • Arrieta-Ibarra I, Goff L, Jiménez-Hernández D, Lanier J, Weyl EG (2018) Should we treat data as labor? Moving beyond" free". aea Papers and Proceedings 108:38–42. https://doi.org/10.1257/pandp.20181003

  • Bianchini S, Müller M, Pelletier P (2022) Artificial intelligence in science: An emerging general method of invention. Res Policy, 51(10)

  • Bordot F (2022) Artificial intelligence, robots and unemployment: Evidence from OECD Countries. J Inn Econ Manag, 37(1):117–138. https://doi.org/10.3917/jie.037.0117

    Article  Google Scholar 

  • Brynjolfsson E, Rock D, Syverson C (2019) 1. Artificial intelligence and the modern productivity paradox: A clash of expectations and statistics. University of Chicago Press

  • Chen L, Bolton P, Holmström BR, Maskin E, Pissarides CA, Spence AM, Yang L (2021) Understanding big data: Data calculus in the digital era. https://doi.org/10.2139/ssrn.3780882

  • Ciarli T, Lorentz A (2010) Product variety and changes in consumption patterns: The effects of structural change on growth. 13th ISS Conference, Aalborg

  • Ciarli T, Lorentz A, Savona M, Valente M (2010) The effect of consumption and production structure on growth and distribution. A micro to macro model. Metroeconomica 61(1):180–218. https://doi.org/10.1111/j.1467-999X.2009.04069.x

    Article  Google Scholar 

  • Cockburn IM, Henderson R, Stern S (2019) 4. The impact of artificial intelligence on innovation: An exploratory analysis. The economics of artificial intelligence. University of Chicago Press (115–148)

  • Delli Gatti D, Desiderio S, Gaffeo E, Cirillo P, Gallegati M (2011) Macroeconomics from the Bottom-up (VOL. 1). Springer Science & Business Media

  • Domini G, Grazzi M, Moschella D, Treibich T (2021) Threats and opportunities in the digital era: automation spikes and employment dynamics. Res Policy 50(7). https://doi.org/10.1016/j.respol.2020.104137

  • Dosi G (1982) Technological paradigms and technological trajectories: a suggested interpretation of the determinants and directions of technical change. Res Policy, 11(3):147–162. https://doi.org/10.1016/0048-7333(82)90016-6

    Article  Google Scholar 

  • Dosi G (1988) Sources, procedures, and microeconomic effects of innovation. J Econ Lit 1120–1171

  • Dosi G, Fagiolo G, Roventini A (2010) Schumpeter meeting Keynes: A policy-friendly model of endogenous growth and business cycles. J Econ Dyn Control, 34(9):1748–1767. https://doi.org/10.1016/j.jedc.2010.06.018

    Article  Google Scholar 

  • Dosi G, Nelson RR (2010) Technical change and industrial dynamics as evolutionary processes. Handb Econ Inn, 1:51–127. https://doi.org/10.1016/S0169-7218(10)01003-8

    Article  Google Scholar 

  • Economist T (2017) The world’s most valuable resource is no longer oil, but data. NY, USA, The Economist, New York

    Google Scholar 

  • Fama EF, French KR (2002) Testing trade-off and pecking order predictions about dividends and debt. Rev Finan Stud 1–33

  • Furman J, Seamans R (2019) AI and the economy. Inn Policy Econ, 19(1):161–191. https://doi.org/10.1086/699936

    Article  Google Scholar 

  • Gnecco G, Landi S, Riccaboni M (2022) Can machines learn creativity needs? An approach based on matrix completion. Ital Econ J 1–41

  • Gundersen OE, Kjensmo S (2018) State of the art: Reproducibility in artificial intelligence. Proceedings of the AAAI conference on artificial intelligence, vol 32. https://doi.org/10.1609/aaai.v32i1.11503

  • Hagendorff T, Meding K (2021) Ethical considerations and statistical analysis of industry involvement in machine learning research. AI Soc 1–11. https://doi.org/10.1007/s00146-021-01284-z

  • Jones CI, Tonetti C (2020) Nonrivalry and the economics of data. Am Econ Rev 110(9):2819–58. https://doi.org/10.1257/aer.20191330

    Article  Google Scholar 

  • Jurowetzki R, Hain D, Mateos-Garcia J, Stathoulopoulos K (2021) The privatization of AI research (-ers): Causes and potential consequences-from university-industry interaction to public research brain-drain? arXiv:2102.01648

  • Klinger J, Mateos-Garcia J, Stathoulopoulos K (2020) A narrowing of AI research? https://doi.org/10.48550/arXiv.2009.10385

  • Klinger J, Mateos-Garcia J, Stathoulopoulos K (2021) Deep learning, deep change? Mapping the evolution and geography of a general purpose technology. Scientometrics, 126:5589–5621

  • Landini F, Lema R, Malerba F (2020) Demand-led catch-up: a history-friendly model of latecomer development in the global green economy. Ind Corp Chang, 29(5):1297–1318

    Article  Google Scholar 

  • LeCun Y, Bengio Y, Hinton G (2015) Deep learning. Nature, 521(7553):436–444. https://doi.org/10.1038/nature14539

    Article  Google Scholar 

  • Malerba F (2007) Innovation and the dynamics and evolution of industries: Progress and challenges. Int J Ind Org 25(4):675–699

    Article  Google Scholar 

  • Malerba F, Nelson R, Orsenigo L, Winter S (1999) ‘History-friendly’ models of industry evolution: the computer industry. Ind Corp Chang, 8(1):3–40. https://doi.org/10.1093/icc/8.1.3

    Article  Google Scholar 

  • Malerba F, Nelson R, Orsenigo L, Winter S (2001) History-friendly models: an overview of the case of the computer industry. J Artif Soc Soc Simul, 4(3):6

    Google Scholar 

  • Malerba F, Nelson R, Orsenigo L, Winter S (2007) Demand, innovation, and the dynamics of market structure: The role of experimental users and diverse preferences. J Evol Econ, 17(4):371–399. https://doi.org/10.1007/s00191-007-0060-x

    Article  Google Scholar 

  • Malerba F, Orsenigo L (1995) Schumpeterian patterns of innovation. Camb J Econ, 19(1):47–65. https://doi.org/10.1093/oxfordjournals.cje.a035308

    Article  Google Scholar 

  • Malerba F, Orsenigo L (1996) The dynamics and evolution of industries. Ind Corp Chang, 5(1):51–87. https://doi.org/10.1093/icc/5.1.51

    Article  Google Scholar 

  • Malerba F, Orsenigo L (1996) Schumpeterian patterns of innovation are technology-specific. Res Policy, 25(3):451–478. https://doi.org/10.1016/0048-7333(95)00840-3

    Article  Google Scholar 

  • Malerba F, Orsenigo L (2002) Innovation and market structure in the dynamics of the pharmaceutical industry and biotechnology: towards a history-friendly model. Ind Corp Chang, 11(4):667–703. https://doi.org/10.1093/icc/11.4.667

    Article  Google Scholar 

  • Mazzucato M, Schaake M, Krier S, Entsminger J (2022) Governing artificial intelligence in the public interest. UCL Institute for Innovation and Public Purpose, Working Paper Series (IIPP WP 2022-12). Retrieved April 2 2023

  • Myers SC (1984) Capital structure puzzle. Nat Bur Econ Res Camb, Mass., USA

  • Nagaraj A, Shears E, de Vaan M (2020) Improving data access democratizes and diversifies science. Proc Nat Acad Sci, 117(38):23490–23498. https://doi.org/10.1073/pnas.2001682117

    Article  Google Scholar 

  • Nelson RR, Sampat BN (2001) Making sense of institutions as a factor shaping economic performance. Rev Econ Ins, 3(5):17–51. https://doi.org/10.1016/S0167-2681(00)00152-9

    Article  Google Scholar 

  • Nelson RR, Winter SG (1982) An evolutionary theory of economic change. Harvard University Press

  • Posner EA, Weyl EG (2019) Radical markets. Princeton University Press

    Book  Google Scholar 

  • Savona M (2019). The value of data: Towards a framework to redistribute it. https://doi.org/10.2139/ssrn.3476668

    Article  Google Scholar 

  • Schumpeter J (1934) The theory of economic development. Harvard Economic Studies, Cambridge, MA

    Google Scholar 

  • Schumpeter JA (1942) Capitalism, socialism and democracy. Harper, New York

    Google Scholar 

  • Silverberg G, Verspagen B (1994) Learning, innovation and economic growth: a long-run model of industrial dynamics. Ind Corp Chang, 3(1):199–223. https://doi.org/10.1093/icc/3.1.199

    Article  Google Scholar 

  • Silverberg G, Verspagen B (1995) An evolutionary model of long term cyclical variations of catching up and falling behind. J Evol Econ, 5(3):209–227. https://doi.org/10.1007/BF01198304

    Article  Google Scholar 

  • Spiekermann M (2019) Data marketplaces: Trends and monetisation of data goods. Intereconomics 54(4):208–216. https://doi.org/10.1007/s10272-019-0826-z

    Article  Google Scholar 

  • Traub J, Quiané-Ruiz J-A, Kaoudi Z, Markl V (2019) Agora: Towards an open ecosystem for democratizing data science & artificial intelligence. arXiv:1909.03026

  • Vannuccini S, Prytkova E (2023) Artificial intelligence’s new clothes? A System Technology Perspective. Journal of Information Technology, 02683962231197824

  • Varian H (2019) 16. Artificial intelligence, economics, and industrial organization. University of Chicago Press

  • Winter SG (1984) Schumpeterian competition in alternative technological regimes. J Econ Behav Org, 5(3–4):287–320

    Article  Google Scholar 

  • Yu Z, Liang Z, Wu P (2021) How data shape actor relations in artificial intelligence innovation systems: an empirical observation from China. Ind Corp Chang, 30(1):251–267. https://doi.org/10.1093/icc/dtaa063

    Article  Google Scholar 

Download references

Acknowledgements

This research has received financial support from the French National Research Agency [reference: DInnAMICS -ANR-18-CE26-0017-01]. The authors are grateful to Nathalie Lazaric, Maria Savona, Tommaso Ciarli, Simone Vannuccini, and Paolo Zeppini for useful and pleasant conversation on the previous version of the paper. This work was presented at the online SPRU Wednesday Seminars, the online International Schumpeter Society Conference, the INFER Workshop on Economic Growth and Macroeconomic Dynamics in Rome, and at the ABM4Policy Workshop in Pisa in 2022. We want to thank all the participants for their comments and suggestions. We are also grateful to the editor and two anonymous referees for their comments, which improved the previous version of the manuscript. Usual disclaimers apply.

Author information

Authors and Affiliations

  1. Bureau d’Economie Théorique et Appliquée, University of Strasbourg, University of Lorraine, CNRS, 23 Avenue de la Forêt Noire, Strasbourg, 67000, France

    Andrea Borsato & André Lorentz

Authors
  1. Andrea Borsato
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. André Lorentz
    View author publications

    You can also search for this author in PubMed Google Scholar

Corresponding author

Correspondence to Andrea Borsato.

Ethics declarations

Conflict of interest

This research has received financial support from the French National Research Agency [reference: DInnAMICS -ANR-18-CE26-0017-01]. The authors have no relevant financial or non-financial interests to disclose.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Rights and permissions

Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Borsato, A., Lorentz, A. Data production and the coevolving AI trajectories: an attempted evolutionary model. J Evol Econ 33, 1427–1472 (2023). https://doi.org/10.1007/s00191-023-00837-3

Download citation

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s00191-023-00837-3

Keywords

JEL Classification

Search

Navigation